diff options
Diffstat (limited to 'ruby/ext')
| -rw-r--r-- | ruby/ext/google/protobuf_c/defs.c | 1763 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/encode_decode.c | 1233 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/extconf.rb | 10 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/map.c | 808 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/message.c | 578 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/protobuf.c | 115 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/protobuf.h | 538 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/repeated_field.c | 651 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/storage.c | 894 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/upb.c | 12812 | ||||
| -rw-r--r-- | ruby/ext/google/protobuf_c/upb.h | 8569 |
11 files changed, 27971 insertions, 0 deletions
diff --git a/ruby/ext/google/protobuf_c/defs.c b/ruby/ext/google/protobuf_c/defs.c new file mode 100644 index 00000000..7e93bafb --- /dev/null +++ b/ruby/ext/google/protobuf_c/defs.c @@ -0,0 +1,1763 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Common utilities. +// ----------------------------------------------------------------------------- + +static const char* get_str(VALUE str) { + Check_Type(str, T_STRING); + return RSTRING_PTR(str); +} + +static VALUE rb_str_maybe_null(const char* s) { + if (s == NULL) { + s = ""; + } + return rb_str_new2(s); +} + +static upb_def* check_notfrozen(const upb_def* def) { + if (upb_def_isfrozen(def)) { + rb_raise(rb_eRuntimeError, + "Attempt to modify a frozen descriptor. Once descriptors are " + "added to the descriptor pool, they may not be modified."); + } + return (upb_def*)def; +} + +static upb_msgdef* check_msg_notfrozen(const upb_msgdef* def) { + return upb_downcast_msgdef_mutable(check_notfrozen((const upb_def*)def)); +} + +static upb_fielddef* check_field_notfrozen(const upb_fielddef* def) { + return upb_downcast_fielddef_mutable(check_notfrozen((const upb_def*)def)); +} + +static upb_oneofdef* check_oneof_notfrozen(const upb_oneofdef* def) { + return (upb_oneofdef*)check_notfrozen((const upb_def*)def); +} + +static upb_enumdef* check_enum_notfrozen(const upb_enumdef* def) { + return (upb_enumdef*)check_notfrozen((const upb_def*)def); +} + +// ----------------------------------------------------------------------------- +// DescriptorPool. +// ----------------------------------------------------------------------------- + +#define DEFINE_CLASS(name, string_name) \ + VALUE c ## name; \ + const rb_data_type_t _ ## name ## _type = { \ + string_name, \ + { name ## _mark, name ## _free, NULL }, \ + }; \ + name* ruby_to_ ## name(VALUE val) { \ + name* ret; \ + TypedData_Get_Struct(val, name, &_ ## name ## _type, ret); \ + return ret; \ + } \ + +#define DEFINE_SELF(type, var, rb_var) \ + type* var = ruby_to_ ## type(rb_var) + +// Global singleton DescriptorPool. The user is free to create others, but this +// is used by generated code. +VALUE generated_pool; + +DEFINE_CLASS(DescriptorPool, "Google::Protobuf::DescriptorPool"); + +void DescriptorPool_mark(void* _self) { +} + +void DescriptorPool_free(void* _self) { + DescriptorPool* self = _self; + upb_symtab_unref(self->symtab, &self->symtab); + xfree(self); +} + +/* + * call-seq: + * DescriptorPool.new => pool + * + * Creates a new, empty, descriptor pool. + */ +VALUE DescriptorPool_alloc(VALUE klass) { + DescriptorPool* self = ALLOC(DescriptorPool); + self->symtab = upb_symtab_new(&self->symtab); + return TypedData_Wrap_Struct(klass, &_DescriptorPool_type, self); +} + +void DescriptorPool_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "DescriptorPool", rb_cObject); + rb_define_alloc_func(klass, DescriptorPool_alloc); + rb_define_method(klass, "add", DescriptorPool_add, 1); + rb_define_method(klass, "build", DescriptorPool_build, 0); + rb_define_method(klass, "lookup", DescriptorPool_lookup, 1); + rb_define_singleton_method(klass, "generated_pool", + DescriptorPool_generated_pool, 0); + cDescriptorPool = klass; + rb_gc_register_address(&cDescriptorPool); + + generated_pool = rb_class_new_instance(0, NULL, klass); + rb_gc_register_address(&generated_pool); +} + +static void add_descriptor_to_pool(DescriptorPool* self, + Descriptor* descriptor) { + CHECK_UPB( + upb_symtab_add(self->symtab, (upb_def**)&descriptor->msgdef, 1, + NULL, &status), + "Adding Descriptor to DescriptorPool failed"); +} + +static void add_enumdesc_to_pool(DescriptorPool* self, + EnumDescriptor* enumdesc) { + CHECK_UPB( + upb_symtab_add(self->symtab, (upb_def**)&enumdesc->enumdef, 1, + NULL, &status), + "Adding EnumDescriptor to DescriptorPool failed"); +} + +/* + * call-seq: + * DescriptorPool.add(descriptor) + * + * Adds the given Descriptor or EnumDescriptor to this pool. All references to + * other types in a Descriptor's fields must be resolvable within this pool or + * an exception will be raised. + */ +VALUE DescriptorPool_add(VALUE _self, VALUE def) { + DEFINE_SELF(DescriptorPool, self, _self); + VALUE def_klass = rb_obj_class(def); + if (def_klass == cDescriptor) { + add_descriptor_to_pool(self, ruby_to_Descriptor(def)); + } else if (def_klass == cEnumDescriptor) { + add_enumdesc_to_pool(self, ruby_to_EnumDescriptor(def)); + } else { + rb_raise(rb_eArgError, + "Second argument must be a Descriptor or EnumDescriptor."); + } + return Qnil; +} + +/* + * call-seq: + * DescriptorPool.build(&block) + * + * Invokes the block with a Builder instance as self. All message and enum types + * added within the block are committed to the pool atomically, and may refer + * (co)recursively to each other. The user should call Builder#add_message and + * Builder#add_enum within the block as appropriate. This is the recommended, + * idiomatic way to define new message and enum types. + */ +VALUE DescriptorPool_build(VALUE _self) { + VALUE ctx = rb_class_new_instance(0, NULL, cBuilder); + VALUE block = rb_block_proc(); + rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); + rb_funcall(ctx, rb_intern("finalize_to_pool"), 1, _self); + return Qnil; +} + +/* + * call-seq: + * DescriptorPool.lookup(name) => descriptor + * + * Finds a Descriptor or EnumDescriptor by name and returns it, or nil if none + * exists with the given name. + */ +VALUE DescriptorPool_lookup(VALUE _self, VALUE name) { + DEFINE_SELF(DescriptorPool, self, _self); + const char* name_str = get_str(name); + const upb_def* def = upb_symtab_lookup(self->symtab, name_str); + if (!def) { + return Qnil; + } + return get_def_obj(def); +} + +/* + * call-seq: + * DescriptorPool.generated_pool => descriptor_pool + * + * Class method that returns the global DescriptorPool. This is a singleton into + * which generated-code message and enum types are registered. The user may also + * register types in this pool for convenience so that they do not have to hold + * a reference to a private pool instance. + */ +VALUE DescriptorPool_generated_pool(VALUE _self) { + return generated_pool; +} + +// ----------------------------------------------------------------------------- +// Descriptor. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(Descriptor, "Google::Protobuf::Descriptor"); + +void Descriptor_mark(void* _self) { + Descriptor* self = _self; + rb_gc_mark(self->klass); + rb_gc_mark(self->typeclass_references); +} + +void Descriptor_free(void* _self) { + Descriptor* self = _self; + upb_msgdef_unref(self->msgdef, &self->msgdef); + if (self->layout) { + free_layout(self->layout); + } + if (self->fill_handlers) { + upb_handlers_unref(self->fill_handlers, &self->fill_handlers); + } + if (self->fill_method) { + upb_pbdecodermethod_unref(self->fill_method, &self->fill_method); + } + if (self->json_fill_method) { + upb_json_parsermethod_unref(self->json_fill_method, + &self->json_fill_method); + } + if (self->pb_serialize_handlers) { + upb_handlers_unref(self->pb_serialize_handlers, + &self->pb_serialize_handlers); + } + if (self->json_serialize_handlers) { + upb_handlers_unref(self->json_serialize_handlers, + &self->json_serialize_handlers); + } + if (self->json_serialize_handlers_preserve) { + upb_handlers_unref(self->json_serialize_handlers_preserve, + &self->json_serialize_handlers_preserve); + } + xfree(self); +} + +/* + * call-seq: + * Descriptor.new => descriptor + * + * Creates a new, empty, message type descriptor. At a minimum, its name must be + * set before it is added to a pool. It cannot be used to create messages until + * it is added to a pool, after which it becomes immutable (as part of a + * finalization process). + */ +VALUE Descriptor_alloc(VALUE klass) { + Descriptor* self = ALLOC(Descriptor); + VALUE ret = TypedData_Wrap_Struct(klass, &_Descriptor_type, self); + self->msgdef = upb_msgdef_new(&self->msgdef); + self->klass = Qnil; + self->layout = NULL; + self->fill_handlers = NULL; + self->fill_method = NULL; + self->json_fill_method = NULL; + self->pb_serialize_handlers = NULL; + self->json_serialize_handlers = NULL; + self->json_serialize_handlers_preserve = NULL; + self->typeclass_references = rb_ary_new(); + return ret; +} + +void Descriptor_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "Descriptor", rb_cObject); + rb_define_alloc_func(klass, Descriptor_alloc); + rb_define_method(klass, "each", Descriptor_each, 0); + rb_define_method(klass, "lookup", Descriptor_lookup, 1); + rb_define_method(klass, "add_field", Descriptor_add_field, 1); + rb_define_method(klass, "add_oneof", Descriptor_add_oneof, 1); + rb_define_method(klass, "each_oneof", Descriptor_each_oneof, 0); + rb_define_method(klass, "lookup_oneof", Descriptor_lookup_oneof, 1); + rb_define_method(klass, "msgclass", Descriptor_msgclass, 0); + rb_define_method(klass, "name", Descriptor_name, 0); + rb_define_method(klass, "name=", Descriptor_name_set, 1); + rb_include_module(klass, rb_mEnumerable); + cDescriptor = klass; + rb_gc_register_address(&cDescriptor); +} + +/* + * call-seq: + * Descriptor.name => name + * + * Returns the name of this message type as a fully-qualfied string (e.g., + * My.Package.MessageType). + */ +VALUE Descriptor_name(VALUE _self) { + DEFINE_SELF(Descriptor, self, _self); + return rb_str_maybe_null(upb_msgdef_fullname(self->msgdef)); +} + +/* + * call-seq: + * Descriptor.name = name + * + * Assigns a name to this message type. The descriptor must not have been added + * to a pool yet. + */ +VALUE Descriptor_name_set(VALUE _self, VALUE str) { + DEFINE_SELF(Descriptor, self, _self); + upb_msgdef* mut_def = check_msg_notfrozen(self->msgdef); + const char* name = get_str(str); + CHECK_UPB( + upb_msgdef_setfullname(mut_def, name, &status), + "Error setting Descriptor name"); + return Qnil; +} + +/* + * call-seq: + * Descriptor.each(&block) + * + * Iterates over fields in this message type, yielding to the block on each one. + */ +VALUE Descriptor_each(VALUE _self) { + DEFINE_SELF(Descriptor, self, _self); + + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, self->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + VALUE obj = get_def_obj(field); + rb_yield(obj); + } + return Qnil; +} + +/* + * call-seq: + * Descriptor.lookup(name) => FieldDescriptor + * + * Returns the field descriptor for the field with the given name, if present, + * or nil if none. + */ +VALUE Descriptor_lookup(VALUE _self, VALUE name) { + DEFINE_SELF(Descriptor, self, _self); + const char* s = get_str(name); + const upb_fielddef* field = upb_msgdef_ntofz(self->msgdef, s); + if (field == NULL) { + return Qnil; + } + return get_def_obj(field); +} + +/* + * call-seq: + * Descriptor.add_field(field) => nil + * + * Adds the given FieldDescriptor to this message type. This descriptor must not + * have been added to a pool yet. Raises an exception if a field with the same + * name or number already exists. Sub-type references (e.g. for fields of type + * message) are not resolved at this point. + */ +VALUE Descriptor_add_field(VALUE _self, VALUE obj) { + DEFINE_SELF(Descriptor, self, _self); + upb_msgdef* mut_def = check_msg_notfrozen(self->msgdef); + FieldDescriptor* def = ruby_to_FieldDescriptor(obj); + upb_fielddef* mut_field_def = check_field_notfrozen(def->fielddef); + CHECK_UPB( + upb_msgdef_addfield(mut_def, mut_field_def, NULL, &status), + "Adding field to Descriptor failed"); + add_def_obj(def->fielddef, obj); + return Qnil; +} + +/* + * call-seq: + * Descriptor.add_oneof(oneof) => nil + * + * Adds the given OneofDescriptor to this message type. This descriptor must not + * have been added to a pool yet. Raises an exception if a oneof with the same + * name already exists, or if any of the oneof's fields' names or numbers + * conflict with an existing field in this message type. All fields in the oneof + * are added to the message descriptor. Sub-type references (e.g. for fields of + * type message) are not resolved at this point. + */ +VALUE Descriptor_add_oneof(VALUE _self, VALUE obj) { + DEFINE_SELF(Descriptor, self, _self); + upb_msgdef* mut_def = check_msg_notfrozen(self->msgdef); + OneofDescriptor* def = ruby_to_OneofDescriptor(obj); + upb_oneofdef* mut_oneof_def = check_oneof_notfrozen(def->oneofdef); + CHECK_UPB( + upb_msgdef_addoneof(mut_def, mut_oneof_def, NULL, &status), + "Adding oneof to Descriptor failed"); + add_def_obj(def->oneofdef, obj); + return Qnil; +} + +/* + * call-seq: + * Descriptor.each_oneof(&block) => nil + * + * Invokes the given block for each oneof in this message type, passing the + * corresponding OneofDescriptor. + */ +VALUE Descriptor_each_oneof(VALUE _self) { + DEFINE_SELF(Descriptor, self, _self); + + upb_msg_oneof_iter it; + for (upb_msg_oneof_begin(&it, self->msgdef); + !upb_msg_oneof_done(&it); + upb_msg_oneof_next(&it)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&it); + VALUE obj = get_def_obj(oneof); + rb_yield(obj); + } + return Qnil; +} + +/* + * call-seq: + * Descriptor.lookup_oneof(name) => OneofDescriptor + * + * Returns the oneof descriptor for the oneof with the given name, if present, + * or nil if none. + */ +VALUE Descriptor_lookup_oneof(VALUE _self, VALUE name) { + DEFINE_SELF(Descriptor, self, _self); + const char* s = get_str(name); + const upb_oneofdef* oneof = upb_msgdef_ntooz(self->msgdef, s); + if (oneof == NULL) { + return Qnil; + } + return get_def_obj(oneof); +} + +/* + * call-seq: + * Descriptor.msgclass => message_klass + * + * Returns the Ruby class created for this message type. Valid only once the + * message type has been added to a pool. + */ +VALUE Descriptor_msgclass(VALUE _self) { + DEFINE_SELF(Descriptor, self, _self); + if (!upb_def_isfrozen((const upb_def*)self->msgdef)) { + rb_raise(rb_eRuntimeError, + "Cannot fetch message class from a Descriptor not yet in a pool."); + } + if (self->klass == Qnil) { + self->klass = build_class_from_descriptor(self); + } + return self->klass; +} + +// ----------------------------------------------------------------------------- +// FieldDescriptor. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(FieldDescriptor, "Google::Protobuf::FieldDescriptor"); + +void FieldDescriptor_mark(void* _self) { +} + +void FieldDescriptor_free(void* _self) { + FieldDescriptor* self = _self; + upb_fielddef_unref(self->fielddef, &self->fielddef); + xfree(self); +} + +/* + * call-seq: + * FieldDescriptor.new => field + * + * Returns a new field descriptor. Its name, type, etc. must be set before it is + * added to a message type. + */ +VALUE FieldDescriptor_alloc(VALUE klass) { + FieldDescriptor* self = ALLOC(FieldDescriptor); + VALUE ret = TypedData_Wrap_Struct(klass, &_FieldDescriptor_type, self); + upb_fielddef* fielddef = upb_fielddef_new(&self->fielddef); + upb_fielddef_setpacked(fielddef, false); + self->fielddef = fielddef; + return ret; +} + +void FieldDescriptor_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "FieldDescriptor", rb_cObject); + rb_define_alloc_func(klass, FieldDescriptor_alloc); + rb_define_method(klass, "name", FieldDescriptor_name, 0); + rb_define_method(klass, "name=", FieldDescriptor_name_set, 1); + rb_define_method(klass, "type", FieldDescriptor_type, 0); + rb_define_method(klass, "type=", FieldDescriptor_type_set, 1); + rb_define_method(klass, "label", FieldDescriptor_label, 0); + rb_define_method(klass, "label=", FieldDescriptor_label_set, 1); + rb_define_method(klass, "number", FieldDescriptor_number, 0); + rb_define_method(klass, "number=", FieldDescriptor_number_set, 1); + rb_define_method(klass, "submsg_name", FieldDescriptor_submsg_name, 0); + rb_define_method(klass, "submsg_name=", FieldDescriptor_submsg_name_set, 1); + rb_define_method(klass, "subtype", FieldDescriptor_subtype, 0); + rb_define_method(klass, "get", FieldDescriptor_get, 1); + rb_define_method(klass, "set", FieldDescriptor_set, 2); + cFieldDescriptor = klass; + rb_gc_register_address(&cFieldDescriptor); +} + +/* + * call-seq: + * FieldDescriptor.name => name + * + * Returns the name of this field. + */ +VALUE FieldDescriptor_name(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + return rb_str_maybe_null(upb_fielddef_name(self->fielddef)); +} + +/* + * call-seq: + * FieldDescriptor.name = name + * + * Sets the name of this field. Cannot be called once the containing message + * type, if any, is added to a pool. + */ +VALUE FieldDescriptor_name_set(VALUE _self, VALUE str) { + DEFINE_SELF(FieldDescriptor, self, _self); + upb_fielddef* mut_def = check_field_notfrozen(self->fielddef); + const char* name = get_str(str); + CHECK_UPB(upb_fielddef_setname(mut_def, name, &status), + "Error setting FieldDescriptor name"); + return Qnil; +} + +upb_fieldtype_t ruby_to_fieldtype(VALUE type) { + if (TYPE(type) != T_SYMBOL) { + rb_raise(rb_eArgError, "Expected symbol for field type."); + } + +#define CONVERT(upb, ruby) \ + if (SYM2ID(type) == rb_intern( # ruby )) { \ + return UPB_TYPE_ ## upb; \ + } + + CONVERT(FLOAT, float); + CONVERT(DOUBLE, double); + CONVERT(BOOL, bool); + CONVERT(STRING, string); + CONVERT(BYTES, bytes); + CONVERT(MESSAGE, message); + CONVERT(ENUM, enum); + CONVERT(INT32, int32); + CONVERT(INT64, int64); + CONVERT(UINT32, uint32); + CONVERT(UINT64, uint64); + +#undef CONVERT + + rb_raise(rb_eArgError, "Unknown field type."); + return 0; +} + +VALUE fieldtype_to_ruby(upb_fieldtype_t type) { + switch (type) { +#define CONVERT(upb, ruby) \ + case UPB_TYPE_ ## upb : return ID2SYM(rb_intern( # ruby )); + CONVERT(FLOAT, float); + CONVERT(DOUBLE, double); + CONVERT(BOOL, bool); + CONVERT(STRING, string); + CONVERT(BYTES, bytes); + CONVERT(MESSAGE, message); + CONVERT(ENUM, enum); + CONVERT(INT32, int32); + CONVERT(INT64, int64); + CONVERT(UINT32, uint32); + CONVERT(UINT64, uint64); +#undef CONVERT + } + return Qnil; +} + +upb_descriptortype_t ruby_to_descriptortype(VALUE type) { + if (TYPE(type) != T_SYMBOL) { + rb_raise(rb_eArgError, "Expected symbol for field type."); + } + +#define CONVERT(upb, ruby) \ + if (SYM2ID(type) == rb_intern( # ruby )) { \ + return UPB_DESCRIPTOR_TYPE_ ## upb; \ + } + + CONVERT(FLOAT, float); + CONVERT(DOUBLE, double); + CONVERT(BOOL, bool); + CONVERT(STRING, string); + CONVERT(BYTES, bytes); + CONVERT(MESSAGE, message); + CONVERT(GROUP, group); + CONVERT(ENUM, enum); + CONVERT(INT32, int32); + CONVERT(INT64, int64); + CONVERT(UINT32, uint32); + CONVERT(UINT64, uint64); + CONVERT(SINT32, sint32); + CONVERT(SINT64, sint64); + CONVERT(FIXED32, fixed32); + CONVERT(FIXED64, fixed64); + CONVERT(SFIXED32, sfixed32); + CONVERT(SFIXED64, sfixed64); + +#undef CONVERT + + rb_raise(rb_eArgError, "Unknown field type."); + return 0; +} + +VALUE descriptortype_to_ruby(upb_descriptortype_t type) { + switch (type) { +#define CONVERT(upb, ruby) \ + case UPB_DESCRIPTOR_TYPE_ ## upb : return ID2SYM(rb_intern( # ruby )); + CONVERT(FLOAT, float); + CONVERT(DOUBLE, double); + CONVERT(BOOL, bool); + CONVERT(STRING, string); + CONVERT(BYTES, bytes); + CONVERT(MESSAGE, message); + CONVERT(GROUP, group); + CONVERT(ENUM, enum); + CONVERT(INT32, int32); + CONVERT(INT64, int64); + CONVERT(UINT32, uint32); + CONVERT(UINT64, uint64); + CONVERT(SINT32, sint32); + CONVERT(SINT64, sint64); + CONVERT(FIXED32, fixed32); + CONVERT(FIXED64, fixed64); + CONVERT(SFIXED32, sfixed32); + CONVERT(SFIXED64, sfixed64); +#undef CONVERT + } + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.type => type + * + * Returns this field's type, as a Ruby symbol, or nil if not yet set. + * + * Valid field types are: + * :int32, :int64, :uint32, :uint64, :float, :double, :bool, :string, + * :bytes, :message. + */ +VALUE FieldDescriptor_type(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + if (!upb_fielddef_typeisset(self->fielddef)) { + return Qnil; + } + return descriptortype_to_ruby(upb_fielddef_descriptortype(self->fielddef)); +} + +/* + * call-seq: + * FieldDescriptor.type = type + * + * Sets this field's type. Cannot be called if field is part of a message type + * already in a pool. + */ +VALUE FieldDescriptor_type_set(VALUE _self, VALUE type) { + DEFINE_SELF(FieldDescriptor, self, _self); + upb_fielddef* mut_def = check_field_notfrozen(self->fielddef); + upb_fielddef_setdescriptortype(mut_def, ruby_to_descriptortype(type)); + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.label => label + * + * Returns this field's label (i.e., plurality), as a Ruby symbol. + * + * Valid field labels are: + * :optional, :repeated + */ +VALUE FieldDescriptor_label(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + switch (upb_fielddef_label(self->fielddef)) { +#define CONVERT(upb, ruby) \ + case UPB_LABEL_ ## upb : return ID2SYM(rb_intern( # ruby )); + + CONVERT(OPTIONAL, optional); + CONVERT(REQUIRED, required); + CONVERT(REPEATED, repeated); + +#undef CONVERT + } + + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.label = label + * + * Sets the label on this field. Cannot be called if field is part of a message + * type already in a pool. + */ +VALUE FieldDescriptor_label_set(VALUE _self, VALUE label) { + DEFINE_SELF(FieldDescriptor, self, _self); + upb_fielddef* mut_def = check_field_notfrozen(self->fielddef); + upb_label_t upb_label = -1; + bool converted = false; + + if (TYPE(label) != T_SYMBOL) { + rb_raise(rb_eArgError, "Expected symbol for field label."); + } + +#define CONVERT(upb, ruby) \ + if (SYM2ID(label) == rb_intern( # ruby )) { \ + upb_label = UPB_LABEL_ ## upb; \ + converted = true; \ + } + + CONVERT(OPTIONAL, optional); + CONVERT(REQUIRED, required); + CONVERT(REPEATED, repeated); + +#undef CONVERT + + if (!converted) { + rb_raise(rb_eArgError, "Unknown field label."); + } + + upb_fielddef_setlabel(mut_def, upb_label); + + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.number => number + * + * Returns the tag number for this field. + */ +VALUE FieldDescriptor_number(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + return INT2NUM(upb_fielddef_number(self->fielddef)); +} + +/* + * call-seq: + * FieldDescriptor.number = number + * + * Sets the tag number for this field. Cannot be called if field is part of a + * message type already in a pool. + */ +VALUE FieldDescriptor_number_set(VALUE _self, VALUE number) { + DEFINE_SELF(FieldDescriptor, self, _self); + upb_fielddef* mut_def = check_field_notfrozen(self->fielddef); + CHECK_UPB(upb_fielddef_setnumber(mut_def, NUM2INT(number), &status), + "Error setting field number"); + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.submsg_name => submsg_name + * + * Returns the name of the message or enum type corresponding to this field, if + * it is a message or enum field (respectively), or nil otherwise. This type + * name will be resolved within the context of the pool to which the containing + * message type is added. + */ +VALUE FieldDescriptor_submsg_name(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + if (!upb_fielddef_hassubdef(self->fielddef)) { + return Qnil; + } + return rb_str_maybe_null(upb_fielddef_subdefname(self->fielddef)); +} + +/* + * call-seq: + * FieldDescriptor.submsg_name = submsg_name + * + * Sets the name of the message or enum type corresponding to this field, if it + * is a message or enum field (respectively). This type name will be resolved + * within the context of the pool to which the containing message type is added. + * Cannot be called on field that are not of message or enum type, or on fields + * that are part of a message type already added to a pool. + */ +VALUE FieldDescriptor_submsg_name_set(VALUE _self, VALUE value) { + DEFINE_SELF(FieldDescriptor, self, _self); + upb_fielddef* mut_def = check_field_notfrozen(self->fielddef); + const char* str = get_str(value); + if (!upb_fielddef_hassubdef(self->fielddef)) { + rb_raise(rb_eTypeError, "FieldDescriptor does not have subdef."); + } + CHECK_UPB(upb_fielddef_setsubdefname(mut_def, str, &status), + "Error setting submessage name"); + return Qnil; +} + +/* + * call-seq: + * FieldDescriptor.subtype => message_or_enum_descriptor + * + * Returns the message or enum descriptor corresponding to this field's type if + * it is a message or enum field, respectively, or nil otherwise. Cannot be + * called *until* the containing message type is added to a pool (and thus + * resolved). + */ +VALUE FieldDescriptor_subtype(VALUE _self) { + DEFINE_SELF(FieldDescriptor, self, _self); + const upb_def* def; + + if (!upb_fielddef_hassubdef(self->fielddef)) { + return Qnil; + } + def = upb_fielddef_subdef(self->fielddef); + if (def == NULL) { + return Qnil; + } + return get_def_obj(def); +} + +/* + * call-seq: + * FieldDescriptor.get(message) => value + * + * Returns the value set for this field on the given message. Raises an + * exception if message is of the wrong type. + */ +VALUE FieldDescriptor_get(VALUE _self, VALUE msg_rb) { + DEFINE_SELF(FieldDescriptor, self, _self); + MessageHeader* msg; + TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); + if (msg->descriptor->msgdef != upb_fielddef_containingtype(self->fielddef)) { + rb_raise(rb_eTypeError, "get method called on wrong message type"); + } + return layout_get(msg->descriptor->layout, Message_data(msg), self->fielddef); +} + +/* + * call-seq: + * FieldDescriptor.set(message, value) + * + * Sets the value corresponding to this field to the given value on the given + * message. Raises an exception if message is of the wrong type. Performs the + * ordinary type-checks for field setting. + */ +VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value) { + DEFINE_SELF(FieldDescriptor, self, _self); + MessageHeader* msg; + TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); + if (msg->descriptor->msgdef != upb_fielddef_containingtype(self->fielddef)) { + rb_raise(rb_eTypeError, "set method called on wrong message type"); + } + layout_set(msg->descriptor->layout, Message_data(msg), self->fielddef, value); + return Qnil; +} + +// ----------------------------------------------------------------------------- +// OneofDescriptor. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(OneofDescriptor, "Google::Protobuf::OneofDescriptor"); + +void OneofDescriptor_mark(void* _self) { +} + +void OneofDescriptor_free(void* _self) { + OneofDescriptor* self = _self; + upb_oneofdef_unref(self->oneofdef, &self->oneofdef); + xfree(self); +} + +/* + * call-seq: + * OneofDescriptor.new => oneof_descriptor + * + * Creates a new, empty, oneof descriptor. The oneof may only be modified prior + * to being added to a message descriptor which is subsequently added to a pool. + */ +VALUE OneofDescriptor_alloc(VALUE klass) { + OneofDescriptor* self = ALLOC(OneofDescriptor); + VALUE ret = TypedData_Wrap_Struct(klass, &_OneofDescriptor_type, self); + self->oneofdef = upb_oneofdef_new(&self->oneofdef); + return ret; +} + +void OneofDescriptor_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "OneofDescriptor", rb_cObject); + rb_define_alloc_func(klass, OneofDescriptor_alloc); + rb_define_method(klass, "name", OneofDescriptor_name, 0); + rb_define_method(klass, "name=", OneofDescriptor_name_set, 1); + rb_define_method(klass, "add_field", OneofDescriptor_add_field, 1); + rb_define_method(klass, "each", OneofDescriptor_each, 0); + rb_include_module(klass, rb_mEnumerable); + cOneofDescriptor = klass; + rb_gc_register_address(&cOneofDescriptor); +} + +/* + * call-seq: + * OneofDescriptor.name => name + * + * Returns the name of this oneof. + */ +VALUE OneofDescriptor_name(VALUE _self) { + DEFINE_SELF(OneofDescriptor, self, _self); + return rb_str_maybe_null(upb_oneofdef_name(self->oneofdef)); +} + +/* + * call-seq: + * OneofDescriptor.name = name + * + * Sets a new name for this oneof. The oneof must not have been added to a + * message descriptor yet. + */ +VALUE OneofDescriptor_name_set(VALUE _self, VALUE value) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneofdef* mut_def = check_oneof_notfrozen(self->oneofdef); + const char* str = get_str(value); + CHECK_UPB(upb_oneofdef_setname(mut_def, str, &status), + "Error setting oneof name"); + return Qnil; +} + +/* + * call-seq: + * OneofDescriptor.add_field(field) => nil + * + * Adds a field to this oneof. The field may have been added to this oneof in + * the past, or the message to which this oneof belongs (if any), but may not + * have already been added to any other oneof or message. Otherwise, an + * exception is raised. + * + * All fields added to the oneof via this method will be automatically added to + * the message to which this oneof belongs, if it belongs to one currently, or + * else will be added to any message to which the oneof is later added at the + * time that it is added. + */ +VALUE OneofDescriptor_add_field(VALUE _self, VALUE obj) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneofdef* mut_def = check_oneof_notfrozen(self->oneofdef); + FieldDescriptor* def = ruby_to_FieldDescriptor(obj); + upb_fielddef* mut_field_def = check_field_notfrozen(def->fielddef); + CHECK_UPB( + upb_oneofdef_addfield(mut_def, mut_field_def, NULL, &status), + "Adding field to OneofDescriptor failed"); + add_def_obj(def->fielddef, obj); + return Qnil; +} + +/* + * call-seq: + * OneofDescriptor.each(&block) => nil + * + * Iterates through fields in this oneof, yielding to the block on each one. + */ +VALUE OneofDescriptor_each(VALUE _self, VALUE field) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneof_iter it; + for (upb_oneof_begin(&it, self->oneofdef); + !upb_oneof_done(&it); + upb_oneof_next(&it)) { + const upb_fielddef* f = upb_oneof_iter_field(&it); + VALUE obj = get_def_obj(f); + rb_yield(obj); + } + return Qnil; +} + +// ----------------------------------------------------------------------------- +// EnumDescriptor. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(EnumDescriptor, "Google::Protobuf::EnumDescriptor"); + +void EnumDescriptor_mark(void* _self) { + EnumDescriptor* self = _self; + rb_gc_mark(self->module); +} + +void EnumDescriptor_free(void* _self) { + EnumDescriptor* self = _self; + upb_enumdef_unref(self->enumdef, &self->enumdef); + xfree(self); +} + +/* + * call-seq: + * EnumDescriptor.new => enum_descriptor + * + * Creates a new, empty, enum descriptor. Must be added to a pool before the + * enum type can be used. The enum type may only be modified prior to adding to + * a pool. + */ +VALUE EnumDescriptor_alloc(VALUE klass) { + EnumDescriptor* self = ALLOC(EnumDescriptor); + VALUE ret = TypedData_Wrap_Struct(klass, &_EnumDescriptor_type, self); + self->enumdef = upb_enumdef_new(&self->enumdef); + self->module = Qnil; + return ret; +} + +void EnumDescriptor_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "EnumDescriptor", rb_cObject); + rb_define_alloc_func(klass, EnumDescriptor_alloc); + rb_define_method(klass, "name", EnumDescriptor_name, 0); + rb_define_method(klass, "name=", EnumDescriptor_name_set, 1); + rb_define_method(klass, "add_value", EnumDescriptor_add_value, 2); + rb_define_method(klass, "lookup_name", EnumDescriptor_lookup_name, 1); + rb_define_method(klass, "lookup_value", EnumDescriptor_lookup_value, 1); + rb_define_method(klass, "each", EnumDescriptor_each, 0); + rb_define_method(klass, "enummodule", EnumDescriptor_enummodule, 0); + rb_include_module(klass, rb_mEnumerable); + cEnumDescriptor = klass; + rb_gc_register_address(&cEnumDescriptor); +} + +/* + * call-seq: + * EnumDescriptor.name => name + * + * Returns the name of this enum type. + */ +VALUE EnumDescriptor_name(VALUE _self) { + DEFINE_SELF(EnumDescriptor, self, _self); + return rb_str_maybe_null(upb_enumdef_fullname(self->enumdef)); +} + +/* + * call-seq: + * EnumDescriptor.name = name + * + * Sets the name of this enum type. Cannot be called if the enum type has + * already been added to a pool. + */ +VALUE EnumDescriptor_name_set(VALUE _self, VALUE str) { + DEFINE_SELF(EnumDescriptor, self, _self); + upb_enumdef* mut_def = check_enum_notfrozen(self->enumdef); + const char* name = get_str(str); + CHECK_UPB(upb_enumdef_setfullname(mut_def, name, &status), + "Error setting EnumDescriptor name"); + return Qnil; +} + +/* + * call-seq: + * EnumDescriptor.add_value(key, value) + * + * Adds a new key => value mapping to this enum type. Key must be given as a + * Ruby symbol. Cannot be called if the enum type has already been added to a + * pool. Will raise an exception if the key or value is already in use. + */ +VALUE EnumDescriptor_add_value(VALUE _self, VALUE name, VALUE number) { + DEFINE_SELF(EnumDescriptor, self, _self); + upb_enumdef* mut_def = check_enum_notfrozen(self->enumdef); + const char* name_str = rb_id2name(SYM2ID(name)); + int32_t val = NUM2INT(number); + CHECK_UPB(upb_enumdef_addval(mut_def, name_str, val, &status), + "Error adding value to enum"); + return Qnil; +} + +/* + * call-seq: + * EnumDescriptor.lookup_name(name) => value + * + * Returns the numeric value corresponding to the given key name (as a Ruby + * symbol), or nil if none. + */ +VALUE EnumDescriptor_lookup_name(VALUE _self, VALUE name) { + DEFINE_SELF(EnumDescriptor, self, _self); + const char* name_str= rb_id2name(SYM2ID(name)); + int32_t val = 0; + if (upb_enumdef_ntoiz(self->enumdef, name_str, &val)) { + return INT2NUM(val); + } else { + return Qnil; + } +} + +/* + * call-seq: + * EnumDescriptor.lookup_value(name) => value + * + * Returns the key name (as a Ruby symbol) corresponding to the integer value, + * or nil if none. + */ +VALUE EnumDescriptor_lookup_value(VALUE _self, VALUE number) { + DEFINE_SELF(EnumDescriptor, self, _self); + int32_t val = NUM2INT(number); + const char* name = upb_enumdef_iton(self->enumdef, val); + if (name != NULL) { + return ID2SYM(rb_intern(name)); + } else { + return Qnil; + } +} + +/* + * call-seq: + * EnumDescriptor.each(&block) + * + * Iterates over key => value mappings in this enum's definition, yielding to + * the block with (key, value) arguments for each one. + */ +VALUE EnumDescriptor_each(VALUE _self) { + DEFINE_SELF(EnumDescriptor, self, _self); + + upb_enum_iter it; + for (upb_enum_begin(&it, self->enumdef); + !upb_enum_done(&it); + upb_enum_next(&it)) { + VALUE key = ID2SYM(rb_intern(upb_enum_iter_name(&it))); + VALUE number = INT2NUM(upb_enum_iter_number(&it)); + rb_yield_values(2, key, number); + } + + return Qnil; +} + +/* + * call-seq: + * EnumDescriptor.enummodule => module + * + * Returns the Ruby module corresponding to this enum type. Cannot be called + * until the enum descriptor has been added to a pool. + */ +VALUE EnumDescriptor_enummodule(VALUE _self) { + DEFINE_SELF(EnumDescriptor, self, _self); + if (!upb_def_isfrozen((const upb_def*)self->enumdef)) { + rb_raise(rb_eRuntimeError, + "Cannot fetch enum module from an EnumDescriptor not yet " + "in a pool."); + } + if (self->module == Qnil) { + self->module = build_module_from_enumdesc(self); + } + return self->module; +} + +// ----------------------------------------------------------------------------- +// MessageBuilderContext. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(MessageBuilderContext, + "Google::Protobuf::Internal::MessageBuilderContext"); + +void MessageBuilderContext_mark(void* _self) { + MessageBuilderContext* self = _self; + rb_gc_mark(self->descriptor); + rb_gc_mark(self->builder); +} + +void MessageBuilderContext_free(void* _self) { + MessageBuilderContext* self = _self; + xfree(self); +} + +VALUE MessageBuilderContext_alloc(VALUE klass) { + MessageBuilderContext* self = ALLOC(MessageBuilderContext); + VALUE ret = TypedData_Wrap_Struct( + klass, &_MessageBuilderContext_type, self); + self->descriptor = Qnil; + self->builder = Qnil; + return ret; +} + +void MessageBuilderContext_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "MessageBuilderContext", rb_cObject); + rb_define_alloc_func(klass, MessageBuilderContext_alloc); + rb_define_method(klass, "initialize", + MessageBuilderContext_initialize, 2); + rb_define_method(klass, "optional", MessageBuilderContext_optional, -1); + rb_define_method(klass, "required", MessageBuilderContext_required, -1); + rb_define_method(klass, "repeated", MessageBuilderContext_repeated, -1); + rb_define_method(klass, "map", MessageBuilderContext_map, -1); + rb_define_method(klass, "oneof", MessageBuilderContext_oneof, 1); + cMessageBuilderContext = klass; + rb_gc_register_address(&cMessageBuilderContext); +} + +/* + * call-seq: + * MessageBuilderContext.new(desc, builder) => context + * + * Create a new message builder context around the given message descriptor and + * builder context. This class is intended to serve as a DSL context to be used + * with #instance_eval. + */ +VALUE MessageBuilderContext_initialize(VALUE _self, + VALUE msgdef, + VALUE builder) { + DEFINE_SELF(MessageBuilderContext, self, _self); + self->descriptor = msgdef; + self->builder = builder; + return Qnil; +} + +static VALUE msgdef_add_field(VALUE msgdef, + const char* label, VALUE name, + VALUE type, VALUE number, + VALUE type_class) { + VALUE fielddef = rb_class_new_instance(0, NULL, cFieldDescriptor); + VALUE name_str = rb_str_new2(rb_id2name(SYM2ID(name))); + + rb_funcall(fielddef, rb_intern("label="), 1, ID2SYM(rb_intern(label))); + rb_funcall(fielddef, rb_intern("name="), 1, name_str); + rb_funcall(fielddef, rb_intern("type="), 1, type); + rb_funcall(fielddef, rb_intern("number="), 1, number); + + if (type_class != Qnil) { + if (TYPE(type_class) != T_STRING) { + rb_raise(rb_eArgError, "Expected string for type class"); + } + // Make it an absolute type name by prepending a dot. + type_class = rb_str_append(rb_str_new2("."), type_class); + rb_funcall(fielddef, rb_intern("submsg_name="), 1, type_class); + } + + rb_funcall(msgdef, rb_intern("add_field"), 1, fielddef); + return fielddef; +} + +/* + * call-seq: + * MessageBuilderContext.optional(name, type, number, type_class = nil) + * + * Defines a new optional field on this message type with the given type, tag + * number, and type class (for message and enum fields). The type must be a Ruby + * symbol (as accepted by FieldDescriptor#type=) and the type_class must be a + * string, if present (as accepted by FieldDescriptor#submsg_name=). + */ +VALUE MessageBuilderContext_optional(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE name, type, number, type_class; + + if (argc < 3) { + rb_raise(rb_eArgError, "Expected at least 3 arguments."); + } + name = argv[0]; + type = argv[1]; + number = argv[2]; + type_class = (argc > 3) ? argv[3] : Qnil; + + return msgdef_add_field(self->descriptor, "optional", + name, type, number, type_class); +} + +/* + * call-seq: + * MessageBuilderContext.required(name, type, number, type_class = nil) + * + * Defines a new required field on this message type with the given type, tag + * number, and type class (for message and enum fields). The type must be a Ruby + * symbol (as accepted by FieldDescriptor#type=) and the type_class must be a + * string, if present (as accepted by FieldDescriptor#submsg_name=). + * + * Proto3 does not have required fields, but this method exists for + * completeness. Any attempt to add a message type with required fields to a + * pool will currently result in an error. + */ +VALUE MessageBuilderContext_required(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE name, type, number, type_class; + + if (argc < 3) { + rb_raise(rb_eArgError, "Expected at least 3 arguments."); + } + name = argv[0]; + type = argv[1]; + number = argv[2]; + type_class = (argc > 3) ? argv[3] : Qnil; + + return msgdef_add_field(self->descriptor, "required", + name, type, number, type_class); +} + +/* + * call-seq: + * MessageBuilderContext.repeated(name, type, number, type_class = nil) + * + * Defines a new repeated field on this message type with the given type, tag + * number, and type class (for message and enum fields). The type must be a Ruby + * symbol (as accepted by FieldDescriptor#type=) and the type_class must be a + * string, if present (as accepted by FieldDescriptor#submsg_name=). + */ +VALUE MessageBuilderContext_repeated(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE name, type, number, type_class; + + if (argc < 3) { + rb_raise(rb_eArgError, "Expected at least 3 arguments."); + } + name = argv[0]; + type = argv[1]; + number = argv[2]; + type_class = (argc > 3) ? argv[3] : Qnil; + + return msgdef_add_field(self->descriptor, "repeated", + name, type, number, type_class); +} + +/* + * call-seq: + * MessageBuilderContext.map(name, key_type, value_type, number, + * value_type_class = nil) + * + * Defines a new map field on this message type with the given key and value + * types, tag number, and type class (for message and enum value types). The key + * type must be :int32/:uint32/:int64/:uint64, :bool, or :string. The value type + * type must be a Ruby symbol (as accepted by FieldDescriptor#type=) and the + * type_class must be a string, if present (as accepted by + * FieldDescriptor#submsg_name=). + */ +VALUE MessageBuilderContext_map(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE name, key_type, value_type, number, type_class; + VALUE mapentry_desc, mapentry_desc_name; + + if (argc < 4) { + rb_raise(rb_eArgError, "Expected at least 4 arguments."); + } + name = argv[0]; + key_type = argv[1]; + value_type = argv[2]; + number = argv[3]; + type_class = (argc > 4) ? argv[4] : Qnil; + + // Validate the key type. We can't accept enums, messages, or floats/doubles + // as map keys. (We exclude these explicitly, and the field-descriptor setter + // below then ensures that the type is one of the remaining valid options.) + if (SYM2ID(key_type) == rb_intern("float") || + SYM2ID(key_type) == rb_intern("double") || + SYM2ID(key_type) == rb_intern("enum") || + SYM2ID(key_type) == rb_intern("message")) { + rb_raise(rb_eArgError, + "Cannot add a map field with a float, double, enum, or message " + "type."); + } + + // Create a new message descriptor for the map entry message, and create a + // repeated submessage field here with that type. + mapentry_desc = rb_class_new_instance(0, NULL, cDescriptor); + mapentry_desc_name = rb_funcall(self->descriptor, rb_intern("name"), 0); + mapentry_desc_name = rb_str_cat2(mapentry_desc_name, "_MapEntry_"); + mapentry_desc_name = rb_str_cat2(mapentry_desc_name, + rb_id2name(SYM2ID(name))); + Descriptor_name_set(mapentry_desc, mapentry_desc_name); + + { + // The 'mapentry' attribute has no Ruby setter because we do not want the + // user attempting to DIY the setup below; we want to ensure that the fields + // are correct. So we reach into the msgdef here to set the bit manually. + Descriptor* mapentry_desc_self = ruby_to_Descriptor(mapentry_desc); + upb_msgdef_setmapentry((upb_msgdef*)mapentry_desc_self->msgdef, true); + } + + { + // optional <type> key = 1; + VALUE key_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + FieldDescriptor_name_set(key_field, rb_str_new2("key")); + FieldDescriptor_label_set(key_field, ID2SYM(rb_intern("optional"))); + FieldDescriptor_number_set(key_field, INT2NUM(1)); + FieldDescriptor_type_set(key_field, key_type); + Descriptor_add_field(mapentry_desc, key_field); + } + + { + // optional <type> value = 2; + VALUE value_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + FieldDescriptor_name_set(value_field, rb_str_new2("value")); + FieldDescriptor_label_set(value_field, ID2SYM(rb_intern("optional"))); + FieldDescriptor_number_set(value_field, INT2NUM(2)); + FieldDescriptor_type_set(value_field, value_type); + if (type_class != Qnil) { + VALUE submsg_name = rb_str_new2("."); // prepend '.' to make absolute. + submsg_name = rb_str_append(submsg_name, type_class); + FieldDescriptor_submsg_name_set(value_field, submsg_name); + } + Descriptor_add_field(mapentry_desc, value_field); + } + + { + // Add the map-entry message type to the current builder, and use the type + // to create the map field itself. + Builder* builder_self = ruby_to_Builder(self->builder); + rb_ary_push(builder_self->pending_list, mapentry_desc); + } + + { + VALUE map_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + VALUE name_str = rb_str_new2(rb_id2name(SYM2ID(name))); + VALUE submsg_name; + + FieldDescriptor_name_set(map_field, name_str); + FieldDescriptor_number_set(map_field, number); + FieldDescriptor_label_set(map_field, ID2SYM(rb_intern("repeated"))); + FieldDescriptor_type_set(map_field, ID2SYM(rb_intern("message"))); + submsg_name = rb_str_new2("."); // prepend '.' to make name absolute. + submsg_name = rb_str_append(submsg_name, mapentry_desc_name); + FieldDescriptor_submsg_name_set(map_field, submsg_name); + Descriptor_add_field(self->descriptor, map_field); + } + + return Qnil; +} + +/* + * call-seq: + * MessageBuilderContext.oneof(name, &block) => nil + * + * Creates a new OneofDescriptor with the given name, creates a + * OneofBuilderContext attached to that OneofDescriptor, evaluates the given + * block in the context of that OneofBuilderContext with #instance_eval, and + * then adds the oneof to the message. + * + * This is the recommended, idiomatic way to build oneof definitions. + */ +VALUE MessageBuilderContext_oneof(VALUE _self, VALUE name) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE oneofdef = rb_class_new_instance(0, NULL, cOneofDescriptor); + VALUE args[2] = { oneofdef, self->builder }; + VALUE ctx = rb_class_new_instance(2, args, cOneofBuilderContext); + VALUE block = rb_block_proc(); + VALUE name_str = rb_str_new2(rb_id2name(SYM2ID(name))); + rb_funcall(oneofdef, rb_intern("name="), 1, name_str); + rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); + Descriptor_add_oneof(self->descriptor, oneofdef); + + return Qnil; +} + +// ----------------------------------------------------------------------------- +// OneofBuilderContext. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(OneofBuilderContext, + "Google::Protobuf::Internal::OneofBuilderContext"); + +void OneofBuilderContext_mark(void* _self) { + OneofBuilderContext* self = _self; + rb_gc_mark(self->descriptor); + rb_gc_mark(self->builder); +} + +void OneofBuilderContext_free(void* _self) { + OneofBuilderContext* self = _self; + xfree(self); +} + +VALUE OneofBuilderContext_alloc(VALUE klass) { + OneofBuilderContext* self = ALLOC(OneofBuilderContext); + VALUE ret = TypedData_Wrap_Struct( + klass, &_OneofBuilderContext_type, self); + self->descriptor = Qnil; + self->builder = Qnil; + return ret; +} + +void OneofBuilderContext_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "OneofBuilderContext", rb_cObject); + rb_define_alloc_func(klass, OneofBuilderContext_alloc); + rb_define_method(klass, "initialize", + OneofBuilderContext_initialize, 2); + rb_define_method(klass, "optional", OneofBuilderContext_optional, -1); + cOneofBuilderContext = klass; + rb_gc_register_address(&cOneofBuilderContext); +} + +/* + * call-seq: + * OneofBuilderContext.new(desc, builder) => context + * + * Create a new oneof builder context around the given oneof descriptor and + * builder context. This class is intended to serve as a DSL context to be used + * with #instance_eval. + */ +VALUE OneofBuilderContext_initialize(VALUE _self, + VALUE oneofdef, + VALUE builder) { + DEFINE_SELF(OneofBuilderContext, self, _self); + self->descriptor = oneofdef; + self->builder = builder; + return Qnil; +} + +/* + * call-seq: + * OneofBuilderContext.optional(name, type, number, type_class = nil) + * + * Defines a new optional field in this oneof with the given type, tag number, + * and type class (for message and enum fields). The type must be a Ruby symbol + * (as accepted by FieldDescriptor#type=) and the type_class must be a string, + * if present (as accepted by FieldDescriptor#submsg_name=). + */ +VALUE OneofBuilderContext_optional(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(OneofBuilderContext, self, _self); + VALUE name, type, number, type_class; + + if (argc < 3) { + rb_raise(rb_eArgError, "Expected at least 3 arguments."); + } + name = argv[0]; + type = argv[1]; + number = argv[2]; + type_class = (argc > 3) ? argv[3] : Qnil; + + return msgdef_add_field(self->descriptor, "optional", + name, type, number, type_class); +} + +// ----------------------------------------------------------------------------- +// EnumBuilderContext. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(EnumBuilderContext, + "Google::Protobuf::Internal::EnumBuilderContext"); + +void EnumBuilderContext_mark(void* _self) { + EnumBuilderContext* self = _self; + rb_gc_mark(self->enumdesc); +} + +void EnumBuilderContext_free(void* _self) { + EnumBuilderContext* self = _self; + xfree(self); +} + +VALUE EnumBuilderContext_alloc(VALUE klass) { + EnumBuilderContext* self = ALLOC(EnumBuilderContext); + VALUE ret = TypedData_Wrap_Struct( + klass, &_EnumBuilderContext_type, self); + self->enumdesc = Qnil; + return ret; +} + +void EnumBuilderContext_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "EnumBuilderContext", rb_cObject); + rb_define_alloc_func(klass, EnumBuilderContext_alloc); + rb_define_method(klass, "initialize", + EnumBuilderContext_initialize, 1); + rb_define_method(klass, "value", EnumBuilderContext_value, 2); + cEnumBuilderContext = klass; + rb_gc_register_address(&cEnumBuilderContext); +} + +/* + * call-seq: + * EnumBuilderContext.new(enumdesc) => context + * + * Create a new builder context around the given enum descriptor. This class is + * intended to serve as a DSL context to be used with #instance_eval. + */ +VALUE EnumBuilderContext_initialize(VALUE _self, VALUE enumdef) { + DEFINE_SELF(EnumBuilderContext, self, _self); + self->enumdesc = enumdef; + return Qnil; +} + +static VALUE enumdef_add_value(VALUE enumdef, + VALUE name, VALUE number) { + rb_funcall(enumdef, rb_intern("add_value"), 2, name, number); + return Qnil; +} + +/* + * call-seq: + * EnumBuilder.add_value(name, number) + * + * Adds the given name => number mapping to the enum type. Name must be a Ruby + * symbol. + */ +VALUE EnumBuilderContext_value(VALUE _self, VALUE name, VALUE number) { + DEFINE_SELF(EnumBuilderContext, self, _self); + return enumdef_add_value(self->enumdesc, name, number); +} + +// ----------------------------------------------------------------------------- +// Builder. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(Builder, "Google::Protobuf::Internal::Builder"); + +void Builder_mark(void* _self) { + Builder* self = _self; + rb_gc_mark(self->pending_list); +} + +void Builder_free(void* _self) { + Builder* self = _self; + xfree(self->defs); + xfree(self); +} + +/* + * call-seq: + * Builder.new => builder + * + * Creates a new Builder. A Builder can accumulate a set of new message and enum + * descriptors and atomically register them into a pool in a way that allows for + * (co)recursive type references. + */ +VALUE Builder_alloc(VALUE klass) { + Builder* self = ALLOC(Builder); + VALUE ret = TypedData_Wrap_Struct( + klass, &_Builder_type, self); + self->pending_list = rb_ary_new(); + self->defs = NULL; + return ret; +} + +void Builder_register(VALUE module) { + VALUE klass = rb_define_class_under(module, "Builder", rb_cObject); + rb_define_alloc_func(klass, Builder_alloc); + rb_define_method(klass, "add_message", Builder_add_message, 1); + rb_define_method(klass, "add_enum", Builder_add_enum, 1); + rb_define_method(klass, "finalize_to_pool", Builder_finalize_to_pool, 1); + cBuilder = klass; + rb_gc_register_address(&cBuilder); +} + +/* + * call-seq: + * Builder.add_message(name, &block) + * + * Creates a new, empty descriptor with the given name, and invokes the block in + * the context of a MessageBuilderContext on that descriptor. The block can then + * call, e.g., MessageBuilderContext#optional and MessageBuilderContext#repeated + * methods to define the message fields. + * + * This is the recommended, idiomatic way to build message definitions. + */ +VALUE Builder_add_message(VALUE _self, VALUE name) { + DEFINE_SELF(Builder, self, _self); + VALUE msgdef = rb_class_new_instance(0, NULL, cDescriptor); + VALUE args[2] = { msgdef, _self }; + VALUE ctx = rb_class_new_instance(2, args, cMessageBuilderContext); + VALUE block = rb_block_proc(); + rb_funcall(msgdef, rb_intern("name="), 1, name); + rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); + rb_ary_push(self->pending_list, msgdef); + return Qnil; +} + +/* + * call-seq: + * Builder.add_enum(name, &block) + * + * Creates a new, empty enum descriptor with the given name, and invokes the + * block in the context of an EnumBuilderContext on that descriptor. The block + * can then call EnumBuilderContext#add_value to define the enum values. + * + * This is the recommended, idiomatic way to build enum definitions. + */ +VALUE Builder_add_enum(VALUE _self, VALUE name) { + DEFINE_SELF(Builder, self, _self); + VALUE enumdef = rb_class_new_instance(0, NULL, cEnumDescriptor); + VALUE ctx = rb_class_new_instance(1, &enumdef, cEnumBuilderContext); + VALUE block = rb_block_proc(); + rb_funcall(enumdef, rb_intern("name="), 1, name); + rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); + rb_ary_push(self->pending_list, enumdef); + return Qnil; +} + +static void validate_msgdef(const upb_msgdef* msgdef) { + // Verify that no required fields exist. proto3 does not support these. + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + if (upb_fielddef_label(field) == UPB_LABEL_REQUIRED) { + rb_raise(rb_eTypeError, "Required fields are unsupported in proto3."); + } + } +} + +static void validate_enumdef(const upb_enumdef* enumdef) { + // Verify that an entry exists with integer value 0. (This is the default + // value.) + const char* lookup = upb_enumdef_iton(enumdef, 0); + if (lookup == NULL) { + rb_raise(rb_eTypeError, + "Enum definition does not contain a value for '0'."); + } +} + +/* + * call-seq: + * Builder.finalize_to_pool(pool) + * + * Adds all accumulated message and enum descriptors created in this builder + * context to the given pool. The operation occurs atomically, and all + * descriptors can refer to each other (including in cycles). This is the only + * way to build (co)recursive message definitions. + * + * This method is usually called automatically by DescriptorPool#build after it + * invokes the given user block in the context of the builder. The user should + * not normally need to call this manually because a Builder is not normally + * created manually. + */ +VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb) { + DEFINE_SELF(Builder, self, _self); + + DescriptorPool* pool = ruby_to_DescriptorPool(pool_rb); + + REALLOC_N(self->defs, upb_def*, RARRAY_LEN(self->pending_list)); + + for (int i = 0; i < RARRAY_LEN(self->pending_list); i++) { + VALUE def_rb = rb_ary_entry(self->pending_list, i); + if (CLASS_OF(def_rb) == cDescriptor) { + self->defs[i] = (upb_def*)ruby_to_Descriptor(def_rb)->msgdef; + validate_msgdef((const upb_msgdef*)self->defs[i]); + } else if (CLASS_OF(def_rb) == cEnumDescriptor) { + self->defs[i] = (upb_def*)ruby_to_EnumDescriptor(def_rb)->enumdef; + validate_enumdef((const upb_enumdef*)self->defs[i]); + } + } + + CHECK_UPB(upb_symtab_add(pool->symtab, (upb_def**)self->defs, + RARRAY_LEN(self->pending_list), NULL, &status), + "Unable to add defs to DescriptorPool"); + + for (int i = 0; i < RARRAY_LEN(self->pending_list); i++) { + VALUE def_rb = rb_ary_entry(self->pending_list, i); + add_def_obj(self->defs[i], def_rb); + } + + self->pending_list = rb_ary_new(); + return Qnil; +} diff --git a/ruby/ext/google/protobuf_c/encode_decode.c b/ruby/ext/google/protobuf_c/encode_decode.c new file mode 100644 index 00000000..f6bea50f --- /dev/null +++ b/ruby/ext/google/protobuf_c/encode_decode.c @@ -0,0 +1,1233 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// This function is equivalent to rb_str_cat(), but unlike the real +// rb_str_cat(), it doesn't leak memory in some versions of Ruby. +// For more information, see: +// https://bugs.ruby-lang.org/issues/11328 +VALUE noleak_rb_str_cat(VALUE rb_str, const char *str, long len) { + char *p; + size_t oldlen = RSTRING_LEN(rb_str); + rb_str_modify_expand(rb_str, len); + p = RSTRING_PTR(rb_str); + memcpy(p + oldlen, str, len); + rb_str_set_len(rb_str, oldlen + len); + return rb_str; +} + +// ----------------------------------------------------------------------------- +// Parsing. +// ----------------------------------------------------------------------------- + +#define DEREF(msg, ofs, type) *(type*)(((uint8_t *)msg) + ofs) + +// Creates a handlerdata that simply contains the offset for this field. +static const void* newhandlerdata(upb_handlers* h, uint32_t ofs) { + size_t* hd_ofs = ALLOC(size_t); + *hd_ofs = ofs; + upb_handlers_addcleanup(h, hd_ofs, free); + return hd_ofs; +} + +typedef struct { + size_t ofs; + const upb_msgdef *md; +} submsg_handlerdata_t; + +// Creates a handlerdata that contains offset and submessage type information. +static const void *newsubmsghandlerdata(upb_handlers* h, uint32_t ofs, + const upb_fielddef* f) { + submsg_handlerdata_t *hd = ALLOC(submsg_handlerdata_t); + hd->ofs = ofs; + hd->md = upb_fielddef_msgsubdef(f); + upb_handlers_addcleanup(h, hd, free); + return hd; +} + +typedef struct { + size_t ofs; // union data slot + size_t case_ofs; // oneof_case field + uint32_t oneof_case_num; // oneof-case number to place in oneof_case field + const upb_msgdef *md; // msgdef, for oneof submessage handler +} oneof_handlerdata_t; + +static const void *newoneofhandlerdata(upb_handlers *h, + uint32_t ofs, + uint32_t case_ofs, + const upb_fielddef *f) { + oneof_handlerdata_t *hd = ALLOC(oneof_handlerdata_t); + hd->ofs = ofs; + hd->case_ofs = case_ofs; + // We reuse the field tag number as a oneof union discriminant tag. Note that + // we don't expose these numbers to the user, so the only requirement is that + // we have some unique ID for each union case/possibility. The field tag + // numbers are already present and are easy to use so there's no reason to + // create a separate ID space. In addition, using the field tag number here + // lets us easily look up the field in the oneof accessor. + hd->oneof_case_num = upb_fielddef_number(f); + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE) { + hd->md = upb_fielddef_msgsubdef(f); + } else { + hd->md = NULL; + } + upb_handlers_addcleanup(h, hd, free); + return hd; +} + +// A handler that starts a repeated field. Gets the Repeated*Field instance for +// this field (such an instance always exists even in an empty message). +static void *startseq_handler(void* closure, const void* hd) { + MessageHeader* msg = closure; + const size_t *ofs = hd; + return (void*)DEREF(msg, *ofs, VALUE); +} + +// Handlers that append primitive values to a repeated field. +#define DEFINE_APPEND_HANDLER(type, ctype) \ + static bool append##type##_handler(void *closure, const void *hd, \ + ctype val) { \ + VALUE ary = (VALUE)closure; \ + RepeatedField_push_native(ary, &val); \ + return true; \ + } + +DEFINE_APPEND_HANDLER(bool, bool) +DEFINE_APPEND_HANDLER(int32, int32_t) +DEFINE_APPEND_HANDLER(uint32, uint32_t) +DEFINE_APPEND_HANDLER(float, float) +DEFINE_APPEND_HANDLER(int64, int64_t) +DEFINE_APPEND_HANDLER(uint64, uint64_t) +DEFINE_APPEND_HANDLER(double, double) + +// Appends a string to a repeated field. +static void* appendstr_handler(void *closure, + const void *hd, + size_t size_hint) { + VALUE ary = (VALUE)closure; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyStringUtf8Encoding); + RepeatedField_push(ary, str); + return (void*)str; +} + +// Appends a 'bytes' string to a repeated field. +static void* appendbytes_handler(void *closure, + const void *hd, + size_t size_hint) { + VALUE ary = (VALUE)closure; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyString8bitEncoding); + RepeatedField_push(ary, str); + return (void*)str; +} + +// Sets a non-repeated string field in a message. +static void* str_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const size_t *ofs = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyStringUtf8Encoding); + DEREF(msg, *ofs, VALUE) = str; + return (void*)str; +} + +// Sets a non-repeated 'bytes' field in a message. +static void* bytes_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const size_t *ofs = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyString8bitEncoding); + DEREF(msg, *ofs, VALUE) = str; + return (void*)str; +} + +static size_t stringdata_handler(void* closure, const void* hd, + const char* str, size_t len, + const upb_bufhandle* handle) { + VALUE rb_str = (VALUE)closure; + noleak_rb_str_cat(rb_str, str, len); + return len; +} + +// Appends a submessage to a repeated field (a regular Ruby array for now). +static void *appendsubmsg_handler(void *closure, const void *hd) { + VALUE ary = (VALUE)closure; + const submsg_handlerdata_t *submsgdata = hd; + VALUE subdesc = + get_def_obj((void*)submsgdata->md); + VALUE subklass = Descriptor_msgclass(subdesc); + MessageHeader* submsg; + + VALUE submsg_rb = rb_class_new_instance(0, NULL, subklass); + RepeatedField_push(ary, submsg_rb); + + TypedData_Get_Struct(submsg_rb, MessageHeader, &Message_type, submsg); + return submsg; +} + +// Sets a non-repeated submessage field in a message. +static void *submsg_handler(void *closure, const void *hd) { + MessageHeader* msg = closure; + const submsg_handlerdata_t* submsgdata = hd; + VALUE subdesc = + get_def_obj((void*)submsgdata->md); + VALUE subklass = Descriptor_msgclass(subdesc); + VALUE submsg_rb; + MessageHeader* submsg; + + if (DEREF(msg, submsgdata->ofs, VALUE) == Qnil) { + DEREF(msg, submsgdata->ofs, VALUE) = + rb_class_new_instance(0, NULL, subklass); + } + + submsg_rb = DEREF(msg, submsgdata->ofs, VALUE); + TypedData_Get_Struct(submsg_rb, MessageHeader, &Message_type, submsg); + return submsg; +} + +// Handler data for startmap/endmap handlers. +typedef struct { + size_t ofs; + upb_fieldtype_t key_field_type; + upb_fieldtype_t value_field_type; + + // We know that we can hold this reference because the handlerdata has the + // same lifetime as the upb_handlers struct, and the upb_handlers struct holds + // a reference to the upb_msgdef, which in turn has references to its subdefs. + const upb_def* value_field_subdef; +} map_handlerdata_t; + +// Temporary frame for map parsing: at the beginning of a map entry message, a +// submsg handler allocates a frame to hold (i) a reference to the Map object +// into which this message will be inserted and (ii) storage slots to +// temporarily hold the key and value for this map entry until the end of the +// submessage. When the submessage ends, another handler is called to insert the +// value into the map. +typedef struct { + VALUE map; + char key_storage[NATIVE_SLOT_MAX_SIZE]; + char value_storage[NATIVE_SLOT_MAX_SIZE]; +} map_parse_frame_t; + +// Handler to begin a map entry: allocates a temporary frame. This is the +// 'startsubmsg' handler on the msgdef that contains the map field. +static void *startmapentry_handler(void *closure, const void *hd) { + MessageHeader* msg = closure; + const map_handlerdata_t* mapdata = hd; + VALUE map_rb = DEREF(msg, mapdata->ofs, VALUE); + + map_parse_frame_t* frame = ALLOC(map_parse_frame_t); + frame->map = map_rb; + + native_slot_init(mapdata->key_field_type, &frame->key_storage); + native_slot_init(mapdata->value_field_type, &frame->value_storage); + + return frame; +} + +// Handler to end a map entry: inserts the value defined during the message into +// the map. This is the 'endmsg' handler on the map entry msgdef. +static bool endmap_handler(void *closure, const void *hd, upb_status* s) { + map_parse_frame_t* frame = closure; + const map_handlerdata_t* mapdata = hd; + + VALUE key = native_slot_get( + mapdata->key_field_type, Qnil, + &frame->key_storage); + + VALUE value_field_typeclass = Qnil; + VALUE value; + + if (mapdata->value_field_type == UPB_TYPE_MESSAGE || + mapdata->value_field_type == UPB_TYPE_ENUM) { + value_field_typeclass = get_def_obj(mapdata->value_field_subdef); + } + + value = native_slot_get( + mapdata->value_field_type, value_field_typeclass, + &frame->value_storage); + + Map_index_set(frame->map, key, value); + free(frame); + + return true; +} + +// Allocates a new map_handlerdata_t given the map entry message definition. If +// the offset of the field within the parent message is also given, that is +// added to the handler data as well. Note that this is called *twice* per map +// field: once in the parent message handler setup when setting the startsubmsg +// handler and once in the map entry message handler setup when setting the +// key/value and endmsg handlers. The reason is that there is no easy way to +// pass the handlerdata down to the sub-message handler setup. +static map_handlerdata_t* new_map_handlerdata( + size_t ofs, + const upb_msgdef* mapentry_def, + Descriptor* desc) { + const upb_fielddef* key_field; + const upb_fielddef* value_field; + map_handlerdata_t* hd = ALLOC(map_handlerdata_t); + hd->ofs = ofs; + key_field = upb_msgdef_itof(mapentry_def, MAP_KEY_FIELD); + assert(key_field != NULL); + hd->key_field_type = upb_fielddef_type(key_field); + value_field = upb_msgdef_itof(mapentry_def, MAP_VALUE_FIELD); + assert(value_field != NULL); + hd->value_field_type = upb_fielddef_type(value_field); + hd->value_field_subdef = upb_fielddef_subdef(value_field); + + return hd; +} + +// Handlers that set primitive values in oneofs. +#define DEFINE_ONEOF_HANDLER(type, ctype) \ + static bool oneof##type##_handler(void *closure, const void *hd, \ + ctype val) { \ + const oneof_handlerdata_t *oneofdata = hd; \ + DEREF(closure, oneofdata->case_ofs, uint32_t) = \ + oneofdata->oneof_case_num; \ + DEREF(closure, oneofdata->ofs, ctype) = val; \ + return true; \ + } + +DEFINE_ONEOF_HANDLER(bool, bool) +DEFINE_ONEOF_HANDLER(int32, int32_t) +DEFINE_ONEOF_HANDLER(uint32, uint32_t) +DEFINE_ONEOF_HANDLER(float, float) +DEFINE_ONEOF_HANDLER(int64, int64_t) +DEFINE_ONEOF_HANDLER(uint64, uint64_t) +DEFINE_ONEOF_HANDLER(double, double) + +#undef DEFINE_ONEOF_HANDLER + +// Handlers for strings in a oneof. +static void *oneofstr_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyStringUtf8Encoding); + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + DEREF(msg, oneofdata->ofs, VALUE) = str; + return (void*)str; +} + +static void *oneofbytes_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyString8bitEncoding); + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + DEREF(msg, oneofdata->ofs, VALUE) = str; + return (void*)str; +} + +// Handler for a submessage field in a oneof. +static void *oneofsubmsg_handler(void *closure, + const void *hd) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + uint32_t oldcase = DEREF(msg, oneofdata->case_ofs, uint32_t); + + VALUE subdesc = + get_def_obj((void*)oneofdata->md); + VALUE subklass = Descriptor_msgclass(subdesc); + VALUE submsg_rb; + MessageHeader* submsg; + + if (oldcase != oneofdata->oneof_case_num || + DEREF(msg, oneofdata->ofs, VALUE) == Qnil) { + DEREF(msg, oneofdata->ofs, VALUE) = + rb_class_new_instance(0, NULL, subklass); + } + // Set the oneof case *after* allocating the new class instance -- otherwise, + // if the Ruby GC is invoked as part of a call into the VM, it might invoke + // our mark routines, and our mark routines might see the case value + // indicating a VALUE is present and expect a valid VALUE. See comment in + // layout_set() for more detail: basically, the change to the value and the + // case must be atomic w.r.t. the Ruby VM. + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + + submsg_rb = DEREF(msg, oneofdata->ofs, VALUE); + TypedData_Get_Struct(submsg_rb, MessageHeader, &Message_type, submsg); + return submsg; +} + +// Set up handlers for a repeated field. +static void add_handlers_for_repeated_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset) { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newhandlerdata(h, offset)); + upb_handlers_setstartseq(h, f, startseq_handler, &attr); + upb_handlerattr_uninit(&attr); + + switch (upb_fielddef_type(f)) { + +#define SET_HANDLER(utype, ltype) \ + case utype: \ + upb_handlers_set##ltype(h, f, append##ltype##_handler, NULL); \ + break; + + SET_HANDLER(UPB_TYPE_BOOL, bool); + SET_HANDLER(UPB_TYPE_INT32, int32); + SET_HANDLER(UPB_TYPE_UINT32, uint32); + SET_HANDLER(UPB_TYPE_ENUM, int32); + SET_HANDLER(UPB_TYPE_FLOAT, float); + SET_HANDLER(UPB_TYPE_INT64, int64); + SET_HANDLER(UPB_TYPE_UINT64, uint64); + SET_HANDLER(UPB_TYPE_DOUBLE, double); + +#undef SET_HANDLER + + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; + upb_handlers_setstartstr(h, f, is_bytes ? + appendbytes_handler : appendstr_handler, + NULL); + upb_handlers_setstring(h, f, stringdata_handler, NULL); + break; + } + case UPB_TYPE_MESSAGE: { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, 0, f)); + upb_handlers_setstartsubmsg(h, f, appendsubmsg_handler, &attr); + upb_handlerattr_uninit(&attr); + break; + } + } +} + +// Set up handlers for a singular field. +static void add_handlers_for_singular_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_UINT32: + case UPB_TYPE_ENUM: + case UPB_TYPE_FLOAT: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT64: + case UPB_TYPE_DOUBLE: + upb_shim_set(h, f, offset, -1); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newhandlerdata(h, offset)); + upb_handlers_setstartstr(h, f, + is_bytes ? bytes_handler : str_handler, + &attr); + upb_handlers_setstring(h, f, stringdata_handler, &attr); + upb_handlerattr_uninit(&attr); + break; + } + case UPB_TYPE_MESSAGE: { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, offset, f)); + upb_handlers_setstartsubmsg(h, f, submsg_handler, &attr); + upb_handlerattr_uninit(&attr); + break; + } + } +} + +// Adds handlers to a map field. +static void add_handlers_for_mapfield(upb_handlers* h, + const upb_fielddef* fielddef, + size_t offset, + Descriptor* desc) { + const upb_msgdef* map_msgdef = upb_fielddef_msgsubdef(fielddef); + map_handlerdata_t* hd = new_map_handlerdata(offset, map_msgdef, desc); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + + upb_handlers_addcleanup(h, hd, free); + upb_handlerattr_sethandlerdata(&attr, hd); + upb_handlers_setstartsubmsg(h, fielddef, startmapentry_handler, &attr); + upb_handlerattr_uninit(&attr); +} + +// Adds handlers to a map-entry msgdef. +static void add_handlers_for_mapentry(const upb_msgdef* msgdef, + upb_handlers* h, + Descriptor* desc) { + const upb_fielddef* key_field = map_entry_key(msgdef); + const upb_fielddef* value_field = map_entry_value(msgdef); + map_handlerdata_t* hd = new_map_handlerdata(0, msgdef, desc); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + + upb_handlers_addcleanup(h, hd, free); + upb_handlerattr_sethandlerdata(&attr, hd); + upb_handlers_setendmsg(h, endmap_handler, &attr); + + add_handlers_for_singular_field( + h, key_field, + offsetof(map_parse_frame_t, key_storage)); + add_handlers_for_singular_field( + h, value_field, + offsetof(map_parse_frame_t, value_storage)); +} + +// Set up handlers for a oneof field. +static void add_handlers_for_oneof_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset, + size_t oneof_case_offset) { + + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata( + &attr, newoneofhandlerdata(h, offset, oneof_case_offset, f)); + + switch (upb_fielddef_type(f)) { + +#define SET_HANDLER(utype, ltype) \ + case utype: \ + upb_handlers_set##ltype(h, f, oneof##ltype##_handler, &attr); \ + break; + + SET_HANDLER(UPB_TYPE_BOOL, bool); + SET_HANDLER(UPB_TYPE_INT32, int32); + SET_HANDLER(UPB_TYPE_UINT32, uint32); + SET_HANDLER(UPB_TYPE_ENUM, int32); + SET_HANDLER(UPB_TYPE_FLOAT, float); + SET_HANDLER(UPB_TYPE_INT64, int64); + SET_HANDLER(UPB_TYPE_UINT64, uint64); + SET_HANDLER(UPB_TYPE_DOUBLE, double); + +#undef SET_HANDLER + + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; + upb_handlers_setstartstr(h, f, is_bytes ? + oneofbytes_handler : oneofstr_handler, + &attr); + upb_handlers_setstring(h, f, stringdata_handler, NULL); + break; + } + case UPB_TYPE_MESSAGE: { + upb_handlers_setstartsubmsg(h, f, oneofsubmsg_handler, &attr); + break; + } + } + + upb_handlerattr_uninit(&attr); +} + + +static void add_handlers_for_message(const void *closure, upb_handlers *h) { + const upb_msgdef* msgdef = upb_handlers_msgdef(h); + Descriptor* desc = ruby_to_Descriptor(get_def_obj((void*)msgdef)); + upb_msg_field_iter i; + + // If this is a mapentry message type, set up a special set of handlers and + // bail out of the normal (user-defined) message type handling. + if (upb_msgdef_mapentry(msgdef)) { + add_handlers_for_mapentry(msgdef, h, desc); + return; + } + + // Ensure layout exists. We may be invoked to create handlers for a given + // message if we are included as a submsg of another message type before our + // class is actually built, so to work around this, we just create the layout + // (and handlers, in the class-building function) on-demand. + if (desc->layout == NULL) { + desc->layout = create_layout(desc->msgdef); + } + + for (upb_msg_field_begin(&i, desc->msgdef); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + size_t offset = desc->layout->fields[upb_fielddef_index(f)].offset + + sizeof(MessageHeader); + + if (upb_fielddef_containingoneof(f)) { + size_t oneof_case_offset = + desc->layout->fields[upb_fielddef_index(f)].case_offset + + sizeof(MessageHeader); + add_handlers_for_oneof_field(h, f, offset, oneof_case_offset); + } else if (is_map_field(f)) { + add_handlers_for_mapfield(h, f, offset, desc); + } else if (upb_fielddef_isseq(f)) { + add_handlers_for_repeated_field(h, f, offset); + } else { + add_handlers_for_singular_field(h, f, offset); + } + } +} + +// Creates upb handlers for populating a message. +static const upb_handlers *new_fill_handlers(Descriptor* desc, + const void* owner) { + // TODO(cfallin, haberman): once upb gets a caching/memoization layer for + // handlers, reuse subdef handlers so that e.g. if we already parse + // B-with-field-of-type-C, we don't have to rebuild the whole hierarchy to + // parse A-with-field-of-type-B-with-field-of-type-C. + return upb_handlers_newfrozen(desc->msgdef, owner, + add_handlers_for_message, NULL); +} + +// Constructs the handlers for filling a message's data into an in-memory +// object. +const upb_handlers* get_fill_handlers(Descriptor* desc) { + if (!desc->fill_handlers) { + desc->fill_handlers = + new_fill_handlers(desc, &desc->fill_handlers); + } + return desc->fill_handlers; +} + +// Constructs the upb decoder method for parsing messages of this type. +// This is called from the message class creation code. +const upb_pbdecodermethod *new_fillmsg_decodermethod(Descriptor* desc, + const void* owner) { + const upb_handlers* handlers = get_fill_handlers(desc); + upb_pbdecodermethodopts opts; + upb_pbdecodermethodopts_init(&opts, handlers); + + return upb_pbdecodermethod_new(&opts, owner); +} + +static const upb_pbdecodermethod *msgdef_decodermethod(Descriptor* desc) { + if (desc->fill_method == NULL) { + desc->fill_method = new_fillmsg_decodermethod( + desc, &desc->fill_method); + } + return desc->fill_method; +} + +static const upb_json_parsermethod *msgdef_jsonparsermethod(Descriptor* desc) { + if (desc->json_fill_method == NULL) { + desc->json_fill_method = + upb_json_parsermethod_new(desc->msgdef, &desc->json_fill_method); + } + return desc->json_fill_method; +} + + +// Stack-allocated context during an encode/decode operation. Contains the upb +// environment and its stack-based allocator, an initial buffer for allocations +// to avoid malloc() when possible, and a template for Ruby exception messages +// if any error occurs. +#define STACK_ENV_STACKBYTES 4096 +typedef struct { + upb_env env; + const char* ruby_error_template; + char allocbuf[STACK_ENV_STACKBYTES]; +} stackenv; + +static void stackenv_init(stackenv* se, const char* errmsg); +static void stackenv_uninit(stackenv* se); + +// Callback invoked by upb if any error occurs during parsing or serialization. +static bool env_error_func(void* ud, const upb_status* status) { + stackenv* se = ud; + // Free the env -- rb_raise will longjmp up the stack past the encode/decode + // function so it would not otherwise have been freed. + stackenv_uninit(se); + + // TODO(haberman): have a way to verify that this is actually a parse error, + // instead of just throwing "parse error" unconditionally. + rb_raise(cParseError, se->ruby_error_template, upb_status_errmsg(status)); + // Never reached: rb_raise() always longjmp()s up the stack, past all of our + // code, back to Ruby. + return false; +} + +static void stackenv_init(stackenv* se, const char* errmsg) { + se->ruby_error_template = errmsg; + upb_env_init2(&se->env, se->allocbuf, sizeof(se->allocbuf), NULL); + upb_env_seterrorfunc(&se->env, env_error_func, se); +} + +static void stackenv_uninit(stackenv* se) { + upb_env_uninit(&se->env); +} + +/* + * call-seq: + * MessageClass.decode(data) => message + * + * Decodes the given data (as a string containing bytes in protocol buffers wire + * format) under the interpretration given by this message class's definition + * and returns a message object with the corresponding field values. + */ +VALUE Message_decode(VALUE klass, VALUE data) { + VALUE descriptor = rb_ivar_get(klass, descriptor_instancevar_interned); + Descriptor* desc = ruby_to_Descriptor(descriptor); + VALUE msgklass = Descriptor_msgclass(descriptor); + VALUE msg_rb; + MessageHeader* msg; + + if (TYPE(data) != T_STRING) { + rb_raise(rb_eArgError, "Expected string for binary protobuf data."); + } + + msg_rb = rb_class_new_instance(0, NULL, msgklass); + TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); + + { + const upb_pbdecodermethod* method = msgdef_decodermethod(desc); + const upb_handlers* h = upb_pbdecodermethod_desthandlers(method); + stackenv se; + upb_sink sink; + upb_pbdecoder* decoder; + stackenv_init(&se, "Error occurred during parsing: %s"); + + upb_sink_reset(&sink, h, msg); + decoder = upb_pbdecoder_create(&se.env, method, &sink); + upb_bufsrc_putbuf(RSTRING_PTR(data), RSTRING_LEN(data), + upb_pbdecoder_input(decoder)); + + stackenv_uninit(&se); + } + + return msg_rb; +} + +/* + * call-seq: + * MessageClass.decode_json(data) => message + * + * Decodes the given data (as a string containing bytes in protocol buffers wire + * format) under the interpretration given by this message class's definition + * and returns a message object with the corresponding field values. + */ +VALUE Message_decode_json(VALUE klass, VALUE data) { + VALUE descriptor = rb_ivar_get(klass, descriptor_instancevar_interned); + Descriptor* desc = ruby_to_Descriptor(descriptor); + VALUE msgklass = Descriptor_msgclass(descriptor); + VALUE msg_rb; + MessageHeader* msg; + + if (TYPE(data) != T_STRING) { + rb_raise(rb_eArgError, "Expected string for JSON data."); + } + // TODO(cfallin): Check and respect string encoding. If not UTF-8, we need to + // convert, because string handlers pass data directly to message string + // fields. + + msg_rb = rb_class_new_instance(0, NULL, msgklass); + TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); + + { + const upb_json_parsermethod* method = msgdef_jsonparsermethod(desc); + stackenv se; + upb_sink sink; + upb_json_parser* parser; + stackenv_init(&se, "Error occurred during parsing: %s"); + + upb_sink_reset(&sink, get_fill_handlers(desc), msg); + parser = upb_json_parser_create(&se.env, method, &sink); + upb_bufsrc_putbuf(RSTRING_PTR(data), RSTRING_LEN(data), + upb_json_parser_input(parser)); + + stackenv_uninit(&se); + } + + return msg_rb; +} + +// ----------------------------------------------------------------------------- +// Serializing. +// ----------------------------------------------------------------------------- +// +// The code below also comes from upb's prototype Ruby binding, developed by +// haberman@. + +/* stringsink *****************************************************************/ + +// This should probably be factored into a common upb component. + +typedef struct { + upb_byteshandler handler; + upb_bytessink sink; + char *ptr; + size_t len, size; +} stringsink; + +static void *stringsink_start(void *_sink, const void *hd, size_t size_hint) { + stringsink *sink = _sink; + sink->len = 0; + return sink; +} + +static size_t stringsink_string(void *_sink, const void *hd, const char *ptr, + size_t len, const upb_bufhandle *handle) { + stringsink *sink = _sink; + size_t new_size = sink->size; + + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + while (sink->len + len > new_size) { + new_size *= 2; + } + + if (new_size != sink->size) { + sink->ptr = realloc(sink->ptr, new_size); + sink->size = new_size; + } + + memcpy(sink->ptr + sink->len, ptr, len); + sink->len += len; + + return len; +} + +void stringsink_init(stringsink *sink) { + upb_byteshandler_init(&sink->handler); + upb_byteshandler_setstartstr(&sink->handler, stringsink_start, NULL); + upb_byteshandler_setstring(&sink->handler, stringsink_string, NULL); + + upb_bytessink_reset(&sink->sink, &sink->handler, sink); + + sink->size = 32; + sink->ptr = malloc(sink->size); + sink->len = 0; +} + +void stringsink_uninit(stringsink *sink) { + free(sink->ptr); +} + +/* msgvisitor *****************************************************************/ + +// TODO: If/when we support proto2 semantics in addition to the current proto3 +// semantics, which means that we have true field presence, we will want to +// modify msgvisitor so that it emits all present fields rather than all +// non-default-value fields. +// +// Likewise, when implementing JSON serialization, we may need to have a +// 'verbose' mode that outputs all fields and a 'concise' mode that outputs only +// those with non-default values. + +static void putmsg(VALUE msg, const Descriptor* desc, + upb_sink *sink, int depth); + +static upb_selector_t getsel(const upb_fielddef *f, upb_handlertype_t type) { + upb_selector_t ret; + bool ok = upb_handlers_getselector(f, type, &ret); + UPB_ASSERT_VAR(ok, ok); + return ret; +} + +static void putstr(VALUE str, const upb_fielddef *f, upb_sink *sink) { + upb_sink subsink; + + if (str == Qnil) return; + + assert(BUILTIN_TYPE(str) == RUBY_T_STRING); + + // Ensure that the string has the correct encoding. We also check at field-set + // time, but the user may have mutated the string object since then. + native_slot_validate_string_encoding(upb_fielddef_type(f), str); + + upb_sink_startstr(sink, getsel(f, UPB_HANDLER_STARTSTR), RSTRING_LEN(str), + &subsink); + upb_sink_putstring(&subsink, getsel(f, UPB_HANDLER_STRING), RSTRING_PTR(str), + RSTRING_LEN(str), NULL); + upb_sink_endstr(sink, getsel(f, UPB_HANDLER_ENDSTR)); +} + +static void putsubmsg(VALUE submsg, const upb_fielddef *f, upb_sink *sink, + int depth) { + upb_sink subsink; + VALUE descriptor; + Descriptor* subdesc; + + if (submsg == Qnil) return; + + descriptor = rb_ivar_get(submsg, descriptor_instancevar_interned); + subdesc = ruby_to_Descriptor(descriptor); + + upb_sink_startsubmsg(sink, getsel(f, UPB_HANDLER_STARTSUBMSG), &subsink); + putmsg(submsg, subdesc, &subsink, depth + 1); + upb_sink_endsubmsg(sink, getsel(f, UPB_HANDLER_ENDSUBMSG)); +} + +static void putary(VALUE ary, const upb_fielddef *f, upb_sink *sink, + int depth) { + upb_sink subsink; + upb_fieldtype_t type = upb_fielddef_type(f); + upb_selector_t sel = 0; + int size; + + if (ary == Qnil) return; + + upb_sink_startseq(sink, getsel(f, UPB_HANDLER_STARTSEQ), &subsink); + + if (upb_fielddef_isprimitive(f)) { + sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + } + + size = NUM2INT(RepeatedField_length(ary)); + for (int i = 0; i < size; i++) { + void* memory = RepeatedField_index_native(ary, i); + switch (type) { +#define T(upbtypeconst, upbtype, ctype) \ + case upbtypeconst: \ + upb_sink_put##upbtype(&subsink, sel, *((ctype *)memory)); \ + break; + + T(UPB_TYPE_FLOAT, float, float) + T(UPB_TYPE_DOUBLE, double, double) + T(UPB_TYPE_BOOL, bool, int8_t) + case UPB_TYPE_ENUM: + T(UPB_TYPE_INT32, int32, int32_t) + T(UPB_TYPE_UINT32, uint32, uint32_t) + T(UPB_TYPE_INT64, int64, int64_t) + T(UPB_TYPE_UINT64, uint64, uint64_t) + + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + putstr(*((VALUE *)memory), f, &subsink); + break; + case UPB_TYPE_MESSAGE: + putsubmsg(*((VALUE *)memory), f, &subsink, depth); + break; + +#undef T + + } + } + upb_sink_endseq(sink, getsel(f, UPB_HANDLER_ENDSEQ)); +} + +static void put_ruby_value(VALUE value, + const upb_fielddef *f, + VALUE type_class, + int depth, + upb_sink *sink) { + upb_selector_t sel = 0; + if (upb_fielddef_isprimitive(f)) { + sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + } + + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + upb_sink_putint32(sink, sel, NUM2INT(value)); + break; + case UPB_TYPE_INT64: + upb_sink_putint64(sink, sel, NUM2LL(value)); + break; + case UPB_TYPE_UINT32: + upb_sink_putuint32(sink, sel, NUM2UINT(value)); + break; + case UPB_TYPE_UINT64: + upb_sink_putuint64(sink, sel, NUM2ULL(value)); + break; + case UPB_TYPE_FLOAT: + upb_sink_putfloat(sink, sel, NUM2DBL(value)); + break; + case UPB_TYPE_DOUBLE: + upb_sink_putdouble(sink, sel, NUM2DBL(value)); + break; + case UPB_TYPE_ENUM: { + if (TYPE(value) == T_SYMBOL) { + value = rb_funcall(type_class, rb_intern("resolve"), 1, value); + } + upb_sink_putint32(sink, sel, NUM2INT(value)); + break; + } + case UPB_TYPE_BOOL: + upb_sink_putbool(sink, sel, value == Qtrue); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + putstr(value, f, sink); + break; + case UPB_TYPE_MESSAGE: + putsubmsg(value, f, sink, depth); + } +} + +static void putmap(VALUE map, const upb_fielddef *f, upb_sink *sink, + int depth) { + Map* self; + upb_sink subsink; + const upb_fielddef* key_field; + const upb_fielddef* value_field; + Map_iter it; + + if (map == Qnil) return; + self = ruby_to_Map(map); + + upb_sink_startseq(sink, getsel(f, UPB_HANDLER_STARTSEQ), &subsink); + + assert(upb_fielddef_type(f) == UPB_TYPE_MESSAGE); + key_field = map_field_key(f); + value_field = map_field_value(f); + + for (Map_begin(map, &it); !Map_done(&it); Map_next(&it)) { + VALUE key = Map_iter_key(&it); + VALUE value = Map_iter_value(&it); + upb_status status; + + upb_sink entry_sink; + upb_sink_startsubmsg(&subsink, getsel(f, UPB_HANDLER_STARTSUBMSG), + &entry_sink); + upb_sink_startmsg(&entry_sink); + + put_ruby_value(key, key_field, Qnil, depth + 1, &entry_sink); + put_ruby_value(value, value_field, self->value_type_class, depth + 1, + &entry_sink); + + upb_sink_endmsg(&entry_sink, &status); + upb_sink_endsubmsg(&subsink, getsel(f, UPB_HANDLER_ENDSUBMSG)); + } + + upb_sink_endseq(sink, getsel(f, UPB_HANDLER_ENDSEQ)); +} + +static void putmsg(VALUE msg_rb, const Descriptor* desc, + upb_sink *sink, int depth) { + MessageHeader* msg; + upb_msg_field_iter i; + upb_status status; + + upb_sink_startmsg(sink); + + // Protect against cycles (possible because users may freely reassign message + // and repeated fields) by imposing a maximum recursion depth. + if (depth > ENCODE_MAX_NESTING) { + rb_raise(rb_eRuntimeError, + "Maximum recursion depth exceeded during encoding."); + } + + TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); + + for (upb_msg_field_begin(&i, desc->msgdef); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + bool is_matching_oneof = false; + uint32_t offset = + desc->layout->fields[upb_fielddef_index(f)].offset + + sizeof(MessageHeader); + + if (upb_fielddef_containingoneof(f)) { + uint32_t oneof_case_offset = + desc->layout->fields[upb_fielddef_index(f)].case_offset + + sizeof(MessageHeader); + // For a oneof, check that this field is actually present -- skip all the + // below if not. + if (DEREF(msg, oneof_case_offset, uint32_t) != + upb_fielddef_number(f)) { + continue; + } + // Otherwise, fall through to the appropriate singular-field handler + // below. + is_matching_oneof = true; + } + + if (is_map_field(f)) { + VALUE map = DEREF(msg, offset, VALUE); + if (map != Qnil) { + putmap(map, f, sink, depth); + } + } else if (upb_fielddef_isseq(f)) { + VALUE ary = DEREF(msg, offset, VALUE); + if (ary != Qnil) { + putary(ary, f, sink, depth); + } + } else if (upb_fielddef_isstring(f)) { + VALUE str = DEREF(msg, offset, VALUE); + if (is_matching_oneof || RSTRING_LEN(str) > 0) { + putstr(str, f, sink); + } + } else if (upb_fielddef_issubmsg(f)) { + putsubmsg(DEREF(msg, offset, VALUE), f, sink, depth); + } else { + upb_selector_t sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + +#define T(upbtypeconst, upbtype, ctype, default_value) \ + case upbtypeconst: { \ + ctype value = DEREF(msg, offset, ctype); \ + if (is_matching_oneof || value != default_value) { \ + upb_sink_put##upbtype(sink, sel, value); \ + } \ + } \ + break; + + switch (upb_fielddef_type(f)) { + T(UPB_TYPE_FLOAT, float, float, 0.0) + T(UPB_TYPE_DOUBLE, double, double, 0.0) + T(UPB_TYPE_BOOL, bool, uint8_t, 0) + case UPB_TYPE_ENUM: + T(UPB_TYPE_INT32, int32, int32_t, 0) + T(UPB_TYPE_UINT32, uint32, uint32_t, 0) + T(UPB_TYPE_INT64, int64, int64_t, 0) + T(UPB_TYPE_UINT64, uint64, uint64_t, 0) + + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + case UPB_TYPE_MESSAGE: rb_raise(rb_eRuntimeError, "Internal error."); + } + +#undef T + + } + } + + upb_sink_endmsg(sink, &status); +} + +static const upb_handlers* msgdef_pb_serialize_handlers(Descriptor* desc) { + if (desc->pb_serialize_handlers == NULL) { + desc->pb_serialize_handlers = + upb_pb_encoder_newhandlers(desc->msgdef, &desc->pb_serialize_handlers); + } + return desc->pb_serialize_handlers; +} + +static const upb_handlers* msgdef_json_serialize_handlers( + Descriptor* desc, bool preserve_proto_fieldnames) { + if (preserve_proto_fieldnames) { + if (desc->json_serialize_handlers == NULL) { + desc->json_serialize_handlers = + upb_json_printer_newhandlers( + desc->msgdef, true, &desc->json_serialize_handlers); + } + return desc->json_serialize_handlers; + } else { + if (desc->json_serialize_handlers_preserve == NULL) { + desc->json_serialize_handlers_preserve = + upb_json_printer_newhandlers( + desc->msgdef, false, &desc->json_serialize_handlers_preserve); + } + return desc->json_serialize_handlers_preserve; + } +} + +/* + * call-seq: + * MessageClass.encode(msg) => bytes + * + * Encodes the given message object to its serialized form in protocol buffers + * wire format. + */ +VALUE Message_encode(VALUE klass, VALUE msg_rb) { + VALUE descriptor = rb_ivar_get(klass, descriptor_instancevar_interned); + Descriptor* desc = ruby_to_Descriptor(descriptor); + + stringsink sink; + stringsink_init(&sink); + + { + const upb_handlers* serialize_handlers = + msgdef_pb_serialize_handlers(desc); + + stackenv se; + upb_pb_encoder* encoder; + VALUE ret; + + stackenv_init(&se, "Error occurred during encoding: %s"); + encoder = upb_pb_encoder_create(&se.env, serialize_handlers, &sink.sink); + + putmsg(msg_rb, desc, upb_pb_encoder_input(encoder), 0); + + ret = rb_str_new(sink.ptr, sink.len); + + stackenv_uninit(&se); + stringsink_uninit(&sink); + + return ret; + } +} + +/* + * call-seq: + * MessageClass.encode_json(msg) => json_string + * + * Encodes the given message object into its serialized JSON representation. + */ +VALUE Message_encode_json(int argc, VALUE* argv, VALUE klass) { + VALUE descriptor = rb_ivar_get(klass, descriptor_instancevar_interned); + Descriptor* desc = ruby_to_Descriptor(descriptor); + VALUE msg_rb; + VALUE preserve_proto_fieldnames = Qfalse; + stringsink sink; + + if (argc < 1 || argc > 2) { + rb_raise(rb_eArgError, "Expected 1 or 2 arguments."); + } + + msg_rb = argv[0]; + + if (argc == 2) { + VALUE hash_args = argv[1]; + if (TYPE(hash_args) != T_HASH) { + rb_raise(rb_eArgError, "Expected hash arguments."); + } + preserve_proto_fieldnames = rb_hash_lookup2( + hash_args, ID2SYM(rb_intern("preserve_proto_fieldnames")), Qfalse); + } + + stringsink_init(&sink); + + { + const upb_handlers* serialize_handlers = + msgdef_json_serialize_handlers(desc, RTEST(preserve_proto_fieldnames)); + upb_json_printer* printer; + stackenv se; + VALUE ret; + + stackenv_init(&se, "Error occurred during encoding: %s"); + printer = upb_json_printer_create(&se.env, serialize_handlers, &sink.sink); + + putmsg(msg_rb, desc, upb_json_printer_input(printer), 0); + + ret = rb_enc_str_new(sink.ptr, sink.len, rb_utf8_encoding()); + + stackenv_uninit(&se); + stringsink_uninit(&sink); + + return ret; + } +} + diff --git a/ruby/ext/google/protobuf_c/extconf.rb b/ruby/ext/google/protobuf_c/extconf.rb new file mode 100644 index 00000000..b368dcc6 --- /dev/null +++ b/ruby/ext/google/protobuf_c/extconf.rb @@ -0,0 +1,10 @@ +#!/usr/bin/ruby + +require 'mkmf' + +$CFLAGS += " -std=c99 -O3 -DNDEBUG" + +$objs = ["protobuf.o", "defs.o", "storage.o", "message.o", + "repeated_field.o", "map.o", "encode_decode.o", "upb.o"] + +create_makefile("google/protobuf_c") diff --git a/ruby/ext/google/protobuf_c/map.c b/ruby/ext/google/protobuf_c/map.c new file mode 100644 index 00000000..92fc7286 --- /dev/null +++ b/ruby/ext/google/protobuf_c/map.c @@ -0,0 +1,808 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Basic map operations on top of upb's strtable. +// +// Note that we roll our own `Map` container here because, as for +// `RepeatedField`, we want a strongly-typed container. This is so that any user +// errors due to incorrect map key or value types are raised as close as +// possible to the error site, rather than at some deferred point (e.g., +// serialization). +// +// We build our `Map` on top of upb_strtable so that we're able to take +// advantage of the native_slot storage abstraction, as RepeatedField does. +// (This is not quite a perfect mapping -- see the key conversions below -- but +// gives us full support and error-checking for all value types for free.) +// ----------------------------------------------------------------------------- + +// Map values are stored using the native_slot abstraction (as with repeated +// field values), but keys are a bit special. Since we use a strtable, we need +// to store keys as sequences of bytes such that equality of those bytes maps +// one-to-one to equality of keys. We store strings directly (i.e., they map to +// their own bytes) and integers as native integers (using the native_slot +// abstraction). + +// Note that there is another tradeoff here in keeping string keys as native +// strings rather than Ruby strings: traversing the Map requires conversion to +// Ruby string values on every traversal, potentially creating more garbage. We +// should consider ways to cache a Ruby version of the key if this becomes an +// issue later. + +// Forms a key to use with the underlying strtable from a Ruby key value. |buf| +// must point to TABLE_KEY_BUF_LENGTH bytes of temporary space, used to +// construct a key byte sequence if needed. |out_key| and |out_length| provide +// the resulting key data/length. +#define TABLE_KEY_BUF_LENGTH 8 // sizeof(uint64_t) +static void table_key(Map* self, VALUE key, + char* buf, + const char** out_key, + size_t* out_length) { + switch (self->key_type) { + case UPB_TYPE_BYTES: + case UPB_TYPE_STRING: + // Strings: use string content directly. + Check_Type(key, T_STRING); + native_slot_validate_string_encoding(self->key_type, key); + *out_key = RSTRING_PTR(key); + *out_length = RSTRING_LEN(key); + break; + + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + native_slot_set(self->key_type, Qnil, buf, key); + *out_key = buf; + *out_length = native_slot_size(self->key_type); + break; + + default: + // Map constructor should not allow a Map with another key type to be + // constructed. + assert(false); + break; + } +} + +static VALUE table_key_to_ruby(Map* self, const char* buf, size_t length) { + switch (self->key_type) { + case UPB_TYPE_BYTES: + case UPB_TYPE_STRING: { + VALUE ret = rb_str_new(buf, length); + rb_enc_associate(ret, + (self->key_type == UPB_TYPE_BYTES) ? + kRubyString8bitEncoding : kRubyStringUtf8Encoding); + return ret; + } + + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + return native_slot_get(self->key_type, Qnil, buf); + + default: + assert(false); + return Qnil; + } +} + +static void* value_memory(upb_value* v) { + return (void*)(&v->val); +} + +// ----------------------------------------------------------------------------- +// Map container type. +// ----------------------------------------------------------------------------- + +const rb_data_type_t Map_type = { + "Google::Protobuf::Map", + { Map_mark, Map_free, NULL }, +}; + +VALUE cMap; + +Map* ruby_to_Map(VALUE _self) { + Map* self; + TypedData_Get_Struct(_self, Map, &Map_type, self); + return self; +} + +void Map_mark(void* _self) { + Map* self = _self; + + rb_gc_mark(self->value_type_class); + + if (self->value_type == UPB_TYPE_STRING || + self->value_type == UPB_TYPE_BYTES || + self->value_type == UPB_TYPE_MESSAGE) { + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + native_slot_mark(self->value_type, mem); + } + } +} + +void Map_free(void* _self) { + Map* self = _self; + upb_strtable_uninit(&self->table); + xfree(self); +} + +VALUE Map_alloc(VALUE klass) { + Map* self = ALLOC(Map); + memset(self, 0, sizeof(Map)); + self->value_type_class = Qnil; + return TypedData_Wrap_Struct(klass, &Map_type, self); +} + +static bool needs_typeclass(upb_fieldtype_t type) { + switch (type) { + case UPB_TYPE_MESSAGE: + case UPB_TYPE_ENUM: + return true; + default: + return false; + } +} + +/* + * call-seq: + * Map.new(key_type, value_type, value_typeclass = nil, init_hashmap = {}) + * => new map + * + * Allocates a new Map container. This constructor may be called with 2, 3, or 4 + * arguments. The first two arguments are always present and are symbols (taking + * on the same values as field-type symbols in message descriptors) that + * indicate the type of the map key and value fields. + * + * The supported key types are: :int32, :int64, :uint32, :uint64, :bool, + * :string, :bytes. + * + * The supported value types are: :int32, :int64, :uint32, :uint64, :bool, + * :string, :bytes, :enum, :message. + * + * The third argument, value_typeclass, must be present if value_type is :enum + * or :message. As in RepeatedField#new, this argument must be a message class + * (for :message) or enum module (for :enum). + * + * The last argument, if present, provides initial content for map. Note that + * this may be an ordinary Ruby hashmap or another Map instance with identical + * key and value types. Also note that this argument may be present whether or + * not value_typeclass is present (and it is unambiguously separate from + * value_typeclass because value_typeclass's presence is strictly determined by + * value_type). The contents of this initial hashmap or Map instance are + * shallow-copied into the new Map: the original map is unmodified, but + * references to underlying objects will be shared if the value type is a + * message type. + */ +VALUE Map_init(int argc, VALUE* argv, VALUE _self) { + Map* self = ruby_to_Map(_self); + int init_value_arg; + + // We take either two args (:key_type, :value_type), three args (:key_type, + // :value_type, "ValueMessageType"), or four args (the above plus an initial + // hashmap). + if (argc < 2 || argc > 4) { + rb_raise(rb_eArgError, "Map constructor expects 2, 3 or 4 arguments."); + } + + self->key_type = ruby_to_fieldtype(argv[0]); + self->value_type = ruby_to_fieldtype(argv[1]); + + // Check that the key type is an allowed type. + switch (self->key_type) { + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + case UPB_TYPE_BOOL: + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + // These are OK. + break; + default: + rb_raise(rb_eArgError, "Invalid key type for map."); + } + + init_value_arg = 2; + if (needs_typeclass(self->value_type) && argc > 2) { + self->value_type_class = argv[2]; + validate_type_class(self->value_type, self->value_type_class); + init_value_arg = 3; + } + + // Table value type is always UINT64: this ensures enough space to store the + // native_slot value. + if (!upb_strtable_init(&self->table, UPB_CTYPE_UINT64)) { + rb_raise(rb_eRuntimeError, "Could not allocate table."); + } + + if (argc > init_value_arg) { + Map_merge_into_self(_self, argv[init_value_arg]); + } + + return Qnil; +} + +/* + * call-seq: + * Map.each(&block) + * + * Invokes &block on each |key, value| pair in the map, in unspecified order. + * Note that Map also includes Enumerable; map thus acts like a normal Ruby + * sequence. + */ +VALUE Map_each(VALUE _self) { + Map* self = ruby_to_Map(_self); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + rb_yield_values(2, key, value); + } + + return Qnil; +} + +/* + * call-seq: + * Map.keys => [list_of_keys] + * + * Returns the list of keys contained in the map, in unspecified order. + */ +VALUE Map_keys(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE ret = rb_ary_new(); + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + rb_ary_push(ret, key); + } + + return ret; +} + +/* + * call-seq: + * Map.values => [list_of_values] + * + * Returns the list of values contained in the map, in unspecified order. + */ +VALUE Map_values(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE ret = rb_ary_new(); + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + rb_ary_push(ret, value); + } + + return ret; +} + +/* + * call-seq: + * Map.[](key) => value + * + * Accesses the element at the given key. Throws an exception if the key type is + * incorrect. Returns nil when the key is not present in the map. + */ +VALUE Map_index(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + upb_value v; + table_key(self, key, keybuf, &keyval, &length); + + if (upb_strtable_lookup2(&self->table, keyval, length, &v)) { + void* mem = value_memory(&v); + return native_slot_get(self->value_type, self->value_type_class, mem); + } else { + return Qnil; + } +} + +/* + * call-seq: + * Map.[]=(key, value) => value + * + * Inserts or overwrites the value at the given key with the given new value. + * Throws an exception if the key type is incorrect. Returns the new value that + * was just inserted. + */ +VALUE Map_index_set(VALUE _self, VALUE key, VALUE value) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + upb_value v; + void* mem; + table_key(self, key, keybuf, &keyval, &length); + + mem = value_memory(&v); + native_slot_set(self->value_type, self->value_type_class, mem, value); + + // Replace any existing value by issuing a 'remove' operation first. + upb_strtable_remove2(&self->table, keyval, length, NULL); + if (!upb_strtable_insert2(&self->table, keyval, length, v)) { + rb_raise(rb_eRuntimeError, "Could not insert into table"); + } + + // Ruby hashmap's :[]= method also returns the inserted value. + return value; +} + +/* + * call-seq: + * Map.has_key?(key) => bool + * + * Returns true if the given key is present in the map. Throws an exception if + * the key has the wrong type. + */ +VALUE Map_has_key(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + table_key(self, key, keybuf, &keyval, &length); + + if (upb_strtable_lookup2(&self->table, keyval, length, NULL)) { + return Qtrue; + } else { + return Qfalse; + } +} + +/* + * call-seq: + * Map.delete(key) => old_value + * + * Deletes the value at the given key, if any, returning either the old value or + * nil if none was present. Throws an exception if the key is of the wrong type. + */ +VALUE Map_delete(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + upb_value v; + table_key(self, key, keybuf, &keyval, &length); + + if (upb_strtable_remove2(&self->table, keyval, length, &v)) { + void* mem = value_memory(&v); + return native_slot_get(self->value_type, self->value_type_class, mem); + } else { + return Qnil; + } +} + +/* + * call-seq: + * Map.clear + * + * Removes all entries from the map. + */ +VALUE Map_clear(VALUE _self) { + Map* self = ruby_to_Map(_self); + + // Uninit and reinit the table -- this is faster than iterating and doing a + // delete-lookup on each key. + upb_strtable_uninit(&self->table); + if (!upb_strtable_init(&self->table, UPB_CTYPE_INT64)) { + rb_raise(rb_eRuntimeError, "Unable to re-initialize table"); + } + return Qnil; +} + +/* + * call-seq: + * Map.length + * + * Returns the number of entries (key-value pairs) in the map. + */ +VALUE Map_length(VALUE _self) { + Map* self = ruby_to_Map(_self); + return ULL2NUM(upb_strtable_count(&self->table)); +} + +static VALUE Map_new_this_type(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Qnil; + VALUE key_type = fieldtype_to_ruby(self->key_type); + VALUE value_type = fieldtype_to_ruby(self->value_type); + if (self->value_type_class != Qnil) { + new_map = rb_funcall(CLASS_OF(_self), rb_intern("new"), 3, + key_type, value_type, self->value_type_class); + } else { + new_map = rb_funcall(CLASS_OF(_self), rb_intern("new"), 2, + key_type, value_type); + } + return new_map; +} + +/* + * call-seq: + * Map.dup => new_map + * + * Duplicates this map with a shallow copy. References to all non-primitive + * element objects (e.g., submessages) are shared. + */ +VALUE Map_dup(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Map_new_this_type(_self); + Map* new_self = ruby_to_Map(new_map); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value dup; + void* dup_mem = value_memory(&dup); + native_slot_dup(self->value_type, dup_mem, mem); + + if (!upb_strtable_insert2(&new_self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + dup)) { + rb_raise(rb_eRuntimeError, "Error inserting value into new table"); + } + } + + return new_map; +} + +// Used by Google::Protobuf.deep_copy but not exposed directly. +VALUE Map_deep_copy(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Map_new_this_type(_self); + Map* new_self = ruby_to_Map(new_map); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value dup; + void* dup_mem = value_memory(&dup); + native_slot_deep_copy(self->value_type, dup_mem, mem); + + if (!upb_strtable_insert2(&new_self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + dup)) { + rb_raise(rb_eRuntimeError, "Error inserting value into new table"); + } + } + + return new_map; +} + +/* + * call-seq: + * Map.==(other) => boolean + * + * Compares this map to another. Maps are equal if they have identical key sets, + * and for each key, the values in both maps compare equal. Elements are + * compared as per normal Ruby semantics, by calling their :== methods (or + * performing a more efficient comparison for primitive types). + * + * Maps with dissimilar key types or value types/typeclasses are never equal, + * even if value comparison (for example, between integers and floats) would + * have otherwise indicated that every element has equal value. + */ +VALUE Map_eq(VALUE _self, VALUE _other) { + Map* self = ruby_to_Map(_self); + Map* other; + upb_strtable_iter it; + + // Allow comparisons to Ruby hashmaps by converting to a temporary Map + // instance. Slow, but workable. + if (TYPE(_other) == T_HASH) { + VALUE other_map = Map_new_this_type(_self); + Map_merge_into_self(other_map, _other); + _other = other_map; + } + + other = ruby_to_Map(_other); + + if (self == other) { + return Qtrue; + } + if (self->key_type != other->key_type || + self->value_type != other->value_type || + self->value_type_class != other->value_type_class) { + return Qfalse; + } + if (upb_strtable_count(&self->table) != upb_strtable_count(&other->table)) { + return Qfalse; + } + + // For each member of self, check that an equal member exists at the same key + // in other. + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value other_v; + void* other_mem = value_memory(&other_v); + + if (!upb_strtable_lookup2(&other->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + &other_v)) { + // Not present in other map. + return Qfalse; + } + + if (!native_slot_eq(self->value_type, mem, other_mem)) { + // Present, but value not equal. + return Qfalse; + } + } + + return Qtrue; +} + +/* + * call-seq: + * Map.hash => hash_value + * + * Returns a hash value based on this map's contents. + */ +VALUE Map_hash(VALUE _self) { + Map* self = ruby_to_Map(_self); + + st_index_t h = rb_hash_start(0); + VALUE hash_sym = rb_intern("hash"); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + h = rb_hash_uint(h, NUM2LONG(rb_funcall(key, hash_sym, 0))); + h = rb_hash_uint(h, NUM2LONG(rb_funcall(value, hash_sym, 0))); + } + + return INT2FIX(h); +} + +/* + * call-seq: + * Map.inspect => string + * + * Returns a string representing this map's elements. It will be formatted as + * "{key => value, key => value, ...}", with each key and value string + * representation computed by its own #inspect method. + */ +VALUE Map_inspect(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE str = rb_str_new2("{"); + + bool first = true; + VALUE inspect_sym = rb_intern("inspect"); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + if (!first) { + str = rb_str_cat2(str, ", "); + } else { + first = false; + } + str = rb_str_append(str, rb_funcall(key, inspect_sym, 0)); + str = rb_str_cat2(str, "=>"); + str = rb_str_append(str, rb_funcall(value, inspect_sym, 0)); + } + + str = rb_str_cat2(str, "}"); + return str; +} + +/* + * call-seq: + * Map.merge(other_map) => map + * + * Copies key/value pairs from other_map into a copy of this map. If a key is + * set in other_map and this map, the value from other_map overwrites the value + * in the new copy of this map. Returns the new copy of this map with merged + * contents. + */ +VALUE Map_merge(VALUE _self, VALUE hashmap) { + VALUE dupped = Map_dup(_self); + return Map_merge_into_self(dupped, hashmap); +} + +static int merge_into_self_callback(VALUE key, VALUE value, VALUE self) { + Map_index_set(self, key, value); + return ST_CONTINUE; +} + +// Used only internally -- shared by #merge and #initialize. +VALUE Map_merge_into_self(VALUE _self, VALUE hashmap) { + if (TYPE(hashmap) == T_HASH) { + rb_hash_foreach(hashmap, merge_into_self_callback, _self); + } else if (RB_TYPE_P(hashmap, T_DATA) && RTYPEDDATA_P(hashmap) && + RTYPEDDATA_TYPE(hashmap) == &Map_type) { + + Map* self = ruby_to_Map(_self); + Map* other = ruby_to_Map(hashmap); + upb_strtable_iter it; + + if (self->key_type != other->key_type || + self->value_type != other->value_type || + self->value_type_class != other->value_type_class) { + rb_raise(rb_eArgError, "Attempt to merge Map with mismatching types"); + } + + for (upb_strtable_begin(&it, &other->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + // Replace any existing value by issuing a 'remove' operation first. + upb_value v; + upb_value oldv; + upb_strtable_remove2(&self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + &oldv); + + v = upb_strtable_iter_value(&it); + upb_strtable_insert2(&self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + v); + } + } else { + rb_raise(rb_eArgError, "Unknown type merging into Map"); + } + return _self; +} + +// Internal method: map iterator initialization (used for serialization). +void Map_begin(VALUE _self, Map_iter* iter) { + Map* self = ruby_to_Map(_self); + iter->self = self; + upb_strtable_begin(&iter->it, &self->table); +} + +void Map_next(Map_iter* iter) { + upb_strtable_next(&iter->it); +} + +bool Map_done(Map_iter* iter) { + return upb_strtable_done(&iter->it); +} + +VALUE Map_iter_key(Map_iter* iter) { + return table_key_to_ruby( + iter->self, + upb_strtable_iter_key(&iter->it), + upb_strtable_iter_keylength(&iter->it)); +} + +VALUE Map_iter_value(Map_iter* iter) { + upb_value v = upb_strtable_iter_value(&iter->it); + void* mem = value_memory(&v); + return native_slot_get(iter->self->value_type, + iter->self->value_type_class, + mem); +} + +void Map_register(VALUE module) { + VALUE klass = rb_define_class_under(module, "Map", rb_cObject); + rb_define_alloc_func(klass, Map_alloc); + cMap = klass; + rb_gc_register_address(&cMap); + + rb_define_method(klass, "initialize", Map_init, -1); + rb_define_method(klass, "each", Map_each, 0); + rb_define_method(klass, "keys", Map_keys, 0); + rb_define_method(klass, "values", Map_values, 0); + rb_define_method(klass, "[]", Map_index, 1); + rb_define_method(klass, "[]=", Map_index_set, 2); + rb_define_method(klass, "has_key?", Map_has_key, 1); + rb_define_method(klass, "delete", Map_delete, 1); + rb_define_method(klass, "clear", Map_clear, 0); + rb_define_method(klass, "length", Map_length, 0); + rb_define_method(klass, "dup", Map_dup, 0); + rb_define_method(klass, "==", Map_eq, 1); + rb_define_method(klass, "hash", Map_hash, 0); + rb_define_method(klass, "inspect", Map_inspect, 0); + rb_define_method(klass, "merge", Map_merge, 1); + rb_include_module(klass, rb_mEnumerable); +} diff --git a/ruby/ext/google/protobuf_c/message.c b/ruby/ext/google/protobuf_c/message.c new file mode 100644 index 00000000..e16250f3 --- /dev/null +++ b/ruby/ext/google/protobuf_c/message.c @@ -0,0 +1,578 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Class/module creation from msgdefs and enumdefs, respectively. +// ----------------------------------------------------------------------------- + +void* Message_data(void* msg) { + return ((uint8_t *)msg) + sizeof(MessageHeader); +} + +void Message_mark(void* _self) { + MessageHeader* self = (MessageHeader *)_self; + layout_mark(self->descriptor->layout, Message_data(self)); +} + +void Message_free(void* self) { + xfree(self); +} + +rb_data_type_t Message_type = { + "Message", + { Message_mark, Message_free, NULL }, +}; + +VALUE Message_alloc(VALUE klass) { + VALUE descriptor = rb_ivar_get(klass, descriptor_instancevar_interned); + Descriptor* desc = ruby_to_Descriptor(descriptor); + MessageHeader* msg = (MessageHeader*)ALLOC_N( + uint8_t, sizeof(MessageHeader) + desc->layout->size); + VALUE ret; + + memset(Message_data(msg), 0, desc->layout->size); + + // We wrap first so that everything in the message object is GC-rooted in case + // a collection happens during object creation in layout_init(). + ret = TypedData_Wrap_Struct(klass, &Message_type, msg); + msg->descriptor = desc; + rb_ivar_set(ret, descriptor_instancevar_interned, descriptor); + + layout_init(desc->layout, Message_data(msg)); + + return ret; +} + +static VALUE which_oneof_field(MessageHeader* self, const upb_oneofdef* o) { + upb_oneof_iter it; + size_t case_ofs; + uint32_t oneof_case; + const upb_fielddef* first_field; + const upb_fielddef* f; + + // If no fields in the oneof, always nil. + if (upb_oneofdef_numfields(o) == 0) { + return Qnil; + } + // Grab the first field in the oneof so we can get its layout info to find the + // oneof_case field. + upb_oneof_begin(&it, o); + assert(!upb_oneof_done(&it)); + first_field = upb_oneof_iter_field(&it); + assert(upb_fielddef_containingoneof(first_field) != NULL); + + case_ofs = + self->descriptor->layout-> + fields[upb_fielddef_index(first_field)].case_offset; + oneof_case = *((uint32_t*)((char*)Message_data(self) + case_ofs)); + + if (oneof_case == ONEOF_CASE_NONE) { + return Qnil; + } + + // oneof_case is a field index, so find that field. + f = upb_oneofdef_itof(o, oneof_case); + assert(f != NULL); + + return ID2SYM(rb_intern(upb_fielddef_name(f))); +} + +/* + * call-seq: + * Message.method_missing(*args) + * + * Provides accessors and setters for message fields according to their field + * names. For any field whose name does not conflict with a built-in method, an + * accessor is provided with the same name as the field, and a setter is + * provided with the name of the field plus the '=' suffix. Thus, given a + * message instance 'msg' with field 'foo', the following code is valid: + * + * msg.foo = 42 + * puts msg.foo + * + * This method also provides read-only accessors for oneofs. If a oneof exists + * with name 'my_oneof', then msg.my_oneof will return a Ruby symbol equal to + * the name of the field in that oneof that is currently set, or nil if none. + */ +VALUE Message_method_missing(int argc, VALUE* argv, VALUE _self) { + MessageHeader* self; + VALUE method_name, method_str; + char* name; + size_t name_len; + bool setter; + const upb_oneofdef* o; + const upb_fielddef* f; + + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + if (argc < 1) { + rb_raise(rb_eArgError, "Expected method name as first argument."); + } + method_name = argv[0]; + if (!SYMBOL_P(method_name)) { + rb_raise(rb_eArgError, "Expected symbol as method name."); + } + method_str = rb_id2str(SYM2ID(method_name)); + name = RSTRING_PTR(method_str); + name_len = RSTRING_LEN(method_str); + setter = false; + + // Setters have names that end in '='. + if (name[name_len - 1] == '=') { + setter = true; + name_len--; + } + + // See if this name corresponds to either a oneof or field in this message. + if (!upb_msgdef_lookupname(self->descriptor->msgdef, name, name_len, &f, + &o)) { + return rb_call_super(argc, argv); + } + + if (o != NULL) { + // This is a oneof -- return which field inside the oneof is set. + if (setter) { + rb_raise(rb_eRuntimeError, "Oneof accessors are read-only."); + } + return which_oneof_field(self, o); + } else { + // This is a field -- get or set the field's value. + assert(f); + if (setter) { + if (argc < 2) { + rb_raise(rb_eArgError, "No value provided to setter."); + } + layout_set(self->descriptor->layout, Message_data(self), f, argv[1]); + return Qnil; + } else { + return layout_get(self->descriptor->layout, Message_data(self), f); + } + } +} + +int Message_initialize_kwarg(VALUE key, VALUE val, VALUE _self) { + MessageHeader* self; + VALUE method_str; + char* name; + const upb_fielddef* f; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + if (!SYMBOL_P(key)) { + rb_raise(rb_eArgError, + "Expected symbols as hash keys in initialization map."); + } + + method_str = rb_id2str(SYM2ID(key)); + name = RSTRING_PTR(method_str); + f = upb_msgdef_ntofz(self->descriptor->msgdef, name); + if (f == NULL) { + rb_raise(rb_eArgError, + "Unknown field name '%s' in initialization map entry.", name); + } + + if (is_map_field(f)) { + VALUE map; + + if (TYPE(val) != T_HASH) { + rb_raise(rb_eArgError, + "Expected Hash object as initializer value for map field '%s'.", name); + } + map = layout_get(self->descriptor->layout, Message_data(self), f); + Map_merge_into_self(map, val); + } else if (upb_fielddef_label(f) == UPB_LABEL_REPEATED) { + VALUE ary; + + if (TYPE(val) != T_ARRAY) { + rb_raise(rb_eArgError, + "Expected array as initializer value for repeated field '%s'.", name); + } + ary = layout_get(self->descriptor->layout, Message_data(self), f); + for (int i = 0; i < RARRAY_LEN(val); i++) { + RepeatedField_push(ary, rb_ary_entry(val, i)); + } + } else { + layout_set(self->descriptor->layout, Message_data(self), f, val); + } + return 0; +} + +/* + * call-seq: + * Message.new(kwargs) => new_message + * + * Creates a new instance of the given message class. Keyword arguments may be + * provided with keywords corresponding to field names. + * + * Note that no literal Message class exists. Only concrete classes per message + * type exist, as provided by the #msgclass method on Descriptors after they + * have been added to a pool. The method definitions described here on the + * Message class are provided on each concrete message class. + */ +VALUE Message_initialize(int argc, VALUE* argv, VALUE _self) { + VALUE hash_args; + + if (argc == 0) { + return Qnil; + } + if (argc != 1) { + rb_raise(rb_eArgError, "Expected 0 or 1 arguments."); + } + hash_args = argv[0]; + if (TYPE(hash_args) != T_HASH) { + rb_raise(rb_eArgError, "Expected hash arguments."); + } + + rb_hash_foreach(hash_args, Message_initialize_kwarg, _self); + return Qnil; +} + +/* + * call-seq: + * Message.dup => new_message + * + * Performs a shallow copy of this message and returns the new copy. + */ +VALUE Message_dup(VALUE _self) { + MessageHeader* self; + VALUE new_msg; + MessageHeader* new_msg_self; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + new_msg = rb_class_new_instance(0, NULL, CLASS_OF(_self)); + TypedData_Get_Struct(new_msg, MessageHeader, &Message_type, new_msg_self); + + layout_dup(self->descriptor->layout, + Message_data(new_msg_self), + Message_data(self)); + + return new_msg; +} + +// Internal only; used by Google::Protobuf.deep_copy. +VALUE Message_deep_copy(VALUE _self) { + MessageHeader* self; + MessageHeader* new_msg_self; + VALUE new_msg; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + new_msg = rb_class_new_instance(0, NULL, CLASS_OF(_self)); + TypedData_Get_Struct(new_msg, MessageHeader, &Message_type, new_msg_self); + + layout_deep_copy(self->descriptor->layout, + Message_data(new_msg_self), + Message_data(self)); + + return new_msg; +} + +/* + * call-seq: + * Message.==(other) => boolean + * + * Performs a deep comparison of this message with another. Messages are equal + * if they have the same type and if each field is equal according to the :== + * method's semantics (a more efficient comparison may actually be done if the + * field is of a primitive type). + */ +VALUE Message_eq(VALUE _self, VALUE _other) { + MessageHeader* self; + MessageHeader* other; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + TypedData_Get_Struct(_other, MessageHeader, &Message_type, other); + + if (self->descriptor != other->descriptor) { + return Qfalse; + } + + return layout_eq(self->descriptor->layout, + Message_data(self), + Message_data(other)); +} + +/* + * call-seq: + * Message.hash => hash_value + * + * Returns a hash value that represents this message's field values. + */ +VALUE Message_hash(VALUE _self) { + MessageHeader* self; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + return layout_hash(self->descriptor->layout, Message_data(self)); +} + +/* + * call-seq: + * Message.inspect => string + * + * Returns a human-readable string representing this message. It will be + * formatted as "<MessageType: field1: value1, field2: value2, ...>". Each + * field's value is represented according to its own #inspect method. + */ +VALUE Message_inspect(VALUE _self) { + MessageHeader* self; + VALUE str; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + str = rb_str_new2("<"); + str = rb_str_append(str, rb_str_new2(rb_class2name(CLASS_OF(_self)))); + str = rb_str_cat2(str, ": "); + str = rb_str_append(str, layout_inspect( + self->descriptor->layout, Message_data(self))); + str = rb_str_cat2(str, ">"); + return str; +} + + +VALUE Message_to_h(VALUE _self) { + MessageHeader* self; + VALUE hash; + upb_msg_field_iter it; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + + hash = rb_hash_new(); + + for (upb_msg_field_begin(&it, self->descriptor->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + VALUE msg_value = layout_get(self->descriptor->layout, Message_data(self), + field); + VALUE msg_key = ID2SYM(rb_intern(upb_fielddef_name(field))); + if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + msg_value = RepeatedField_to_ary(msg_value); + } + rb_hash_aset(hash, msg_key, msg_value); + } + return hash; +} + + + +/* + * call-seq: + * Message.[](index) => value + * + * Accesses a field's value by field name. The provided field name should be a + * string. + */ +VALUE Message_index(VALUE _self, VALUE field_name) { + MessageHeader* self; + const upb_fielddef* field; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + Check_Type(field_name, T_STRING); + field = upb_msgdef_ntofz(self->descriptor->msgdef, RSTRING_PTR(field_name)); + if (field == NULL) { + return Qnil; + } + return layout_get(self->descriptor->layout, Message_data(self), field); +} + +/* + * call-seq: + * Message.[]=(index, value) + * + * Sets a field's value by field name. The provided field name should be a + * string. + */ +VALUE Message_index_set(VALUE _self, VALUE field_name, VALUE value) { + MessageHeader* self; + const upb_fielddef* field; + TypedData_Get_Struct(_self, MessageHeader, &Message_type, self); + Check_Type(field_name, T_STRING); + field = upb_msgdef_ntofz(self->descriptor->msgdef, RSTRING_PTR(field_name)); + if (field == NULL) { + rb_raise(rb_eArgError, "Unknown field: %s", RSTRING_PTR(field_name)); + } + layout_set(self->descriptor->layout, Message_data(self), field, value); + return Qnil; +} + +/* + * call-seq: + * Message.descriptor => descriptor + * + * Class method that returns the Descriptor instance corresponding to this + * message class's type. + */ +VALUE Message_descriptor(VALUE klass) { + return rb_ivar_get(klass, descriptor_instancevar_interned); +} + +VALUE build_class_from_descriptor(Descriptor* desc) { + const char *name; + VALUE klass; + + if (desc->layout == NULL) { + desc->layout = create_layout(desc->msgdef); + } + if (desc->fill_method == NULL) { + desc->fill_method = new_fillmsg_decodermethod(desc, &desc->fill_method); + } + + name = upb_msgdef_fullname(desc->msgdef); + if (name == NULL) { + rb_raise(rb_eRuntimeError, "Descriptor does not have assigned name."); + } + + klass = rb_define_class_id( + // Docs say this parameter is ignored. User will assign return value to + // their own toplevel constant class name. + rb_intern("Message"), + rb_cObject); + rb_ivar_set(klass, descriptor_instancevar_interned, + get_def_obj(desc->msgdef)); + rb_define_alloc_func(klass, Message_alloc); + rb_require("google/protobuf/message_exts"); + rb_include_module(klass, rb_eval_string("Google::Protobuf::MessageExts")); + rb_extend_object( + klass, rb_eval_string("Google::Protobuf::MessageExts::ClassMethods")); + + rb_define_method(klass, "method_missing", + Message_method_missing, -1); + rb_define_method(klass, "initialize", Message_initialize, -1); + rb_define_method(klass, "dup", Message_dup, 0); + // Also define #clone so that we don't inherit Object#clone. + rb_define_method(klass, "clone", Message_dup, 0); + rb_define_method(klass, "==", Message_eq, 1); + rb_define_method(klass, "hash", Message_hash, 0); + rb_define_method(klass, "to_h", Message_to_h, 0); + rb_define_method(klass, "to_hash", Message_to_h, 0); + rb_define_method(klass, "inspect", Message_inspect, 0); + rb_define_method(klass, "[]", Message_index, 1); + rb_define_method(klass, "[]=", Message_index_set, 2); + rb_define_singleton_method(klass, "decode", Message_decode, 1); + rb_define_singleton_method(klass, "encode", Message_encode, 1); + rb_define_singleton_method(klass, "decode_json", Message_decode_json, 1); + rb_define_singleton_method(klass, "encode_json", Message_encode_json, -1); + rb_define_singleton_method(klass, "descriptor", Message_descriptor, 0); + + return klass; +} + +/* + * call-seq: + * Enum.lookup(number) => name + * + * This module method, provided on each generated enum module, looks up an enum + * value by number and returns its name as a Ruby symbol, or nil if not found. + */ +VALUE enum_lookup(VALUE self, VALUE number) { + int32_t num = NUM2INT(number); + VALUE desc = rb_ivar_get(self, descriptor_instancevar_interned); + EnumDescriptor* enumdesc = ruby_to_EnumDescriptor(desc); + + const char* name = upb_enumdef_iton(enumdesc->enumdef, num); + if (name == NULL) { + return Qnil; + } else { + return ID2SYM(rb_intern(name)); + } +} + +/* + * call-seq: + * Enum.resolve(name) => number + * + * This module method, provided on each generated enum module, looks up an enum + * value by name (as a Ruby symbol) and returns its name, or nil if not found. + */ +VALUE enum_resolve(VALUE self, VALUE sym) { + const char* name = rb_id2name(SYM2ID(sym)); + VALUE desc = rb_ivar_get(self, descriptor_instancevar_interned); + EnumDescriptor* enumdesc = ruby_to_EnumDescriptor(desc); + + int32_t num = 0; + bool found = upb_enumdef_ntoiz(enumdesc->enumdef, name, &num); + if (!found) { + return Qnil; + } else { + return INT2NUM(num); + } +} + +/* + * call-seq: + * Enum.descriptor + * + * This module method, provided on each generated enum module, returns the + * EnumDescriptor corresponding to this enum type. + */ +VALUE enum_descriptor(VALUE self) { + return rb_ivar_get(self, descriptor_instancevar_interned); +} + +VALUE build_module_from_enumdesc(EnumDescriptor* enumdesc) { + VALUE mod = rb_define_module_id( + rb_intern(upb_enumdef_fullname(enumdesc->enumdef))); + + upb_enum_iter it; + for (upb_enum_begin(&it, enumdesc->enumdef); + !upb_enum_done(&it); + upb_enum_next(&it)) { + const char* name = upb_enum_iter_name(&it); + int32_t value = upb_enum_iter_number(&it); + if (name[0] < 'A' || name[0] > 'Z') { + rb_raise(rb_eTypeError, + "Enum value '%s' does not start with an uppercase letter " + "as is required for Ruby constants.", + name); + } + rb_define_const(mod, name, INT2NUM(value)); + } + + rb_define_singleton_method(mod, "lookup", enum_lookup, 1); + rb_define_singleton_method(mod, "resolve", enum_resolve, 1); + rb_define_singleton_method(mod, "descriptor", enum_descriptor, 0); + rb_ivar_set(mod, descriptor_instancevar_interned, + get_def_obj(enumdesc->enumdef)); + + return mod; +} + +/* + * call-seq: + * Google::Protobuf.deep_copy(obj) => copy_of_obj + * + * Performs a deep copy of a RepeatedField instance, a Map instance, or a + * message object, recursively copying its members. + */ +VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj) { + VALUE klass = CLASS_OF(obj); + if (klass == cRepeatedField) { + return RepeatedField_deep_copy(obj); + } else if (klass == cMap) { + return Map_deep_copy(obj); + } else { + return Message_deep_copy(obj); + } +} diff --git a/ruby/ext/google/protobuf_c/protobuf.c b/ruby/ext/google/protobuf_c/protobuf.c new file mode 100644 index 00000000..7cde4aec --- /dev/null +++ b/ruby/ext/google/protobuf_c/protobuf.c @@ -0,0 +1,115 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Global map from upb {msg,enum}defs to wrapper Descriptor/EnumDescriptor +// instances. +// ----------------------------------------------------------------------------- + +// This is a hash table from def objects (encoded by converting pointers to +// Ruby integers) to MessageDef/EnumDef instances (as Ruby values). +VALUE upb_def_to_ruby_obj_map; + +VALUE cError; +VALUE cParseError; + +void add_def_obj(const void* def, VALUE value) { + rb_hash_aset(upb_def_to_ruby_obj_map, ULL2NUM((intptr_t)def), value); +} + +VALUE get_def_obj(const void* def) { + return rb_hash_aref(upb_def_to_ruby_obj_map, ULL2NUM((intptr_t)def)); +} + +// ----------------------------------------------------------------------------- +// Utilities. +// ----------------------------------------------------------------------------- + +// Raises a Ruby error if |status| is not OK, using its error message. +void check_upb_status(const upb_status* status, const char* msg) { + if (!upb_ok(status)) { + rb_raise(rb_eRuntimeError, "%s: %s\n", msg, upb_status_errmsg(status)); + } +} + +// String encodings: we look these up once, at load time, and then cache them +// here. +rb_encoding* kRubyStringUtf8Encoding; +rb_encoding* kRubyStringASCIIEncoding; +rb_encoding* kRubyString8bitEncoding; + +// Ruby-interned string: "descriptor". We use this identifier to store an +// instance variable on message classes we create in order to link them back to +// their descriptors. +// +// We intern this once at module load time then use the interned identifier at +// runtime in order to avoid the cost of repeatedly interning in hot paths. +const char* kDescriptorInstanceVar = "descriptor"; +ID descriptor_instancevar_interned; + +// ----------------------------------------------------------------------------- +// Initialization/entry point. +// ----------------------------------------------------------------------------- + +// This must be named "Init_protobuf_c" because the Ruby module is named +// "protobuf_c" -- the VM looks for this symbol in our .so. +void Init_protobuf_c() { + VALUE google = rb_define_module("Google"); + VALUE protobuf = rb_define_module_under(google, "Protobuf"); + VALUE internal = rb_define_module_under(protobuf, "Internal"); + + descriptor_instancevar_interned = rb_intern(kDescriptorInstanceVar); + DescriptorPool_register(protobuf); + Descriptor_register(protobuf); + FieldDescriptor_register(protobuf); + OneofDescriptor_register(protobuf); + EnumDescriptor_register(protobuf); + MessageBuilderContext_register(internal); + OneofBuilderContext_register(internal); + EnumBuilderContext_register(internal); + Builder_register(internal); + RepeatedField_register(protobuf); + Map_register(protobuf); + + cError = rb_const_get(protobuf, rb_intern("Error")); + cParseError = rb_const_get(protobuf, rb_intern("ParseError")); + + rb_define_singleton_method(protobuf, "deep_copy", + Google_Protobuf_deep_copy, 1); + + kRubyStringUtf8Encoding = rb_utf8_encoding(); + kRubyStringASCIIEncoding = rb_usascii_encoding(); + kRubyString8bitEncoding = rb_ascii8bit_encoding(); + + upb_def_to_ruby_obj_map = rb_hash_new(); + rb_gc_register_address(&upb_def_to_ruby_obj_map); +} diff --git a/ruby/ext/google/protobuf_c/protobuf.h b/ruby/ext/google/protobuf_c/protobuf.h new file mode 100644 index 00000000..2834c894 --- /dev/null +++ b/ruby/ext/google/protobuf_c/protobuf.h @@ -0,0 +1,538 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__ +#define __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__ + +#include <ruby/ruby.h> +#include <ruby/vm.h> +#include <ruby/encoding.h> + +#include "upb.h" + +// Forward decls. +struct DescriptorPool; +struct Descriptor; +struct FieldDescriptor; +struct EnumDescriptor; +struct MessageLayout; +struct MessageField; +struct MessageHeader; +struct MessageBuilderContext; +struct EnumBuilderContext; +struct Builder; + +typedef struct DescriptorPool DescriptorPool; +typedef struct Descriptor Descriptor; +typedef struct FieldDescriptor FieldDescriptor; +typedef struct OneofDescriptor OneofDescriptor; +typedef struct EnumDescriptor EnumDescriptor; +typedef struct MessageLayout MessageLayout; +typedef struct MessageField MessageField; +typedef struct MessageHeader MessageHeader; +typedef struct MessageBuilderContext MessageBuilderContext; +typedef struct OneofBuilderContext OneofBuilderContext; +typedef struct EnumBuilderContext EnumBuilderContext; +typedef struct Builder Builder; + +/* + It can be a bit confusing how the C structs defined below and the Ruby + objects interact and hold references to each other. First, a few principles: + + - Ruby's "TypedData" abstraction lets a Ruby VALUE hold a pointer to a C + struct (or arbitrary memory chunk), own it, and free it when collected. + Thus, each struct below will have a corresponding Ruby object + wrapping/owning it. + + - To get back from an underlying upb {msg,enum}def to the Ruby object, we + keep a global hashmap, accessed by get_def_obj/add_def_obj below. + + The in-memory structure is then something like: + + Ruby | upb + | + DescriptorPool ------------|-----------> upb_symtab____________________ + | | (message types) \ + | v \ + Descriptor ---------------|-----------> upb_msgdef (enum types)| + |--> msgclass | | ^ | + | (dynamically built) | | | (submsg fields) | + |--> MessageLayout | | | / + |--------------------------|> decoder method| | / + \--------------------------|> serialize | | / + | handlers v | / + FieldDescriptor -----------|-----------> upb_fielddef / + | | / + | v (enum fields) / + EnumDescriptor ------------|-----------> upb_enumdef <----------' + | + | + ^ | \___/ + `---------------|-----------------' (get_def_obj map) + */ + +// ----------------------------------------------------------------------------- +// Ruby class structure definitions. +// ----------------------------------------------------------------------------- + +struct DescriptorPool { + upb_symtab* symtab; +}; + +struct Descriptor { + const upb_msgdef* msgdef; + MessageLayout* layout; + VALUE klass; // begins as nil + const upb_handlers* fill_handlers; + const upb_pbdecodermethod* fill_method; + const upb_json_parsermethod* json_fill_method; + const upb_handlers* pb_serialize_handlers; + const upb_handlers* json_serialize_handlers; + const upb_handlers* json_serialize_handlers_preserve; + // Handlers hold type class references for sub-message fields directly in some + // cases. We need to keep these rooted because they might otherwise be + // collected. + VALUE typeclass_references; +}; + +struct FieldDescriptor { + const upb_fielddef* fielddef; +}; + +struct OneofDescriptor { + const upb_oneofdef* oneofdef; +}; + +struct EnumDescriptor { + const upb_enumdef* enumdef; + VALUE module; // begins as nil +}; + +struct MessageBuilderContext { + VALUE descriptor; + VALUE builder; +}; + +struct OneofBuilderContext { + VALUE descriptor; + VALUE builder; +}; + +struct EnumBuilderContext { + VALUE enumdesc; +}; + +struct Builder { + VALUE pending_list; + upb_def** defs; // used only while finalizing +}; + +extern VALUE cDescriptorPool; +extern VALUE cDescriptor; +extern VALUE cFieldDescriptor; +extern VALUE cEnumDescriptor; +extern VALUE cMessageBuilderContext; +extern VALUE cOneofBuilderContext; +extern VALUE cEnumBuilderContext; +extern VALUE cBuilder; + +extern VALUE cError; +extern VALUE cParseError; + +// We forward-declare all of the Ruby method implementations here because we +// sometimes call the methods directly across .c files, rather than going +// through Ruby's method dispatching (e.g. during message parse). It's cleaner +// to keep the list of object methods together than to split them between +// static-in-file definitions and header declarations. + +void DescriptorPool_mark(void* _self); +void DescriptorPool_free(void* _self); +VALUE DescriptorPool_alloc(VALUE klass); +void DescriptorPool_register(VALUE module); +DescriptorPool* ruby_to_DescriptorPool(VALUE value); +VALUE DescriptorPool_add(VALUE _self, VALUE def); +VALUE DescriptorPool_build(VALUE _self); +VALUE DescriptorPool_lookup(VALUE _self, VALUE name); +VALUE DescriptorPool_generated_pool(VALUE _self); + +void Descriptor_mark(void* _self); +void Descriptor_free(void* _self); +VALUE Descriptor_alloc(VALUE klass); +void Descriptor_register(VALUE module); +Descriptor* ruby_to_Descriptor(VALUE value); +VALUE Descriptor_name(VALUE _self); +VALUE Descriptor_name_set(VALUE _self, VALUE str); +VALUE Descriptor_each(VALUE _self); +VALUE Descriptor_lookup(VALUE _self, VALUE name); +VALUE Descriptor_add_field(VALUE _self, VALUE obj); +VALUE Descriptor_add_oneof(VALUE _self, VALUE obj); +VALUE Descriptor_each_oneof(VALUE _self); +VALUE Descriptor_lookup_oneof(VALUE _self, VALUE name); +VALUE Descriptor_msgclass(VALUE _self); +extern const rb_data_type_t _Descriptor_type; + +void FieldDescriptor_mark(void* _self); +void FieldDescriptor_free(void* _self); +VALUE FieldDescriptor_alloc(VALUE klass); +void FieldDescriptor_register(VALUE module); +FieldDescriptor* ruby_to_FieldDescriptor(VALUE value); +VALUE FieldDescriptor_name(VALUE _self); +VALUE FieldDescriptor_name_set(VALUE _self, VALUE str); +VALUE FieldDescriptor_type(VALUE _self); +VALUE FieldDescriptor_type_set(VALUE _self, VALUE type); +VALUE FieldDescriptor_label(VALUE _self); +VALUE FieldDescriptor_label_set(VALUE _self, VALUE label); +VALUE FieldDescriptor_number(VALUE _self); +VALUE FieldDescriptor_number_set(VALUE _self, VALUE number); +VALUE FieldDescriptor_submsg_name(VALUE _self); +VALUE FieldDescriptor_submsg_name_set(VALUE _self, VALUE value); +VALUE FieldDescriptor_subtype(VALUE _self); +VALUE FieldDescriptor_get(VALUE _self, VALUE msg_rb); +VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value); +upb_fieldtype_t ruby_to_fieldtype(VALUE type); +VALUE fieldtype_to_ruby(upb_fieldtype_t type); + +void OneofDescriptor_mark(void* _self); +void OneofDescriptor_free(void* _self); +VALUE OneofDescriptor_alloc(VALUE klass); +void OneofDescriptor_register(VALUE module); +OneofDescriptor* ruby_to_OneofDescriptor(VALUE value); +VALUE OneofDescriptor_name(VALUE _self); +VALUE OneofDescriptor_name_set(VALUE _self, VALUE value); +VALUE OneofDescriptor_add_field(VALUE _self, VALUE field); +VALUE OneofDescriptor_each(VALUE _self, VALUE field); + +void EnumDescriptor_mark(void* _self); +void EnumDescriptor_free(void* _self); +VALUE EnumDescriptor_alloc(VALUE klass); +void EnumDescriptor_register(VALUE module); +EnumDescriptor* ruby_to_EnumDescriptor(VALUE value); +VALUE EnumDescriptor_name(VALUE _self); +VALUE EnumDescriptor_name_set(VALUE _self, VALUE str); +VALUE EnumDescriptor_add_value(VALUE _self, VALUE name, VALUE number); +VALUE EnumDescriptor_lookup_name(VALUE _self, VALUE name); +VALUE EnumDescriptor_lookup_value(VALUE _self, VALUE number); +VALUE EnumDescriptor_each(VALUE _self); +VALUE EnumDescriptor_enummodule(VALUE _self); +extern const rb_data_type_t _EnumDescriptor_type; + +void MessageBuilderContext_mark(void* _self); +void MessageBuilderContext_free(void* _self); +VALUE MessageBuilderContext_alloc(VALUE klass); +void MessageBuilderContext_register(VALUE module); +MessageBuilderContext* ruby_to_MessageBuilderContext(VALUE value); +VALUE MessageBuilderContext_initialize(VALUE _self, + VALUE descriptor, + VALUE builder); +VALUE MessageBuilderContext_optional(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_required(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_repeated(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_map(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_oneof(VALUE _self, VALUE name); + +void OneofBuilderContext_mark(void* _self); +void OneofBuilderContext_free(void* _self); +VALUE OneofBuilderContext_alloc(VALUE klass); +void OneofBuilderContext_register(VALUE module); +OneofBuilderContext* ruby_to_OneofBuilderContext(VALUE value); +VALUE OneofBuilderContext_initialize(VALUE _self, + VALUE descriptor, + VALUE builder); +VALUE OneofBuilderContext_optional(int argc, VALUE* argv, VALUE _self); + +void EnumBuilderContext_mark(void* _self); +void EnumBuilderContext_free(void* _self); +VALUE EnumBuilderContext_alloc(VALUE klass); +void EnumBuilderContext_register(VALUE module); +EnumBuilderContext* ruby_to_EnumBuilderContext(VALUE value); +VALUE EnumBuilderContext_initialize(VALUE _self, VALUE enumdesc); +VALUE EnumBuilderContext_value(VALUE _self, VALUE name, VALUE number); + +void Builder_mark(void* _self); +void Builder_free(void* _self); +VALUE Builder_alloc(VALUE klass); +void Builder_register(VALUE module); +Builder* ruby_to_Builder(VALUE value); +VALUE Builder_add_message(VALUE _self, VALUE name); +VALUE Builder_add_enum(VALUE _self, VALUE name); +VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb); + +// ----------------------------------------------------------------------------- +// Native slot storage abstraction. +// ----------------------------------------------------------------------------- + +#define NATIVE_SLOT_MAX_SIZE sizeof(uint64_t) + +size_t native_slot_size(upb_fieldtype_t type); +void native_slot_set(upb_fieldtype_t type, + VALUE type_class, + void* memory, + VALUE value); +// Atomically (with respect to Ruby VM calls) either update the value and set a +// oneof case, or do neither. If |case_memory| is null, then no case value is +// set. +void native_slot_set_value_and_case(upb_fieldtype_t type, + VALUE type_class, + void* memory, + VALUE value, + uint32_t* case_memory, + uint32_t case_number); +VALUE native_slot_get(upb_fieldtype_t type, + VALUE type_class, + const void* memory); +void native_slot_init(upb_fieldtype_t type, void* memory); +void native_slot_mark(upb_fieldtype_t type, void* memory); +void native_slot_dup(upb_fieldtype_t type, void* to, void* from); +void native_slot_deep_copy(upb_fieldtype_t type, void* to, void* from); +bool native_slot_eq(upb_fieldtype_t type, void* mem1, void* mem2); + +void native_slot_validate_string_encoding(upb_fieldtype_t type, VALUE value); +void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE value); + +extern rb_encoding* kRubyStringUtf8Encoding; +extern rb_encoding* kRubyStringASCIIEncoding; +extern rb_encoding* kRubyString8bitEncoding; + +VALUE field_type_class(const upb_fielddef* field); + +#define MAP_KEY_FIELD 1 +#define MAP_VALUE_FIELD 2 + +// Oneof case slot value to indicate that no oneof case is set. The value `0` is +// safe because field numbers are used as case identifiers, and no field can +// have a number of 0. +#define ONEOF_CASE_NONE 0 + +// These operate on a map field (i.e., a repeated field of submessages whose +// submessage type is a map-entry msgdef). +bool is_map_field(const upb_fielddef* field); +const upb_fielddef* map_field_key(const upb_fielddef* field); +const upb_fielddef* map_field_value(const upb_fielddef* field); + +// These operate on a map-entry msgdef. +const upb_fielddef* map_entry_key(const upb_msgdef* msgdef); +const upb_fielddef* map_entry_value(const upb_msgdef* msgdef); + +// ----------------------------------------------------------------------------- +// Repeated field container type. +// ----------------------------------------------------------------------------- + +typedef struct { + upb_fieldtype_t field_type; + VALUE field_type_class; + void* elements; + int size; + int capacity; +} RepeatedField; + +void RepeatedField_mark(void* self); +void RepeatedField_free(void* self); +VALUE RepeatedField_alloc(VALUE klass); +VALUE RepeatedField_init(int argc, VALUE* argv, VALUE self); +void RepeatedField_register(VALUE module); + +extern const rb_data_type_t RepeatedField_type; +extern VALUE cRepeatedField; + +RepeatedField* ruby_to_RepeatedField(VALUE value); + +VALUE RepeatedField_each(VALUE _self); +VALUE RepeatedField_index(int argc, VALUE* argv, VALUE _self); +void* RepeatedField_index_native(VALUE _self, int index); +VALUE RepeatedField_index_set(VALUE _self, VALUE _index, VALUE val); +void RepeatedField_reserve(RepeatedField* self, int new_size); +VALUE RepeatedField_push(VALUE _self, VALUE val); +void RepeatedField_push_native(VALUE _self, void* data); +VALUE RepeatedField_pop_one(VALUE _self); +VALUE RepeatedField_insert(int argc, VALUE* argv, VALUE _self); +VALUE RepeatedField_replace(VALUE _self, VALUE list); +VALUE RepeatedField_clear(VALUE _self); +VALUE RepeatedField_length(VALUE _self); +VALUE RepeatedField_dup(VALUE _self); +VALUE RepeatedField_deep_copy(VALUE _self); +VALUE RepeatedField_to_ary(VALUE _self); +VALUE RepeatedField_eq(VALUE _self, VALUE _other); +VALUE RepeatedField_hash(VALUE _self); +VALUE RepeatedField_inspect(VALUE _self); +VALUE RepeatedField_plus(VALUE _self, VALUE list); + +// Defined in repeated_field.c; also used by Map. +void validate_type_class(upb_fieldtype_t type, VALUE klass); + +// ----------------------------------------------------------------------------- +// Map container type. +// ----------------------------------------------------------------------------- + +typedef struct { + upb_fieldtype_t key_type; + upb_fieldtype_t value_type; + VALUE value_type_class; + upb_strtable table; +} Map; + +void Map_mark(void* self); +void Map_free(void* self); +VALUE Map_alloc(VALUE klass); +VALUE Map_init(int argc, VALUE* argv, VALUE self); +void Map_register(VALUE module); + +extern const rb_data_type_t Map_type; +extern VALUE cMap; + +Map* ruby_to_Map(VALUE value); + +VALUE Map_each(VALUE _self); +VALUE Map_keys(VALUE _self); +VALUE Map_values(VALUE _self); +VALUE Map_index(VALUE _self, VALUE key); +VALUE Map_index_set(VALUE _self, VALUE key, VALUE value); +VALUE Map_has_key(VALUE _self, VALUE key); +VALUE Map_delete(VALUE _self, VALUE key); +VALUE Map_clear(VALUE _self); +VALUE Map_length(VALUE _self); +VALUE Map_dup(VALUE _self); +VALUE Map_deep_copy(VALUE _self); +VALUE Map_eq(VALUE _self, VALUE _other); +VALUE Map_hash(VALUE _self); +VALUE Map_inspect(VALUE _self); +VALUE Map_merge(VALUE _self, VALUE hashmap); +VALUE Map_merge_into_self(VALUE _self, VALUE hashmap); + +typedef struct { + Map* self; + upb_strtable_iter it; +} Map_iter; + +void Map_begin(VALUE _self, Map_iter* iter); +void Map_next(Map_iter* iter); +bool Map_done(Map_iter* iter); +VALUE Map_iter_key(Map_iter* iter); +VALUE Map_iter_value(Map_iter* iter); + +// ----------------------------------------------------------------------------- +// Message layout / storage. +// ----------------------------------------------------------------------------- + +#define MESSAGE_FIELD_NO_CASE ((size_t)-1) + +struct MessageField { + size_t offset; + size_t case_offset; // for oneofs, a uint32. Else, MESSAGE_FIELD_NO_CASE. +}; + +struct MessageLayout { + const upb_msgdef* msgdef; + MessageField* fields; + size_t size; +}; + +MessageLayout* create_layout(const upb_msgdef* msgdef); +void free_layout(MessageLayout* layout); +VALUE layout_get(MessageLayout* layout, + const void* storage, + const upb_fielddef* field); +void layout_set(MessageLayout* layout, + void* storage, + const upb_fielddef* field, + VALUE val); +void layout_init(MessageLayout* layout, void* storage); +void layout_mark(MessageLayout* layout, void* storage); +void layout_dup(MessageLayout* layout, void* to, void* from); +void layout_deep_copy(MessageLayout* layout, void* to, void* from); +VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2); +VALUE layout_hash(MessageLayout* layout, void* storage); +VALUE layout_inspect(MessageLayout* layout, void* storage); + +// ----------------------------------------------------------------------------- +// Message class creation. +// ----------------------------------------------------------------------------- + +struct MessageHeader { + Descriptor* descriptor; // kept alive by self.class.descriptor reference. + // Data comes after this. +}; + +extern rb_data_type_t Message_type; + +VALUE build_class_from_descriptor(Descriptor* descriptor); +void* Message_data(void* msg); +void Message_mark(void* self); +void Message_free(void* self); +VALUE Message_alloc(VALUE klass); +VALUE Message_method_missing(int argc, VALUE* argv, VALUE _self); +VALUE Message_initialize(int argc, VALUE* argv, VALUE _self); +VALUE Message_dup(VALUE _self); +VALUE Message_deep_copy(VALUE _self); +VALUE Message_eq(VALUE _self, VALUE _other); +VALUE Message_hash(VALUE _self); +VALUE Message_inspect(VALUE _self); +VALUE Message_index(VALUE _self, VALUE field_name); +VALUE Message_index_set(VALUE _self, VALUE field_name, VALUE value); +VALUE Message_descriptor(VALUE klass); +VALUE Message_decode(VALUE klass, VALUE data); +VALUE Message_encode(VALUE klass, VALUE msg_rb); +VALUE Message_decode_json(VALUE klass, VALUE data); +VALUE Message_encode_json(int argc, VALUE* argv, VALUE klass); + +VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj); + +VALUE build_module_from_enumdesc(EnumDescriptor* enumdef); +VALUE enum_lookup(VALUE self, VALUE number); +VALUE enum_resolve(VALUE self, VALUE sym); + +const upb_pbdecodermethod *new_fillmsg_decodermethod( + Descriptor* descriptor, const void *owner); + +// Maximum depth allowed during encoding, to avoid stack overflows due to +// cycles. +#define ENCODE_MAX_NESTING 63 + +// ----------------------------------------------------------------------------- +// Global map from upb {msg,enum}defs to wrapper Descriptor/EnumDescriptor +// instances. +// ----------------------------------------------------------------------------- +void add_def_obj(const void* def, VALUE value); +VALUE get_def_obj(const void* def); + +// ----------------------------------------------------------------------------- +// Utilities. +// ----------------------------------------------------------------------------- + +void check_upb_status(const upb_status* status, const char* msg); + +#define CHECK_UPB(code, msg) do { \ + upb_status status = UPB_STATUS_INIT; \ + code; \ + check_upb_status(&status, msg); \ +} while (0) + +extern ID descriptor_instancevar_interned; + +#endif // __GOOGLE_PROTOBUF_RUBY_PROTOBUF_H__ diff --git a/ruby/ext/google/protobuf_c/repeated_field.c b/ruby/ext/google/protobuf_c/repeated_field.c new file mode 100644 index 00000000..83afbc91 --- /dev/null +++ b/ruby/ext/google/protobuf_c/repeated_field.c @@ -0,0 +1,651 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Repeated field container type. +// ----------------------------------------------------------------------------- + +const rb_data_type_t RepeatedField_type = { + "Google::Protobuf::RepeatedField", + { RepeatedField_mark, RepeatedField_free, NULL }, +}; + +VALUE cRepeatedField; + +RepeatedField* ruby_to_RepeatedField(VALUE _self) { + RepeatedField* self; + TypedData_Get_Struct(_self, RepeatedField, &RepeatedField_type, self); + return self; +} + +void* RepeatedField_memoryat(RepeatedField* self, int index, int element_size) { + return ((uint8_t *)self->elements) + index * element_size; +} + +static int index_position(VALUE _index, RepeatedField* repeated_field) { + int index = NUM2INT(_index); + if (index < 0 && repeated_field->size > 0) { + index = repeated_field->size + index; + } + return index; +} + +VALUE RepeatedField_subarray(VALUE _self, long beg, long len) { + RepeatedField* self = ruby_to_RepeatedField(_self); + int element_size = native_slot_size(self->field_type); + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + + size_t off = beg * element_size; + VALUE ary = rb_ary_new2(len); + for (int i = beg; i < beg + len; i++, off += element_size) { + void* mem = ((uint8_t *)self->elements) + off; + VALUE elem = native_slot_get(field_type, field_type_class, mem); + rb_ary_push(ary, elem); + } + return ary; +} + +/* + * call-seq: + * RepeatedField.each(&block) + * + * Invokes the block once for each element of the repeated field. RepeatedField + * also includes Enumerable; combined with this method, the repeated field thus + * acts like an ordinary Ruby sequence. + */ +VALUE RepeatedField_each(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + int element_size = native_slot_size(field_type); + + size_t off = 0; + for (int i = 0; i < self->size; i++, off += element_size) { + void* memory = (void *) (((uint8_t *)self->elements) + off); + VALUE val = native_slot_get(field_type, field_type_class, memory); + rb_yield(val); + } + return _self; +} + + +/* + * call-seq: + * RepeatedField.[](index) => value + * + * Accesses the element at the given index. Returns nil on out-of-bounds + */ +VALUE RepeatedField_index(int argc, VALUE* argv, VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + int element_size = native_slot_size(self->field_type); + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + + VALUE arg = argv[0]; + long beg, len; + + if (argc == 1){ + if (FIXNUM_P(arg)) { + /* standard case */ + void* memory; + int index = index_position(argv[0], self); + if (index < 0 || index >= self->size) { + return Qnil; + } + memory = RepeatedField_memoryat(self, index, element_size); + return native_slot_get(field_type, field_type_class, memory); + }else{ + /* check if idx is Range */ + switch (rb_range_beg_len(arg, &beg, &len, self->size, 0)) { + case Qfalse: + break; + case Qnil: + return Qnil; + default: + return RepeatedField_subarray(_self, beg, len); + } + } + } + /* assume 2 arguments */ + beg = NUM2LONG(argv[0]); + len = NUM2LONG(argv[1]); + if (beg < 0) { + beg += self->size; + } + if (beg >= self->size) { + return Qnil; + } + return RepeatedField_subarray(_self, beg, len); +} + +/* + * call-seq: + * RepeatedField.[]=(index, value) + * + * Sets the element at the given index. On out-of-bounds assignments, extends + * the array and fills the hole (if any) with default values. + */ +VALUE RepeatedField_index_set(VALUE _self, VALUE _index, VALUE val) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + int element_size = native_slot_size(field_type); + void* memory; + + int index = index_position(_index, self); + if (index < 0 || index >= (INT_MAX - 1)) { + return Qnil; + } + if (index >= self->size) { + upb_fieldtype_t field_type = self->field_type; + int element_size = native_slot_size(field_type); + RepeatedField_reserve(self, index + 1); + for (int i = self->size; i <= index; i++) { + void* elem = RepeatedField_memoryat(self, i, element_size); + native_slot_init(field_type, elem); + } + self->size = index + 1; + } + + memory = RepeatedField_memoryat(self, index, element_size); + native_slot_set(field_type, field_type_class, memory, val); + return Qnil; +} + +static int kInitialSize = 8; + +void RepeatedField_reserve(RepeatedField* self, int new_size) { + void* old_elems = self->elements; + int elem_size = native_slot_size(self->field_type); + if (new_size <= self->capacity) { + return; + } + if (self->capacity == 0) { + self->capacity = kInitialSize; + } + while (self->capacity < new_size) { + self->capacity *= 2; + } + self->elements = ALLOC_N(uint8_t, elem_size * self->capacity); + if (old_elems != NULL) { + memcpy(self->elements, old_elems, self->size * elem_size); + xfree(old_elems); + } +} + +/* + * call-seq: + * RepeatedField.push(value) + * + * Adds a new element to the repeated field. + */ +VALUE RepeatedField_push(VALUE _self, VALUE val) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + int element_size = native_slot_size(field_type); + void* memory; + + RepeatedField_reserve(self, self->size + 1); + memory = (void *) (((uint8_t *)self->elements) + self->size * element_size); + native_slot_set(field_type, self->field_type_class, memory, val); + // native_slot_set may raise an error; bump size only after set. + self->size++; + return _self; +} + + +// Used by parsing handlers. +void RepeatedField_push_native(VALUE _self, void* data) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + int element_size = native_slot_size(field_type); + void* memory; + + RepeatedField_reserve(self, self->size + 1); + memory = (void *) (((uint8_t *)self->elements) + self->size * element_size); + memcpy(memory, data, element_size); + self->size++; +} + +void* RepeatedField_index_native(VALUE _self, int index) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + int element_size = native_slot_size(field_type); + return RepeatedField_memoryat(self, index, element_size); +} + +/* + * Private ruby method, used by RepeatedField.pop + */ +VALUE RepeatedField_pop_one(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + int element_size = native_slot_size(field_type); + int index; + void* memory; + VALUE ret; + + if (self->size == 0) { + return Qnil; + } + index = self->size - 1; + memory = RepeatedField_memoryat(self, index, element_size); + ret = native_slot_get(field_type, field_type_class, memory); + self->size--; + return ret; +} + +/* + * call-seq: + * RepeatedField.replace(list) + * + * Replaces the contents of the repeated field with the given list of elements. + */ +VALUE RepeatedField_replace(VALUE _self, VALUE list) { + RepeatedField* self = ruby_to_RepeatedField(_self); + Check_Type(list, T_ARRAY); + self->size = 0; + for (int i = 0; i < RARRAY_LEN(list); i++) { + RepeatedField_push(_self, rb_ary_entry(list, i)); + } + return list; +} + +/* + * call-seq: + * RepeatedField.clear + * + * Clears (removes all elements from) this repeated field. + */ +VALUE RepeatedField_clear(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + self->size = 0; + return _self; +} + +/* + * call-seq: + * RepeatedField.length + * + * Returns the length of this repeated field. + */ +VALUE RepeatedField_length(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + return INT2NUM(self->size); +} + +static VALUE RepeatedField_new_this_type(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + VALUE new_rptfield = Qnil; + VALUE element_type = fieldtype_to_ruby(self->field_type); + if (self->field_type_class != Qnil) { + new_rptfield = rb_funcall(CLASS_OF(_self), rb_intern("new"), 2, + element_type, self->field_type_class); + } else { + new_rptfield = rb_funcall(CLASS_OF(_self), rb_intern("new"), 1, + element_type); + } + return new_rptfield; +} + +/* + * call-seq: + * RepeatedField.dup => repeated_field + * + * Duplicates this repeated field with a shallow copy. References to all + * non-primitive element objects (e.g., submessages) are shared. + */ +VALUE RepeatedField_dup(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + VALUE new_rptfield = RepeatedField_new_this_type(_self); + RepeatedField* new_rptfield_self = ruby_to_RepeatedField(new_rptfield); + upb_fieldtype_t field_type = self->field_type; + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + RepeatedField_reserve(new_rptfield_self, self->size); + for (int i = 0; i < self->size; i++, off += elem_size) { + void* to_mem = (uint8_t *)new_rptfield_self->elements + off; + void* from_mem = (uint8_t *)self->elements + off; + native_slot_dup(field_type, to_mem, from_mem); + new_rptfield_self->size++; + } + + return new_rptfield; +} + +// Internal only: used by Google::Protobuf.deep_copy. +VALUE RepeatedField_deep_copy(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + VALUE new_rptfield = RepeatedField_new_this_type(_self); + RepeatedField* new_rptfield_self = ruby_to_RepeatedField(new_rptfield); + upb_fieldtype_t field_type = self->field_type; + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + RepeatedField_reserve(new_rptfield_self, self->size); + for (int i = 0; i < self->size; i++, off += elem_size) { + void* to_mem = (uint8_t *)new_rptfield_self->elements + off; + void* from_mem = (uint8_t *)self->elements + off; + native_slot_deep_copy(field_type, to_mem, from_mem); + new_rptfield_self->size++; + } + + return new_rptfield; +} + +/* + * call-seq: + * RepeatedField.to_ary => array + * + * Used when converted implicitly into array, e.g. compared to an Array. + * Also called as a fallback of Object#to_a + */ +VALUE RepeatedField_to_ary(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + VALUE ary = rb_ary_new2(self->size); + for (int i = 0; i < self->size; i++, off += elem_size) { + void* mem = ((uint8_t *)self->elements) + off; + VALUE elem = native_slot_get(field_type, self->field_type_class, mem); + rb_ary_push(ary, elem); + } + return ary; +} + +/* + * call-seq: + * RepeatedField.==(other) => boolean + * + * Compares this repeated field to another. Repeated fields are equal if their + * element types are equal, their lengths are equal, and each element is equal. + * Elements are compared as per normal Ruby semantics, by calling their :== + * methods (or performing a more efficient comparison for primitive types). + * + * Repeated fields with dissimilar element types are never equal, even if value + * comparison (for example, between integers and floats) would have otherwise + * indicated that every element has equal value. + */ +VALUE RepeatedField_eq(VALUE _self, VALUE _other) { + RepeatedField* self; + RepeatedField* other; + + if (_self == _other) { + return Qtrue; + } + + if (TYPE(_other) == T_ARRAY) { + VALUE self_ary = RepeatedField_to_ary(_self); + return rb_equal(self_ary, _other); + } + + self = ruby_to_RepeatedField(_self); + other = ruby_to_RepeatedField(_other); + if (self->field_type != other->field_type || + self->field_type_class != other->field_type_class || + self->size != other->size) { + return Qfalse; + } + + { + upb_fieldtype_t field_type = self->field_type; + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + for (int i = 0; i < self->size; i++, off += elem_size) { + void* self_mem = ((uint8_t *)self->elements) + off; + void* other_mem = ((uint8_t *)other->elements) + off; + if (!native_slot_eq(field_type, self_mem, other_mem)) { + return Qfalse; + } + } + return Qtrue; + } +} + +/* + * call-seq: + * RepeatedField.hash => hash_value + * + * Returns a hash value computed from this repeated field's elements. + */ +VALUE RepeatedField_hash(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + + VALUE hash = LL2NUM(0); + + upb_fieldtype_t field_type = self->field_type; + VALUE field_type_class = self->field_type_class; + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + for (int i = 0; i < self->size; i++, off += elem_size) { + void* mem = ((uint8_t *)self->elements) + off; + VALUE elem = native_slot_get(field_type, field_type_class, mem); + hash = rb_funcall(hash, rb_intern("<<"), 1, INT2NUM(2)); + hash = rb_funcall(hash, rb_intern("^"), 1, + rb_funcall(elem, rb_intern("hash"), 0)); + } + + return hash; +} + +/* + * call-seq: + * RepeatedField.+(other) => repeated field + * + * Returns a new repeated field that contains the concatenated list of this + * repeated field's elements and other's elements. The other (second) list may + * be either another repeated field or a Ruby array. + */ +VALUE RepeatedField_plus(VALUE _self, VALUE list) { + VALUE dupped = RepeatedField_dup(_self); + + if (TYPE(list) == T_ARRAY) { + for (int i = 0; i < RARRAY_LEN(list); i++) { + VALUE elem = rb_ary_entry(list, i); + RepeatedField_push(dupped, elem); + } + } else if (RB_TYPE_P(list, T_DATA) && RTYPEDDATA_P(list) && + RTYPEDDATA_TYPE(list) == &RepeatedField_type) { + RepeatedField* self = ruby_to_RepeatedField(_self); + RepeatedField* list_rptfield = ruby_to_RepeatedField(list); + if (self->field_type != list_rptfield->field_type || + self->field_type_class != list_rptfield->field_type_class) { + rb_raise(rb_eArgError, + "Attempt to append RepeatedField with different element type."); + } + for (int i = 0; i < list_rptfield->size; i++) { + void* mem = RepeatedField_index_native(list, i); + RepeatedField_push_native(dupped, mem); + } + } else { + rb_raise(rb_eArgError, "Unknown type appending to RepeatedField"); + } + + return dupped; +} + +/* + * call-seq: + * RepeatedField.concat(other) => self + * + * concats the passed in array to self. Returns a Ruby array. + */ +VALUE RepeatedField_concat(VALUE _self, VALUE list) { + Check_Type(list, T_ARRAY); + for (int i = 0; i < RARRAY_LEN(list); i++) { + RepeatedField_push(_self, rb_ary_entry(list, i)); + } + return _self; +} + + +void validate_type_class(upb_fieldtype_t type, VALUE klass) { + if (rb_ivar_get(klass, descriptor_instancevar_interned) == Qnil) { + rb_raise(rb_eArgError, + "Type class has no descriptor. Please pass a " + "class or enum as returned by the DescriptorPool."); + } + if (type == UPB_TYPE_MESSAGE) { + VALUE desc = rb_ivar_get(klass, descriptor_instancevar_interned); + if (!RB_TYPE_P(desc, T_DATA) || !RTYPEDDATA_P(desc) || + RTYPEDDATA_TYPE(desc) != &_Descriptor_type) { + rb_raise(rb_eArgError, "Descriptor has an incorrect type."); + } + if (rb_get_alloc_func(klass) != &Message_alloc) { + rb_raise(rb_eArgError, + "Message class was not returned by the DescriptorPool."); + } + } else if (type == UPB_TYPE_ENUM) { + VALUE enumdesc = rb_ivar_get(klass, descriptor_instancevar_interned); + if (!RB_TYPE_P(enumdesc, T_DATA) || !RTYPEDDATA_P(enumdesc) || + RTYPEDDATA_TYPE(enumdesc) != &_EnumDescriptor_type) { + rb_raise(rb_eArgError, "Descriptor has an incorrect type."); + } + } +} + +void RepeatedField_init_args(int argc, VALUE* argv, + VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + VALUE ary = Qnil; + if (argc < 1) { + rb_raise(rb_eArgError, "Expected at least 1 argument."); + } + self->field_type = ruby_to_fieldtype(argv[0]); + + if (self->field_type == UPB_TYPE_MESSAGE || + self->field_type == UPB_TYPE_ENUM) { + if (argc < 2) { + rb_raise(rb_eArgError, "Expected at least 2 arguments for message/enum."); + } + self->field_type_class = argv[1]; + if (argc > 2) { + ary = argv[2]; + } + validate_type_class(self->field_type, self->field_type_class); + } else { + if (argc > 2) { + rb_raise(rb_eArgError, "Too many arguments: expected 1 or 2."); + } + if (argc > 1) { + ary = argv[1]; + } + } + + if (ary != Qnil) { + if (!RB_TYPE_P(ary, T_ARRAY)) { + rb_raise(rb_eArgError, "Expected array as initialize argument"); + } + for (int i = 0; i < RARRAY_LEN(ary); i++) { + RepeatedField_push(_self, rb_ary_entry(ary, i)); + } + } +} + +// Mark, free, alloc, init and class setup functions. + +void RepeatedField_mark(void* _self) { + RepeatedField* self = (RepeatedField*)_self; + upb_fieldtype_t field_type = self->field_type; + int element_size = native_slot_size(field_type); + rb_gc_mark(self->field_type_class); + for (int i = 0; i < self->size; i++) { + void* memory = (((uint8_t *)self->elements) + i * element_size); + native_slot_mark(self->field_type, memory); + } +} + +void RepeatedField_free(void* _self) { + RepeatedField* self = (RepeatedField*)_self; + xfree(self->elements); + xfree(self); +} + +/* + * call-seq: + * RepeatedField.new(type, type_class = nil, initial_elems = []) + * + * Creates a new repeated field. The provided type must be a Ruby symbol, and + * can take on the same values as those accepted by FieldDescriptor#type=. If + * the type is :message or :enum, type_class must be non-nil, and must be the + * Ruby class or module returned by Descriptor#msgclass or + * EnumDescriptor#enummodule, respectively. An initial list of elements may also + * be provided. + */ +VALUE RepeatedField_alloc(VALUE klass) { + RepeatedField* self = ALLOC(RepeatedField); + self->elements = NULL; + self->size = 0; + self->capacity = 0; + self->field_type = -1; + self->field_type_class = Qnil; + return TypedData_Wrap_Struct(klass, &RepeatedField_type, self); +} + +VALUE RepeatedField_init(int argc, VALUE* argv, VALUE self) { + RepeatedField_init_args(argc, argv, self); + return Qnil; +} + +void RepeatedField_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "RepeatedField", rb_cObject); + rb_define_alloc_func(klass, RepeatedField_alloc); + cRepeatedField = klass; + rb_gc_register_address(&cRepeatedField); + + rb_define_method(klass, "initialize", + RepeatedField_init, -1); + rb_define_method(klass, "each", RepeatedField_each, 0); + rb_define_method(klass, "[]", RepeatedField_index, -1); + rb_define_method(klass, "at", RepeatedField_index, -1); + rb_define_method(klass, "[]=", RepeatedField_index_set, 2); + rb_define_method(klass, "push", RepeatedField_push, 1); + rb_define_method(klass, "<<", RepeatedField_push, 1); + rb_define_private_method(klass, "pop_one", RepeatedField_pop_one, 0); + rb_define_method(klass, "replace", RepeatedField_replace, 1); + rb_define_method(klass, "clear", RepeatedField_clear, 0); + rb_define_method(klass, "length", RepeatedField_length, 0); + rb_define_method(klass, "size", RepeatedField_length, 0); + rb_define_method(klass, "dup", RepeatedField_dup, 0); + // Also define #clone so that we don't inherit Object#clone. + rb_define_method(klass, "clone", RepeatedField_dup, 0); + rb_define_method(klass, "==", RepeatedField_eq, 1); + rb_define_method(klass, "to_ary", RepeatedField_to_ary, 0); + rb_define_method(klass, "hash", RepeatedField_hash, 0); + rb_define_method(klass, "+", RepeatedField_plus, 1); + rb_define_method(klass, "concat", RepeatedField_concat, 1); + rb_include_module(klass, rb_mEnumerable); +} diff --git a/ruby/ext/google/protobuf_c/storage.c b/ruby/ext/google/protobuf_c/storage.c new file mode 100644 index 00000000..1c839781 --- /dev/null +++ b/ruby/ext/google/protobuf_c/storage.c @@ -0,0 +1,894 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +#include <math.h> + +#include <ruby/encoding.h> + +// ----------------------------------------------------------------------------- +// Ruby <-> native slot management. +// ----------------------------------------------------------------------------- + +#define DEREF(memory, type) *(type*)(memory) + +size_t native_slot_size(upb_fieldtype_t type) { + switch (type) { + case UPB_TYPE_FLOAT: return 4; + case UPB_TYPE_DOUBLE: return 8; + case UPB_TYPE_BOOL: return 1; + case UPB_TYPE_STRING: return sizeof(VALUE); + case UPB_TYPE_BYTES: return sizeof(VALUE); + case UPB_TYPE_MESSAGE: return sizeof(VALUE); + case UPB_TYPE_ENUM: return 4; + case UPB_TYPE_INT32: return 4; + case UPB_TYPE_INT64: return 8; + case UPB_TYPE_UINT32: return 4; + case UPB_TYPE_UINT64: return 8; + default: return 0; + } +} + +static VALUE value_from_default(const upb_fielddef *field) { + switch (upb_fielddef_type(field)) { + case UPB_TYPE_FLOAT: return DBL2NUM(upb_fielddef_defaultfloat(field)); + case UPB_TYPE_DOUBLE: return DBL2NUM(upb_fielddef_defaultdouble(field)); + case UPB_TYPE_BOOL: + return upb_fielddef_defaultbool(field) ? Qtrue : Qfalse; + case UPB_TYPE_MESSAGE: return Qnil; + case UPB_TYPE_ENUM: { + const upb_enumdef *enumdef = upb_fielddef_enumsubdef(field); + int32_t num = upb_fielddef_defaultint32(field); + const char *label = upb_enumdef_iton(enumdef, num); + if (label) { + return ID2SYM(rb_intern(label)); + } else { + return INT2NUM(num); + } + } + case UPB_TYPE_INT32: return INT2NUM(upb_fielddef_defaultint32(field)); + case UPB_TYPE_INT64: return LL2NUM(upb_fielddef_defaultint64(field));; + case UPB_TYPE_UINT32: return UINT2NUM(upb_fielddef_defaultuint32(field)); + case UPB_TYPE_UINT64: return ULL2NUM(upb_fielddef_defaultuint64(field)); + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + size_t size; + const char *str = upb_fielddef_defaultstr(field, &size); + return rb_str_new(str, size); + } + default: return Qnil; + } +} + +static bool is_ruby_num(VALUE value) { + return (TYPE(value) == T_FLOAT || + TYPE(value) == T_FIXNUM || + TYPE(value) == T_BIGNUM); +} + +void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE val) { + if (!is_ruby_num(val)) { + rb_raise(rb_eTypeError, "Expected number type for integral field."); + } + + // NUM2{INT,UINT,LL,ULL} macros do the appropriate range checks on upper + // bound; we just need to do precision checks (i.e., disallow rounding) and + // check for < 0 on unsigned types. + if (TYPE(val) == T_FLOAT) { + double dbl_val = NUM2DBL(val); + if (floor(dbl_val) != dbl_val) { + rb_raise(rb_eRangeError, + "Non-integral floating point value assigned to integer field."); + } + } + if (type == UPB_TYPE_UINT32 || type == UPB_TYPE_UINT64) { + if (NUM2DBL(val) < 0) { + rb_raise(rb_eRangeError, + "Assigning negative value to unsigned integer field."); + } + } +} + +void native_slot_validate_string_encoding(upb_fieldtype_t type, VALUE value) { + bool bad_encoding = false; + rb_encoding* string_encoding = rb_enc_from_index(ENCODING_GET(value)); + if (type == UPB_TYPE_STRING) { + bad_encoding = + string_encoding != kRubyStringUtf8Encoding && + string_encoding != kRubyStringASCIIEncoding; + } else { + bad_encoding = + string_encoding != kRubyString8bitEncoding; + } + // Check that encoding is UTF-8 or ASCII (for string fields) or ASCII-8BIT + // (for bytes fields). + if (bad_encoding) { + rb_raise(rb_eTypeError, "Encoding for '%s' fields must be %s (was %s)", + (type == UPB_TYPE_STRING) ? "string" : "bytes", + (type == UPB_TYPE_STRING) ? "UTF-8 or ASCII" : "ASCII-8BIT", + rb_enc_name(string_encoding)); + } +} + +void native_slot_set(upb_fieldtype_t type, VALUE type_class, + void* memory, VALUE value) { + native_slot_set_value_and_case(type, type_class, memory, value, NULL, 0); +} + +void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class, + void* memory, VALUE value, + uint32_t* case_memory, + uint32_t case_number) { + // Note that in order to atomically change the value in memory and the case + // value (w.r.t. Ruby VM calls), we must set the value at |memory| only after + // all Ruby VM calls are complete. The case is then set at the bottom of this + // function. + switch (type) { + case UPB_TYPE_FLOAT: + if (!is_ruby_num(value)) { + rb_raise(rb_eTypeError, "Expected number type for float field."); + } + DEREF(memory, float) = NUM2DBL(value); + break; + case UPB_TYPE_DOUBLE: + if (!is_ruby_num(value)) { + rb_raise(rb_eTypeError, "Expected number type for double field."); + } + DEREF(memory, double) = NUM2DBL(value); + break; + case UPB_TYPE_BOOL: { + int8_t val = -1; + if (value == Qtrue) { + val = 1; + } else if (value == Qfalse) { + val = 0; + } else { + rb_raise(rb_eTypeError, "Invalid argument for boolean field."); + } + DEREF(memory, int8_t) = val; + break; + } + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + if (CLASS_OF(value) != rb_cString) { + rb_raise(rb_eTypeError, "Invalid argument for string field."); + } + native_slot_validate_string_encoding(type, value); + DEREF(memory, VALUE) = value; + break; + } + case UPB_TYPE_MESSAGE: { + if (CLASS_OF(value) == CLASS_OF(Qnil)) { + value = Qnil; + } else if (CLASS_OF(value) != type_class) { + rb_raise(rb_eTypeError, + "Invalid type %s to assign to submessage field.", + rb_class2name(CLASS_OF(value))); + } + DEREF(memory, VALUE) = value; + break; + } + case UPB_TYPE_ENUM: { + int32_t int_val = 0; + if (!is_ruby_num(value) && TYPE(value) != T_SYMBOL) { + rb_raise(rb_eTypeError, + "Expected number or symbol type for enum field."); + } + if (TYPE(value) == T_SYMBOL) { + // Ensure that the given symbol exists in the enum module. + VALUE lookup = rb_funcall(type_class, rb_intern("resolve"), 1, value); + if (lookup == Qnil) { + rb_raise(rb_eRangeError, "Unknown symbol value for enum field."); + } else { + int_val = NUM2INT(lookup); + } + } else { + native_slot_check_int_range_precision(UPB_TYPE_INT32, value); + int_val = NUM2INT(value); + } + DEREF(memory, int32_t) = int_val; + break; + } + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + native_slot_check_int_range_precision(type, value); + switch (type) { + case UPB_TYPE_INT32: + DEREF(memory, int32_t) = NUM2INT(value); + break; + case UPB_TYPE_INT64: + DEREF(memory, int64_t) = NUM2LL(value); + break; + case UPB_TYPE_UINT32: + DEREF(memory, uint32_t) = NUM2UINT(value); + break; + case UPB_TYPE_UINT64: + DEREF(memory, uint64_t) = NUM2ULL(value); + break; + default: + break; + } + break; + default: + break; + } + + if (case_memory != NULL) { + *case_memory = case_number; + } +} + +VALUE native_slot_get(upb_fieldtype_t type, + VALUE type_class, + const void* memory) { + switch (type) { + case UPB_TYPE_FLOAT: + return DBL2NUM(DEREF(memory, float)); + case UPB_TYPE_DOUBLE: + return DBL2NUM(DEREF(memory, double)); + case UPB_TYPE_BOOL: + return DEREF(memory, int8_t) ? Qtrue : Qfalse; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + case UPB_TYPE_MESSAGE: + return DEREF(memory, VALUE); + case UPB_TYPE_ENUM: { + int32_t val = DEREF(memory, int32_t); + VALUE symbol = enum_lookup(type_class, INT2NUM(val)); + if (symbol == Qnil) { + return INT2NUM(val); + } else { + return symbol; + } + } + case UPB_TYPE_INT32: + return INT2NUM(DEREF(memory, int32_t)); + case UPB_TYPE_INT64: + return LL2NUM(DEREF(memory, int64_t)); + case UPB_TYPE_UINT32: + return UINT2NUM(DEREF(memory, uint32_t)); + case UPB_TYPE_UINT64: + return ULL2NUM(DEREF(memory, uint64_t)); + default: + return Qnil; + } +} + +void native_slot_init(upb_fieldtype_t type, void* memory) { + switch (type) { + case UPB_TYPE_FLOAT: + DEREF(memory, float) = 0.0; + break; + case UPB_TYPE_DOUBLE: + DEREF(memory, double) = 0.0; + break; + case UPB_TYPE_BOOL: + DEREF(memory, int8_t) = 0; + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + DEREF(memory, VALUE) = rb_str_new2(""); + rb_enc_associate(DEREF(memory, VALUE), (type == UPB_TYPE_BYTES) ? + kRubyString8bitEncoding : kRubyStringUtf8Encoding); + break; + case UPB_TYPE_MESSAGE: + DEREF(memory, VALUE) = Qnil; + break; + case UPB_TYPE_ENUM: + case UPB_TYPE_INT32: + DEREF(memory, int32_t) = 0; + break; + case UPB_TYPE_INT64: + DEREF(memory, int64_t) = 0; + break; + case UPB_TYPE_UINT32: + DEREF(memory, uint32_t) = 0; + break; + case UPB_TYPE_UINT64: + DEREF(memory, uint64_t) = 0; + break; + default: + break; + } +} + +void native_slot_mark(upb_fieldtype_t type, void* memory) { + switch (type) { + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + case UPB_TYPE_MESSAGE: + rb_gc_mark(DEREF(memory, VALUE)); + break; + default: + break; + } +} + +void native_slot_dup(upb_fieldtype_t type, void* to, void* from) { + memcpy(to, from, native_slot_size(type)); +} + +void native_slot_deep_copy(upb_fieldtype_t type, void* to, void* from) { + switch (type) { + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + VALUE from_val = DEREF(from, VALUE); + DEREF(to, VALUE) = (from_val != Qnil) ? + rb_funcall(from_val, rb_intern("dup"), 0) : Qnil; + break; + } + case UPB_TYPE_MESSAGE: { + VALUE from_val = DEREF(from, VALUE); + DEREF(to, VALUE) = (from_val != Qnil) ? + Message_deep_copy(from_val) : Qnil; + break; + } + default: + memcpy(to, from, native_slot_size(type)); + } +} + +bool native_slot_eq(upb_fieldtype_t type, void* mem1, void* mem2) { + switch (type) { + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + case UPB_TYPE_MESSAGE: { + VALUE val1 = DEREF(mem1, VALUE); + VALUE val2 = DEREF(mem2, VALUE); + VALUE ret = rb_funcall(val1, rb_intern("=="), 1, val2); + return ret == Qtrue; + } + default: + return !memcmp(mem1, mem2, native_slot_size(type)); + } +} + +// ----------------------------------------------------------------------------- +// Map field utilities. +// ----------------------------------------------------------------------------- + +const upb_msgdef* tryget_map_entry_msgdef(const upb_fielddef* field) { + const upb_msgdef* subdef; + if (upb_fielddef_label(field) != UPB_LABEL_REPEATED || + upb_fielddef_type(field) != UPB_TYPE_MESSAGE) { + return NULL; + } + subdef = upb_fielddef_msgsubdef(field); + return upb_msgdef_mapentry(subdef) ? subdef : NULL; +} + +const upb_msgdef *map_entry_msgdef(const upb_fielddef* field) { + const upb_msgdef* subdef = tryget_map_entry_msgdef(field); + assert(subdef); + return subdef; +} + +bool is_map_field(const upb_fielddef *field) { + return tryget_map_entry_msgdef(field) != NULL; +} + +const upb_fielddef* map_field_key(const upb_fielddef* field) { + const upb_msgdef* subdef = map_entry_msgdef(field); + return map_entry_key(subdef); +} + +const upb_fielddef* map_field_value(const upb_fielddef* field) { + const upb_msgdef* subdef = map_entry_msgdef(field); + return map_entry_value(subdef); +} + +const upb_fielddef* map_entry_key(const upb_msgdef* msgdef) { + const upb_fielddef* key_field = upb_msgdef_itof(msgdef, MAP_KEY_FIELD); + assert(key_field != NULL); + return key_field; +} + +const upb_fielddef* map_entry_value(const upb_msgdef* msgdef) { + const upb_fielddef* value_field = upb_msgdef_itof(msgdef, MAP_VALUE_FIELD); + assert(value_field != NULL); + return value_field; +} + +// ----------------------------------------------------------------------------- +// Memory layout management. +// ----------------------------------------------------------------------------- + +static size_t align_up_to(size_t offset, size_t granularity) { + // Granularity must be a power of two. + return (offset + granularity - 1) & ~(granularity - 1); +} + +MessageLayout* create_layout(const upb_msgdef* msgdef) { + MessageLayout* layout = ALLOC(MessageLayout); + int nfields = upb_msgdef_numfields(msgdef); + upb_msg_field_iter it; + upb_msg_oneof_iter oit; + size_t off = 0; + + layout->fields = ALLOC_N(MessageField, nfields); + + for (upb_msg_field_begin(&it, msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + size_t field_size; + + if (upb_fielddef_containingoneof(field)) { + // Oneofs are handled separately below. + continue; + } + + // Allocate |field_size| bytes for this field in the layout. + field_size = 0; + if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + field_size = sizeof(VALUE); + } else { + field_size = native_slot_size(upb_fielddef_type(field)); + } + // Align current offset up to |size| granularity. + off = align_up_to(off, field_size); + layout->fields[upb_fielddef_index(field)].offset = off; + layout->fields[upb_fielddef_index(field)].case_offset = + MESSAGE_FIELD_NO_CASE; + off += field_size; + } + + // Handle oneofs now -- we iterate over oneofs specifically and allocate only + // one slot per oneof. + // + // We assign all value slots first, then pack the 'case' fields at the end, + // since in the common case (modern 64-bit platform) these are 8 bytes and 4 + // bytes respectively and we want to avoid alignment overhead. + // + // Note that we reserve 4 bytes (a uint32) per 'case' slot because the value + // space for oneof cases is conceptually as wide as field tag numbers. In + // practice, it's unlikely that a oneof would have more than e.g. 256 or 64K + // members (8 or 16 bits respectively), so conceivably we could assign + // consecutive case numbers and then pick a smaller oneof case slot size, but + // the complexity to implement this indirection is probably not worthwhile. + for (upb_msg_oneof_begin(&oit, msgdef); + !upb_msg_oneof_done(&oit); + upb_msg_oneof_next(&oit)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit); + upb_oneof_iter fit; + + // Always allocate NATIVE_SLOT_MAX_SIZE bytes, but share the slot between + // all fields. + size_t field_size = NATIVE_SLOT_MAX_SIZE; + // Align the offset. + off = align_up_to(off, field_size); + // Assign all fields in the oneof this same offset. + for (upb_oneof_begin(&fit, oneof); + !upb_oneof_done(&fit); + upb_oneof_next(&fit)) { + const upb_fielddef* field = upb_oneof_iter_field(&fit); + layout->fields[upb_fielddef_index(field)].offset = off; + } + off += field_size; + } + + // Now the case fields. + for (upb_msg_oneof_begin(&oit, msgdef); + !upb_msg_oneof_done(&oit); + upb_msg_oneof_next(&oit)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit); + upb_oneof_iter fit; + + size_t field_size = sizeof(uint32_t); + // Align the offset. + off = (off + field_size - 1) & ~(field_size - 1); + // Assign all fields in the oneof this same offset. + for (upb_oneof_begin(&fit, oneof); + !upb_oneof_done(&fit); + upb_oneof_next(&fit)) { + const upb_fielddef* field = upb_oneof_iter_field(&fit); + layout->fields[upb_fielddef_index(field)].case_offset = off; + } + off += field_size; + } + + layout->size = off; + + layout->msgdef = msgdef; + upb_msgdef_ref(layout->msgdef, &layout->msgdef); + + return layout; +} + +void free_layout(MessageLayout* layout) { + xfree(layout->fields); + upb_msgdef_unref(layout->msgdef, &layout->msgdef); + xfree(layout); +} + +VALUE field_type_class(const upb_fielddef* field) { + VALUE type_class = Qnil; + if (upb_fielddef_type(field) == UPB_TYPE_MESSAGE) { + VALUE submsgdesc = + get_def_obj(upb_fielddef_subdef(field)); + type_class = Descriptor_msgclass(submsgdesc); + } else if (upb_fielddef_type(field) == UPB_TYPE_ENUM) { + VALUE subenumdesc = + get_def_obj(upb_fielddef_subdef(field)); + type_class = EnumDescriptor_enummodule(subenumdesc); + } + return type_class; +} + +static void* slot_memory(MessageLayout* layout, + const void* storage, + const upb_fielddef* field) { + return ((uint8_t *)storage) + + layout->fields[upb_fielddef_index(field)].offset; +} + +static uint32_t* slot_oneof_case(MessageLayout* layout, + const void* storage, + const upb_fielddef* field) { + return (uint32_t *)(((uint8_t *)storage) + + layout->fields[upb_fielddef_index(field)].case_offset); +} + + +VALUE layout_get(MessageLayout* layout, + const void* storage, + const upb_fielddef* field) { + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + if (*oneof_case != upb_fielddef_number(field)) { + return value_from_default(field); + } + return native_slot_get(upb_fielddef_type(field), + field_type_class(field), + memory); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + return *((VALUE *)memory); + } else { + return native_slot_get(upb_fielddef_type(field), + field_type_class(field), + memory); + } +} + +static void check_repeated_field_type(VALUE val, const upb_fielddef* field) { + RepeatedField* self; + assert(upb_fielddef_label(field) == UPB_LABEL_REPEATED); + + if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) || + RTYPEDDATA_TYPE(val) != &RepeatedField_type) { + rb_raise(rb_eTypeError, "Expected repeated field array"); + } + + self = ruby_to_RepeatedField(val); + if (self->field_type != upb_fielddef_type(field)) { + rb_raise(rb_eTypeError, "Repeated field array has wrong element type"); + } + + if (self->field_type == UPB_TYPE_MESSAGE || + self->field_type == UPB_TYPE_ENUM) { + if (self->field_type_class != + get_def_obj(upb_fielddef_subdef(field))) { + rb_raise(rb_eTypeError, + "Repeated field array has wrong message/enum class"); + } + } +} + +static void check_map_field_type(VALUE val, const upb_fielddef* field) { + const upb_fielddef* key_field = map_field_key(field); + const upb_fielddef* value_field = map_field_value(field); + Map* self; + + if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) || + RTYPEDDATA_TYPE(val) != &Map_type) { + rb_raise(rb_eTypeError, "Expected Map instance"); + } + + self = ruby_to_Map(val); + if (self->key_type != upb_fielddef_type(key_field)) { + rb_raise(rb_eTypeError, "Map key type does not match field's key type"); + } + if (self->value_type != upb_fielddef_type(value_field)) { + rb_raise(rb_eTypeError, "Map value type does not match field's value type"); + } + if (upb_fielddef_type(value_field) == UPB_TYPE_MESSAGE || + upb_fielddef_type(value_field) == UPB_TYPE_ENUM) { + if (self->value_type_class != + get_def_obj(upb_fielddef_subdef(value_field))) { + rb_raise(rb_eTypeError, + "Map value type has wrong message/enum class"); + } + } +} + + +void layout_set(MessageLayout* layout, + void* storage, + const upb_fielddef* field, + VALUE val) { + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + if (val == Qnil) { + // Assigning nil to a oneof field clears the oneof completely. + *oneof_case = ONEOF_CASE_NONE; + memset(memory, 0, NATIVE_SLOT_MAX_SIZE); + } else { + // The transition between field types for a single oneof (union) slot is + // somewhat complex because we need to ensure that a GC triggered at any + // point by a call into the Ruby VM sees a valid state for this field and + // does not either go off into the weeds (following what it thinks is a + // VALUE but is actually a different field type) or miss an object (seeing + // what it thinks is a primitive field but is actually a VALUE for the new + // field type). + // + // In order for the transition to be safe, the oneof case slot must be in + // sync with the value slot whenever the Ruby VM has been called. Thus, we + // use native_slot_set_value_and_case(), which ensures that both the value + // and case number are altered atomically (w.r.t. the Ruby VM). + native_slot_set_value_and_case( + upb_fielddef_type(field), field_type_class(field), + memory, val, + oneof_case, upb_fielddef_number(field)); + } + } else if (is_map_field(field)) { + check_map_field_type(val, field); + DEREF(memory, VALUE) = val; + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + check_repeated_field_type(val, field); + DEREF(memory, VALUE) = val; + } else { + native_slot_set(upb_fielddef_type(field), field_type_class(field), + memory, val); + } +} + +void layout_init(MessageLayout* layout, + void* storage) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + memset(memory, 0, NATIVE_SLOT_MAX_SIZE); + *oneof_case = ONEOF_CASE_NONE; + } else if (is_map_field(field)) { + VALUE map = Qnil; + + const upb_fielddef* key_field = map_field_key(field); + const upb_fielddef* value_field = map_field_value(field); + VALUE type_class = field_type_class(value_field); + + if (type_class != Qnil) { + VALUE args[3] = { + fieldtype_to_ruby(upb_fielddef_type(key_field)), + fieldtype_to_ruby(upb_fielddef_type(value_field)), + type_class, + }; + map = rb_class_new_instance(3, args, cMap); + } else { + VALUE args[2] = { + fieldtype_to_ruby(upb_fielddef_type(key_field)), + fieldtype_to_ruby(upb_fielddef_type(value_field)), + }; + map = rb_class_new_instance(2, args, cMap); + } + + DEREF(memory, VALUE) = map; + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + VALUE ary = Qnil; + + VALUE type_class = field_type_class(field); + + if (type_class != Qnil) { + VALUE args[2] = { + fieldtype_to_ruby(upb_fielddef_type(field)), + type_class, + }; + ary = rb_class_new_instance(2, args, cRepeatedField); + } else { + VALUE args[1] = { fieldtype_to_ruby(upb_fielddef_type(field)) }; + ary = rb_class_new_instance(1, args, cRepeatedField); + } + + DEREF(memory, VALUE) = ary; + } else { + native_slot_init(upb_fielddef_type(field), memory); + } + } +} + +void layout_mark(MessageLayout* layout, void* storage) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + if (*oneof_case == upb_fielddef_number(field)) { + native_slot_mark(upb_fielddef_type(field), memory); + } + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + rb_gc_mark(DEREF(memory, VALUE)); + } else { + native_slot_mark(upb_fielddef_type(field), memory); + } + } +} + +void layout_dup(MessageLayout* layout, void* to, void* from) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + + void* to_memory = slot_memory(layout, to, field); + uint32_t* to_oneof_case = slot_oneof_case(layout, to, field); + void* from_memory = slot_memory(layout, from, field); + uint32_t* from_oneof_case = slot_oneof_case(layout, from, field); + + if (upb_fielddef_containingoneof(field)) { + if (*from_oneof_case == upb_fielddef_number(field)) { + *to_oneof_case = *from_oneof_case; + native_slot_dup(upb_fielddef_type(field), to_memory, from_memory); + } + } else if (is_map_field(field)) { + DEREF(to_memory, VALUE) = Map_dup(DEREF(from_memory, VALUE)); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + DEREF(to_memory, VALUE) = RepeatedField_dup(DEREF(from_memory, VALUE)); + } else { + native_slot_dup(upb_fielddef_type(field), to_memory, from_memory); + } + } +} + +void layout_deep_copy(MessageLayout* layout, void* to, void* from) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + + void* to_memory = slot_memory(layout, to, field); + uint32_t* to_oneof_case = slot_oneof_case(layout, to, field); + void* from_memory = slot_memory(layout, from, field); + uint32_t* from_oneof_case = slot_oneof_case(layout, from, field); + + if (upb_fielddef_containingoneof(field)) { + if (*from_oneof_case == upb_fielddef_number(field)) { + *to_oneof_case = *from_oneof_case; + native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory); + } + } else if (is_map_field(field)) { + DEREF(to_memory, VALUE) = + Map_deep_copy(DEREF(from_memory, VALUE)); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + DEREF(to_memory, VALUE) = + RepeatedField_deep_copy(DEREF(from_memory, VALUE)); + } else { + native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory); + } + } +} + +VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + + void* msg1_memory = slot_memory(layout, msg1, field); + uint32_t* msg1_oneof_case = slot_oneof_case(layout, msg1, field); + void* msg2_memory = slot_memory(layout, msg2, field); + uint32_t* msg2_oneof_case = slot_oneof_case(layout, msg2, field); + + if (upb_fielddef_containingoneof(field)) { + if (*msg1_oneof_case != *msg2_oneof_case || + (*msg1_oneof_case == upb_fielddef_number(field) && + !native_slot_eq(upb_fielddef_type(field), + msg1_memory, + msg2_memory))) { + return Qfalse; + } + } else if (is_map_field(field)) { + if (!Map_eq(DEREF(msg1_memory, VALUE), + DEREF(msg2_memory, VALUE))) { + return Qfalse; + } + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + if (!RepeatedField_eq(DEREF(msg1_memory, VALUE), + DEREF(msg2_memory, VALUE))) { + return Qfalse; + } + } else { + if (!native_slot_eq(upb_fielddef_type(field), + msg1_memory, msg2_memory)) { + return Qfalse; + } + } + } + return Qtrue; +} + +VALUE layout_hash(MessageLayout* layout, void* storage) { + upb_msg_field_iter it; + st_index_t h = rb_hash_start(0); + VALUE hash_sym = rb_intern("hash"); + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + VALUE field_val = layout_get(layout, storage, field); + h = rb_hash_uint(h, NUM2LONG(rb_funcall(field_val, hash_sym, 0))); + } + h = rb_hash_end(h); + + return INT2FIX(h); +} + +VALUE layout_inspect(MessageLayout* layout, void* storage) { + VALUE str = rb_str_new2(""); + + upb_msg_field_iter it; + bool first = true; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { + const upb_fielddef* field = upb_msg_iter_field(&it); + VALUE field_val = layout_get(layout, storage, field); + + if (!first) { + str = rb_str_cat2(str, ", "); + } else { + first = false; + } + str = rb_str_cat2(str, upb_fielddef_name(field)); + str = rb_str_cat2(str, ": "); + + str = rb_str_append(str, rb_funcall(field_val, rb_intern("inspect"), 0)); + } + + return str; +} diff --git a/ruby/ext/google/protobuf_c/upb.c b/ruby/ext/google/protobuf_c/upb.c new file mode 100644 index 00000000..74a2a1db --- /dev/null +++ b/ruby/ext/google/protobuf_c/upb.c @@ -0,0 +1,12812 @@ +// Amalgamated source file +#include "upb.h" + + +#include <ctype.h> +#include <stdlib.h> +#include <string.h> + +typedef struct { + size_t len; + char str[1]; /* Null-terminated string data follows. */ +} str_t; + +static str_t *newstr(const char *data, size_t len) { + str_t *ret = upb_gmalloc(sizeof(*ret) + len); + if (!ret) return NULL; + ret->len = len; + memcpy(ret->str, data, len); + ret->str[len] = '\0'; + return ret; +} + +static void freestr(str_t *s) { upb_gfree(s); } + +/* isalpha() etc. from <ctype.h> are locale-dependent, which we don't want. */ +static bool upb_isbetween(char c, char low, char high) { + return c >= low && c <= high; +} + +static bool upb_isletter(char c) { + return upb_isbetween(c, 'A', 'Z') || upb_isbetween(c, 'a', 'z') || c == '_'; +} + +static bool upb_isalphanum(char c) { + return upb_isletter(c) || upb_isbetween(c, '0', '9'); +} + +static bool upb_isident(const char *str, size_t len, bool full, upb_status *s) { + bool start = true; + size_t i; + for (i = 0; i < len; i++) { + char c = str[i]; + if (c == '.') { + if (start || !full) { + upb_status_seterrf(s, "invalid name: unexpected '.' (%s)", str); + return false; + } + start = true; + } else if (start) { + if (!upb_isletter(c)) { + upb_status_seterrf( + s, "invalid name: path components must start with a letter (%s)", + str); + return false; + } + start = false; + } else { + if (!upb_isalphanum(c)) { + upb_status_seterrf(s, "invalid name: non-alphanumeric character (%s)", + str); + return false; + } + } + } + return !start; +} + +static bool upb_isoneof(const upb_refcounted *def) { + return def->vtbl == &upb_oneofdef_vtbl; +} + +static bool upb_isfield(const upb_refcounted *def) { + return def->vtbl == &upb_fielddef_vtbl; +} + +static const upb_oneofdef *upb_trygetoneof(const upb_refcounted *def) { + return upb_isoneof(def) ? (const upb_oneofdef*)def : NULL; +} + +static const upb_fielddef *upb_trygetfield(const upb_refcounted *def) { + return upb_isfield(def) ? (const upb_fielddef*)def : NULL; +} + + +/* upb_def ********************************************************************/ + +upb_deftype_t upb_def_type(const upb_def *d) { return d->type; } + +const char *upb_def_fullname(const upb_def *d) { return d->fullname; } + +const char *upb_def_name(const upb_def *d) { + const char *p; + + if (d->fullname == NULL) { + return NULL; + } else if ((p = strrchr(d->fullname, '.')) == NULL) { + /* No '.' in the name, return the full string. */ + return d->fullname; + } else { + /* Return one past the last '.'. */ + return p + 1; + } +} + +bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s) { + assert(!upb_def_isfrozen(def)); + if (!upb_isident(fullname, strlen(fullname), true, s)) { + return false; + } + + fullname = upb_gstrdup(fullname); + if (!fullname) { + upb_upberr_setoom(s); + return false; + } + + upb_gfree((void*)def->fullname); + def->fullname = fullname; + return true; +} + +const upb_filedef *upb_def_file(const upb_def *d) { return d->file; } + +upb_def *upb_def_dup(const upb_def *def, const void *o) { + switch (def->type) { + case UPB_DEF_MSG: + return upb_msgdef_upcast_mutable( + upb_msgdef_dup(upb_downcast_msgdef(def), o)); + case UPB_DEF_FIELD: + return upb_fielddef_upcast_mutable( + upb_fielddef_dup(upb_downcast_fielddef(def), o)); + case UPB_DEF_ENUM: + return upb_enumdef_upcast_mutable( + upb_enumdef_dup(upb_downcast_enumdef(def), o)); + default: assert(false); return NULL; + } +} + +static bool upb_def_init(upb_def *def, upb_deftype_t type, + const struct upb_refcounted_vtbl *vtbl, + const void *owner) { + if (!upb_refcounted_init(upb_def_upcast_mutable(def), vtbl, owner)) return false; + def->type = type; + def->fullname = NULL; + def->came_from_user = false; + def->file = NULL; + return true; +} + +static void upb_def_uninit(upb_def *def) { + upb_gfree((void*)def->fullname); +} + +static const char *msgdef_name(const upb_msgdef *m) { + const char *name = upb_def_fullname(upb_msgdef_upcast(m)); + return name ? name : "(anonymous)"; +} + +static bool upb_validate_field(upb_fielddef *f, upb_status *s) { + if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "fielddef must have name and number set"); + return false; + } + + if (!f->type_is_set_) { + upb_status_seterrmsg(s, "fielddef type was not initialized"); + return false; + } + + if (upb_fielddef_lazy(f) && + upb_fielddef_descriptortype(f) != UPB_DESCRIPTOR_TYPE_MESSAGE) { + upb_status_seterrmsg(s, + "only length-delimited submessage fields may be lazy"); + return false; + } + + if (upb_fielddef_hassubdef(f)) { + const upb_def *subdef; + + if (f->subdef_is_symbolic) { + upb_status_seterrf(s, "field '%s.%s' has not been resolved", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + + subdef = upb_fielddef_subdef(f); + if (subdef == NULL) { + upb_status_seterrf(s, "field %s.%s is missing required subdef", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + + if (!upb_def_isfrozen(subdef) && !subdef->came_from_user) { + upb_status_seterrf(s, + "subdef of field %s.%s is not frozen or being frozen", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + } + + if (upb_fielddef_type(f) == UPB_TYPE_ENUM) { + bool has_default_name = upb_fielddef_enumhasdefaultstr(f); + bool has_default_number = upb_fielddef_enumhasdefaultint32(f); + + /* Previously verified by upb_validate_enumdef(). */ + assert(upb_enumdef_numvals(upb_fielddef_enumsubdef(f)) > 0); + + /* We've already validated that we have an associated enumdef and that it + * has at least one member, so at least one of these should be true. + * Because if the user didn't set anything, we'll pick up the enum's + * default, but if the user *did* set something we should at least pick up + * the one they set (int32 or string). */ + assert(has_default_name || has_default_number); + + if (!has_default_name) { + upb_status_seterrf(s, + "enum default for field %s.%s (%d) is not in the enum", + msgdef_name(f->msg.def), upb_fielddef_name(f), + upb_fielddef_defaultint32(f)); + return false; + } + + if (!has_default_number) { + upb_status_seterrf(s, + "enum default for field %s.%s (%s) is not in the enum", + msgdef_name(f->msg.def), upb_fielddef_name(f), + upb_fielddef_defaultstr(f, NULL)); + return false; + } + + /* Lift the effective numeric default into the field's default slot, in case + * we were only getting it "by reference" from the enumdef. */ + upb_fielddef_setdefaultint32(f, upb_fielddef_defaultint32(f)); + } + + /* Ensure that MapEntry submessages only appear as repeated fields, not + * optional/required (singular) fields. */ + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE && + upb_fielddef_msgsubdef(f) != NULL) { + const upb_msgdef *subdef = upb_fielddef_msgsubdef(f); + if (upb_msgdef_mapentry(subdef) && !upb_fielddef_isseq(f)) { + upb_status_seterrf(s, + "Field %s refers to mapentry message but is not " + "a repeated field", + upb_fielddef_name(f) ? upb_fielddef_name(f) : + "(unnamed)"); + return false; + } + } + + return true; +} + +static bool upb_validate_enumdef(const upb_enumdef *e, upb_status *s) { + if (upb_enumdef_numvals(e) == 0) { + upb_status_seterrf(s, "enum %s has no members (must have at least one)", + upb_enumdef_fullname(e)); + return false; + } + + return true; +} + +/* All submessage fields are lower than all other fields. + * Secondly, fields are increasing in order. */ +uint32_t field_rank(const upb_fielddef *f) { + uint32_t ret = upb_fielddef_number(f); + const uint32_t high_bit = 1 << 30; + assert(ret < high_bit); + if (!upb_fielddef_issubmsg(f)) + ret |= high_bit; + return ret; +} + +int cmp_fields(const void *p1, const void *p2) { + const upb_fielddef *f1 = *(upb_fielddef*const*)p1; + const upb_fielddef *f2 = *(upb_fielddef*const*)p2; + return field_rank(f1) - field_rank(f2); +} + +static bool assign_msg_indices(upb_msgdef *m, upb_status *s) { + /* Sort fields. upb internally relies on UPB_TYPE_MESSAGE fields having the + * lowest indexes, but we do not publicly guarantee this. */ + upb_msg_field_iter j; + int i; + uint32_t selector; + int n = upb_msgdef_numfields(m); + upb_fielddef **fields; + + if (n == 0) { + m->selector_count = UPB_STATIC_SELECTOR_COUNT; + m->submsg_field_count = 0; + return true; + } + + fields = upb_gmalloc(n * sizeof(*fields)); + if (!fields) { + upb_upberr_setoom(s); + return false; + } + + m->submsg_field_count = 0; + for(i = 0, upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j), i++) { + upb_fielddef *f = upb_msg_iter_field(&j); + assert(f->msg.def == m); + if (!upb_validate_field(f, s)) { + upb_gfree(fields); + return false; + } + if (upb_fielddef_issubmsg(f)) { + m->submsg_field_count++; + } + fields[i] = f; + } + + qsort(fields, n, sizeof(*fields), cmp_fields); + + selector = UPB_STATIC_SELECTOR_COUNT + m->submsg_field_count; + for (i = 0; i < n; i++) { + upb_fielddef *f = fields[i]; + f->index_ = i; + f->selector_base = selector + upb_handlers_selectorbaseoffset(f); + selector += upb_handlers_selectorcount(f); + } + m->selector_count = selector; + +#ifndef NDEBUG + { + /* Verify that all selectors for the message are distinct. */ +#define TRY(type) \ + if (upb_handlers_getselector(f, type, &sel)) upb_inttable_insert(&t, sel, v); + + upb_inttable t; + upb_value v; + upb_selector_t sel; + + upb_inttable_init(&t, UPB_CTYPE_BOOL); + v = upb_value_bool(true); + upb_inttable_insert(&t, UPB_STARTMSG_SELECTOR, v); + upb_inttable_insert(&t, UPB_ENDMSG_SELECTOR, v); + for(upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + upb_fielddef *f = upb_msg_iter_field(&j); + /* These calls will assert-fail in upb_table if the value already + * exists. */ + TRY(UPB_HANDLER_INT32); + TRY(UPB_HANDLER_INT64) + TRY(UPB_HANDLER_UINT32) + TRY(UPB_HANDLER_UINT64) + TRY(UPB_HANDLER_FLOAT) + TRY(UPB_HANDLER_DOUBLE) + TRY(UPB_HANDLER_BOOL) + TRY(UPB_HANDLER_STARTSTR) + TRY(UPB_HANDLER_STRING) + TRY(UPB_HANDLER_ENDSTR) + TRY(UPB_HANDLER_STARTSUBMSG) + TRY(UPB_HANDLER_ENDSUBMSG) + TRY(UPB_HANDLER_STARTSEQ) + TRY(UPB_HANDLER_ENDSEQ) + } + upb_inttable_uninit(&t); + } +#undef TRY +#endif + + upb_gfree(fields); + return true; +} + +bool _upb_def_validate(upb_def *const*defs, size_t n, upb_status *s) { + size_t i; + + /* First perform validation, in two passes so we can check that we have a + * transitive closure without needing to search. */ + for (i = 0; i < n; i++) { + upb_def *def = defs[i]; + if (upb_def_isfrozen(def)) { + /* Could relax this requirement if it's annoying. */ + upb_status_seterrmsg(s, "def is already frozen"); + goto err; + } else if (def->type == UPB_DEF_FIELD) { + upb_status_seterrmsg(s, "standalone fielddefs can not be frozen"); + goto err; + } else if (def->type == UPB_DEF_ENUM) { + if (!upb_validate_enumdef(upb_dyncast_enumdef(def), s)) { + goto err; + } + } else { + /* Set now to detect transitive closure in the second pass. */ + def->came_from_user = true; + } + } + + /* Second pass of validation. Also assign selector bases and indexes, and + * compact tables. */ + for (i = 0; i < n; i++) { + upb_def *def = defs[i]; + upb_msgdef *m = upb_dyncast_msgdef_mutable(def); + upb_enumdef *e = upb_dyncast_enumdef_mutable(def); + if (m) { + upb_inttable_compact(&m->itof); + if (!assign_msg_indices(m, s)) { + goto err; + } + } else if (e) { + upb_inttable_compact(&e->iton); + } + } + + return true; + +err: + for (i = 0; i < n; i++) { + upb_def *def = defs[i]; + def->came_from_user = false; + } + assert(!(s && upb_ok(s))); + return false; +} + +bool upb_def_freeze(upb_def *const* defs, size_t n, upb_status *s) { + /* Def graph contains FieldDefs between each MessageDef, so double the + * limit. */ + const size_t maxdepth = UPB_MAX_MESSAGE_DEPTH * 2; + + if (!_upb_def_validate(defs, n, s)) { + return false; + } + + + /* Validation all passed; freeze the objects. */ + return upb_refcounted_freeze((upb_refcounted *const*)defs, n, s, maxdepth); +} + + +/* upb_enumdef ****************************************************************/ + +static void upb_enumdef_free(upb_refcounted *r) { + upb_enumdef *e = (upb_enumdef*)r; + upb_inttable_iter i; + upb_inttable_begin(&i, &e->iton); + for( ; !upb_inttable_done(&i); upb_inttable_next(&i)) { + /* To clean up the upb_gstrdup() from upb_enumdef_addval(). */ + upb_gfree(upb_value_getcstr(upb_inttable_iter_value(&i))); + } + upb_strtable_uninit(&e->ntoi); + upb_inttable_uninit(&e->iton); + upb_def_uninit(upb_enumdef_upcast_mutable(e)); + upb_gfree(e); +} + +const struct upb_refcounted_vtbl upb_enumdef_vtbl = {NULL, &upb_enumdef_free}; + +upb_enumdef *upb_enumdef_new(const void *owner) { + upb_enumdef *e = upb_gmalloc(sizeof(*e)); + if (!e) return NULL; + + if (!upb_def_init(upb_enumdef_upcast_mutable(e), UPB_DEF_ENUM, + &upb_enumdef_vtbl, owner)) { + goto err2; + } + + if (!upb_strtable_init(&e->ntoi, UPB_CTYPE_INT32)) goto err2; + if (!upb_inttable_init(&e->iton, UPB_CTYPE_CSTR)) goto err1; + return e; + +err1: + upb_strtable_uninit(&e->ntoi); +err2: + upb_gfree(e); + return NULL; +} + +upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner) { + upb_enum_iter i; + upb_enumdef *new_e = upb_enumdef_new(owner); + if (!new_e) return NULL; + for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) { + bool success = upb_enumdef_addval( + new_e, upb_enum_iter_name(&i),upb_enum_iter_number(&i), NULL); + if (!success) { + upb_enumdef_unref(new_e, owner); + return NULL; + } + } + return new_e; +} + +bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status) { + upb_def *d = upb_enumdef_upcast_mutable(e); + return upb_def_freeze(&d, 1, status); +} + +const char *upb_enumdef_fullname(const upb_enumdef *e) { + return upb_def_fullname(upb_enumdef_upcast(e)); +} + +const char *upb_enumdef_name(const upb_enumdef *e) { + return upb_def_name(upb_enumdef_upcast(e)); +} + +bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname, + upb_status *s) { + return upb_def_setfullname(upb_enumdef_upcast_mutable(e), fullname, s); +} + +bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num, + upb_status *status) { + char *name2; + + if (!upb_isident(name, strlen(name), false, status)) { + return false; + } + + if (upb_enumdef_ntoiz(e, name, NULL)) { + upb_status_seterrf(status, "name '%s' is already defined", name); + return false; + } + + if (!upb_strtable_insert(&e->ntoi, name, upb_value_int32(num))) { + upb_status_seterrmsg(status, "out of memory"); + return false; + } + + if (!upb_inttable_lookup(&e->iton, num, NULL)) { + name2 = upb_gstrdup(name); + if (!name2 || !upb_inttable_insert(&e->iton, num, upb_value_cstr(name2))) { + upb_status_seterrmsg(status, "out of memory"); + upb_strtable_remove(&e->ntoi, name, NULL); + return false; + } + } + + if (upb_enumdef_numvals(e) == 1) { + bool ok = upb_enumdef_setdefault(e, num, NULL); + UPB_ASSERT_VAR(ok, ok); + } + + return true; +} + +int32_t upb_enumdef_default(const upb_enumdef *e) { + assert(upb_enumdef_iton(e, e->defaultval)); + return e->defaultval; +} + +bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s) { + assert(!upb_enumdef_isfrozen(e)); + if (!upb_enumdef_iton(e, val)) { + upb_status_seterrf(s, "number '%d' is not in the enum.", val); + return false; + } + e->defaultval = val; + return true; +} + +int upb_enumdef_numvals(const upb_enumdef *e) { + return upb_strtable_count(&e->ntoi); +} + +void upb_enum_begin(upb_enum_iter *i, const upb_enumdef *e) { + /* We iterate over the ntoi table, to account for duplicate numbers. */ + upb_strtable_begin(i, &e->ntoi); +} + +void upb_enum_next(upb_enum_iter *iter) { upb_strtable_next(iter); } +bool upb_enum_done(upb_enum_iter *iter) { return upb_strtable_done(iter); } + +bool upb_enumdef_ntoi(const upb_enumdef *def, const char *name, + size_t len, int32_t *num) { + upb_value v; + if (!upb_strtable_lookup2(&def->ntoi, name, len, &v)) { + return false; + } + if (num) *num = upb_value_getint32(v); + return true; +} + +const char *upb_enumdef_iton(const upb_enumdef *def, int32_t num) { + upb_value v; + return upb_inttable_lookup32(&def->iton, num, &v) ? + upb_value_getcstr(v) : NULL; +} + +const char *upb_enum_iter_name(upb_enum_iter *iter) { + return upb_strtable_iter_key(iter); +} + +int32_t upb_enum_iter_number(upb_enum_iter *iter) { + return upb_value_getint32(upb_strtable_iter_value(iter)); +} + + +/* upb_fielddef ***************************************************************/ + +static void upb_fielddef_init_default(upb_fielddef *f); + +static void upb_fielddef_uninit_default(upb_fielddef *f) { + if (f->type_is_set_ && f->default_is_string && f->defaultval.bytes) + freestr(f->defaultval.bytes); +} + +const char *upb_fielddef_fullname(const upb_fielddef *e) { + return upb_def_fullname(upb_fielddef_upcast(e)); +} + +static void visitfield(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_fielddef *f = (const upb_fielddef*)r; + if (upb_fielddef_containingtype(f)) { + visit(r, upb_msgdef_upcast2(upb_fielddef_containingtype(f)), closure); + } + if (upb_fielddef_containingoneof(f)) { + visit(r, upb_oneofdef_upcast(upb_fielddef_containingoneof(f)), closure); + } + if (upb_fielddef_subdef(f)) { + visit(r, upb_def_upcast(upb_fielddef_subdef(f)), closure); + } +} + +static void freefield(upb_refcounted *r) { + upb_fielddef *f = (upb_fielddef*)r; + upb_fielddef_uninit_default(f); + if (f->subdef_is_symbolic) + upb_gfree(f->sub.name); + upb_def_uninit(upb_fielddef_upcast_mutable(f)); + upb_gfree(f); +} + +static const char *enumdefaultstr(const upb_fielddef *f) { + const upb_enumdef *e; + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + e = upb_fielddef_enumsubdef(f); + if (f->default_is_string && f->defaultval.bytes) { + /* Default was explicitly set as a string. */ + str_t *s = f->defaultval.bytes; + return s->str; + } else if (e) { + if (!f->default_is_string) { + /* Default was explicitly set as an integer; look it up in enumdef. */ + const char *name = upb_enumdef_iton(e, f->defaultval.sint); + if (name) { + return name; + } + } else { + /* Default is completely unset; pull enumdef default. */ + if (upb_enumdef_numvals(e) > 0) { + const char *name = upb_enumdef_iton(e, upb_enumdef_default(e)); + assert(name); + return name; + } + } + } + return NULL; +} + +static bool enumdefaultint32(const upb_fielddef *f, int32_t *val) { + const upb_enumdef *e; + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + e = upb_fielddef_enumsubdef(f); + if (!f->default_is_string) { + /* Default was explicitly set as an integer. */ + *val = f->defaultval.sint; + return true; + } else if (e) { + if (f->defaultval.bytes) { + /* Default was explicitly set as a str; try to lookup corresponding int. */ + str_t *s = f->defaultval.bytes; + if (upb_enumdef_ntoiz(e, s->str, val)) { + return true; + } + } else { + /* Default is unset; try to pull in enumdef default. */ + if (upb_enumdef_numvals(e) > 0) { + *val = upb_enumdef_default(e); + return true; + } + } + } + return false; +} + +const struct upb_refcounted_vtbl upb_fielddef_vtbl = {visitfield, freefield}; + +upb_fielddef *upb_fielddef_new(const void *o) { + upb_fielddef *f = upb_gmalloc(sizeof(*f)); + if (!f) return NULL; + if (!upb_def_init(upb_fielddef_upcast_mutable(f), UPB_DEF_FIELD, + &upb_fielddef_vtbl, o)) { + upb_gfree(f); + return NULL; + } + f->msg.def = NULL; + f->sub.def = NULL; + f->oneof = NULL; + f->subdef_is_symbolic = false; + f->msg_is_symbolic = false; + f->label_ = UPB_LABEL_OPTIONAL; + f->type_ = UPB_TYPE_INT32; + f->number_ = 0; + f->type_is_set_ = false; + f->tagdelim = false; + f->is_extension_ = false; + f->lazy_ = false; + f->packed_ = true; + + /* For the moment we default this to UPB_INTFMT_VARIABLE, since it will work + * with all integer types and is in some since more "default" since the most + * normal-looking proto2 types int32/int64/uint32/uint64 use variable. + * + * Other options to consider: + * - there is no default; users must set this manually (like type). + * - default signed integers to UPB_INTFMT_ZIGZAG, since it's more likely to + * be an optimal default for signed integers. */ + f->intfmt = UPB_INTFMT_VARIABLE; + return f; +} + +upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner) { + const char *srcname; + upb_fielddef *newf = upb_fielddef_new(owner); + if (!newf) return NULL; + upb_fielddef_settype(newf, upb_fielddef_type(f)); + upb_fielddef_setlabel(newf, upb_fielddef_label(f)); + upb_fielddef_setnumber(newf, upb_fielddef_number(f), NULL); + upb_fielddef_setname(newf, upb_fielddef_name(f), NULL); + if (f->default_is_string && f->defaultval.bytes) { + str_t *s = f->defaultval.bytes; + upb_fielddef_setdefaultstr(newf, s->str, s->len, NULL); + } else { + newf->default_is_string = f->default_is_string; + newf->defaultval = f->defaultval; + } + + if (f->subdef_is_symbolic) { + srcname = f->sub.name; /* Might be NULL. */ + } else { + srcname = f->sub.def ? upb_def_fullname(f->sub.def) : NULL; + } + if (srcname) { + char *newname = upb_gmalloc(strlen(f->sub.def->fullname) + 2); + if (!newname) { + upb_fielddef_unref(newf, owner); + return NULL; + } + strcpy(newname, "."); + strcat(newname, f->sub.def->fullname); + upb_fielddef_setsubdefname(newf, newname, NULL); + upb_gfree(newname); + } + + return newf; +} + +bool upb_fielddef_typeisset(const upb_fielddef *f) { + return f->type_is_set_; +} + +upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f) { + assert(f->type_is_set_); + return f->type_; +} + +uint32_t upb_fielddef_index(const upb_fielddef *f) { + return f->index_; +} + +upb_label_t upb_fielddef_label(const upb_fielddef *f) { + return f->label_; +} + +upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f) { + return f->intfmt; +} + +bool upb_fielddef_istagdelim(const upb_fielddef *f) { + return f->tagdelim; +} + +uint32_t upb_fielddef_number(const upb_fielddef *f) { + return f->number_; +} + +bool upb_fielddef_isextension(const upb_fielddef *f) { + return f->is_extension_; +} + +bool upb_fielddef_lazy(const upb_fielddef *f) { + return f->lazy_; +} + +bool upb_fielddef_packed(const upb_fielddef *f) { + return f->packed_; +} + +const char *upb_fielddef_name(const upb_fielddef *f) { + return upb_def_fullname(upb_fielddef_upcast(f)); +} + +size_t upb_fielddef_getjsonname(const upb_fielddef *f, char *buf, size_t len) { + const char *name = upb_fielddef_name(f); + size_t src, dst = 0; + bool ucase_next = false; + +#define WRITE(byte) \ + ++dst; \ + if (dst < len) buf[dst - 1] = byte; \ + else if (dst == len) buf[dst - 1] = '\0' + + if (!name) { + WRITE('\0'); + return 0; + } + + /* Implement the transformation as described in the spec: + * 1. upper case all letters after an underscore. + * 2. remove all underscores. + */ + for (src = 0; name[src]; src++) { + if (name[src] == '_') { + ucase_next = true; + continue; + } + + if (ucase_next) { + WRITE(toupper(name[src])); + ucase_next = false; + } else { + WRITE(name[src]); + } + } + + WRITE('\0'); + return dst; + +#undef WRITE +} + +const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f) { + return f->msg_is_symbolic ? NULL : f->msg.def; +} + +const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f) { + return f->oneof; +} + +upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f) { + return (upb_msgdef*)upb_fielddef_containingtype(f); +} + +const char *upb_fielddef_containingtypename(upb_fielddef *f) { + return f->msg_is_symbolic ? f->msg.name : NULL; +} + +static void release_containingtype(upb_fielddef *f) { + if (f->msg_is_symbolic) upb_gfree(f->msg.name); +} + +bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name, + upb_status *s) { + char *name_copy; + assert(!upb_fielddef_isfrozen(f)); + if (upb_fielddef_containingtype(f)) { + upb_status_seterrmsg(s, "field has already been added to a message."); + return false; + } + /* TODO: validate name (upb_isident() doesn't quite work atm because this name + * may have a leading "."). */ + + name_copy = upb_gstrdup(name); + if (!name_copy) { + upb_upberr_setoom(s); + return false; + } + + release_containingtype(f); + f->msg.name = name_copy; + f->msg_is_symbolic = true; + return true; +} + +bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s) { + if (upb_fielddef_containingtype(f) || upb_fielddef_containingoneof(f)) { + upb_status_seterrmsg(s, "Already added to message or oneof"); + return false; + } + return upb_def_setfullname(upb_fielddef_upcast_mutable(f), name, s); +} + +static void chkdefaulttype(const upb_fielddef *f, upb_fieldtype_t type) { + UPB_UNUSED(f); + UPB_UNUSED(type); + assert(f->type_is_set_ && upb_fielddef_type(f) == type); +} + +int64_t upb_fielddef_defaultint64(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_INT64); + return f->defaultval.sint; +} + +int32_t upb_fielddef_defaultint32(const upb_fielddef *f) { + if (f->type_is_set_ && upb_fielddef_type(f) == UPB_TYPE_ENUM) { + int32_t val; + bool ok = enumdefaultint32(f, &val); + UPB_ASSERT_VAR(ok, ok); + return val; + } else { + chkdefaulttype(f, UPB_TYPE_INT32); + return f->defaultval.sint; + } +} + +uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_UINT64); + return f->defaultval.uint; +} + +uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_UINT32); + return f->defaultval.uint; +} + +bool upb_fielddef_defaultbool(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_BOOL); + return f->defaultval.uint; +} + +float upb_fielddef_defaultfloat(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_FLOAT); + return f->defaultval.flt; +} + +double upb_fielddef_defaultdouble(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_DOUBLE); + return f->defaultval.dbl; +} + +const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len) { + assert(f->type_is_set_); + assert(upb_fielddef_type(f) == UPB_TYPE_STRING || + upb_fielddef_type(f) == UPB_TYPE_BYTES || + upb_fielddef_type(f) == UPB_TYPE_ENUM); + + if (upb_fielddef_type(f) == UPB_TYPE_ENUM) { + const char *ret = enumdefaultstr(f); + assert(ret); + /* Enum defaults can't have embedded NULLs. */ + if (len) *len = strlen(ret); + return ret; + } + + if (f->default_is_string) { + str_t *str = f->defaultval.bytes; + if (len) *len = str->len; + return str->str; + } + + return NULL; +} + +static void upb_fielddef_init_default(upb_fielddef *f) { + f->default_is_string = false; + switch (upb_fielddef_type(f)) { + case UPB_TYPE_DOUBLE: f->defaultval.dbl = 0; break; + case UPB_TYPE_FLOAT: f->defaultval.flt = 0; break; + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: f->defaultval.sint = 0; break; + case UPB_TYPE_UINT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_BOOL: f->defaultval.uint = 0; break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + f->defaultval.bytes = newstr("", 0); + f->default_is_string = true; + break; + case UPB_TYPE_MESSAGE: break; + case UPB_TYPE_ENUM: + /* This is our special sentinel that indicates "not set" for an enum. */ + f->default_is_string = true; + f->defaultval.bytes = NULL; + break; + } +} + +const upb_def *upb_fielddef_subdef(const upb_fielddef *f) { + return f->subdef_is_symbolic ? NULL : f->sub.def; +} + +const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f) { + const upb_def *def = upb_fielddef_subdef(f); + return def ? upb_dyncast_msgdef(def) : NULL; +} + +const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f) { + const upb_def *def = upb_fielddef_subdef(f); + return def ? upb_dyncast_enumdef(def) : NULL; +} + +upb_def *upb_fielddef_subdef_mutable(upb_fielddef *f) { + return (upb_def*)upb_fielddef_subdef(f); +} + +const char *upb_fielddef_subdefname(const upb_fielddef *f) { + if (f->subdef_is_symbolic) { + return f->sub.name; + } else if (f->sub.def) { + return upb_def_fullname(f->sub.def); + } else { + return NULL; + } +} + +bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s) { + if (upb_fielddef_containingtype(f)) { + upb_status_seterrmsg( + s, "cannot change field number after adding to a message"); + return false; + } + if (number == 0 || number > UPB_MAX_FIELDNUMBER) { + upb_status_seterrf(s, "invalid field number (%u)", number); + return false; + } + f->number_ = number; + return true; +} + +void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checktype(type)); + upb_fielddef_uninit_default(f); + f->type_ = type; + f->type_is_set_ = true; + upb_fielddef_init_default(f); +} + +void upb_fielddef_setdescriptortype(upb_fielddef *f, int type) { + assert(!upb_fielddef_isfrozen(f)); + switch (type) { + case UPB_DESCRIPTOR_TYPE_DOUBLE: + upb_fielddef_settype(f, UPB_TYPE_DOUBLE); + break; + case UPB_DESCRIPTOR_TYPE_FLOAT: + upb_fielddef_settype(f, UPB_TYPE_FLOAT); + break; + case UPB_DESCRIPTOR_TYPE_INT64: + case UPB_DESCRIPTOR_TYPE_SFIXED64: + case UPB_DESCRIPTOR_TYPE_SINT64: + upb_fielddef_settype(f, UPB_TYPE_INT64); + break; + case UPB_DESCRIPTOR_TYPE_UINT64: + case UPB_DESCRIPTOR_TYPE_FIXED64: + upb_fielddef_settype(f, UPB_TYPE_UINT64); + break; + case UPB_DESCRIPTOR_TYPE_INT32: + case UPB_DESCRIPTOR_TYPE_SFIXED32: + case UPB_DESCRIPTOR_TYPE_SINT32: + upb_fielddef_settype(f, UPB_TYPE_INT32); + break; + case UPB_DESCRIPTOR_TYPE_UINT32: + case UPB_DESCRIPTOR_TYPE_FIXED32: + upb_fielddef_settype(f, UPB_TYPE_UINT32); + break; + case UPB_DESCRIPTOR_TYPE_BOOL: + upb_fielddef_settype(f, UPB_TYPE_BOOL); + break; + case UPB_DESCRIPTOR_TYPE_STRING: + upb_fielddef_settype(f, UPB_TYPE_STRING); + break; + case UPB_DESCRIPTOR_TYPE_BYTES: + upb_fielddef_settype(f, UPB_TYPE_BYTES); + break; + case UPB_DESCRIPTOR_TYPE_GROUP: + case UPB_DESCRIPTOR_TYPE_MESSAGE: + upb_fielddef_settype(f, UPB_TYPE_MESSAGE); + break; + case UPB_DESCRIPTOR_TYPE_ENUM: + upb_fielddef_settype(f, UPB_TYPE_ENUM); + break; + default: assert(false); + } + + if (type == UPB_DESCRIPTOR_TYPE_FIXED64 || + type == UPB_DESCRIPTOR_TYPE_FIXED32 || + type == UPB_DESCRIPTOR_TYPE_SFIXED64 || + type == UPB_DESCRIPTOR_TYPE_SFIXED32) { + upb_fielddef_setintfmt(f, UPB_INTFMT_FIXED); + } else if (type == UPB_DESCRIPTOR_TYPE_SINT64 || + type == UPB_DESCRIPTOR_TYPE_SINT32) { + upb_fielddef_setintfmt(f, UPB_INTFMT_ZIGZAG); + } else { + upb_fielddef_setintfmt(f, UPB_INTFMT_VARIABLE); + } + + upb_fielddef_settagdelim(f, type == UPB_DESCRIPTOR_TYPE_GROUP); +} + +upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_FLOAT: return UPB_DESCRIPTOR_TYPE_FLOAT; + case UPB_TYPE_DOUBLE: return UPB_DESCRIPTOR_TYPE_DOUBLE; + case UPB_TYPE_BOOL: return UPB_DESCRIPTOR_TYPE_BOOL; + case UPB_TYPE_STRING: return UPB_DESCRIPTOR_TYPE_STRING; + case UPB_TYPE_BYTES: return UPB_DESCRIPTOR_TYPE_BYTES; + case UPB_TYPE_ENUM: return UPB_DESCRIPTOR_TYPE_ENUM; + case UPB_TYPE_INT32: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT32; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_SFIXED32; + case UPB_INTFMT_ZIGZAG: return UPB_DESCRIPTOR_TYPE_SINT32; + } + case UPB_TYPE_INT64: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT64; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_SFIXED64; + case UPB_INTFMT_ZIGZAG: return UPB_DESCRIPTOR_TYPE_SINT64; + } + case UPB_TYPE_UINT32: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT32; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_FIXED32; + case UPB_INTFMT_ZIGZAG: return -1; + } + case UPB_TYPE_UINT64: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT64; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_FIXED64; + case UPB_INTFMT_ZIGZAG: return -1; + } + case UPB_TYPE_MESSAGE: + return upb_fielddef_istagdelim(f) ? + UPB_DESCRIPTOR_TYPE_GROUP : UPB_DESCRIPTOR_TYPE_MESSAGE; + } + return 0; +} + +void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension) { + assert(!upb_fielddef_isfrozen(f)); + f->is_extension_ = is_extension; +} + +void upb_fielddef_setlazy(upb_fielddef *f, bool lazy) { + assert(!upb_fielddef_isfrozen(f)); + f->lazy_ = lazy; +} + +void upb_fielddef_setpacked(upb_fielddef *f, bool packed) { + assert(!upb_fielddef_isfrozen(f)); + f->packed_ = packed; +} + +void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checklabel(label)); + f->label_ = label; +} + +void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checkintfmt(fmt)); + f->intfmt = fmt; +} + +void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim) { + assert(!upb_fielddef_isfrozen(f)); + f->tagdelim = tag_delim; + f->tagdelim = tag_delim; +} + +static bool checksetdefault(upb_fielddef *f, upb_fieldtype_t type) { + if (!f->type_is_set_ || upb_fielddef_isfrozen(f) || + upb_fielddef_type(f) != type) { + assert(false); + return false; + } + if (f->default_is_string) { + str_t *s = f->defaultval.bytes; + assert(s || type == UPB_TYPE_ENUM); + if (s) freestr(s); + } + f->default_is_string = false; + return true; +} + +void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t value) { + if (checksetdefault(f, UPB_TYPE_INT64)) + f->defaultval.sint = value; +} + +void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t value) { + if ((upb_fielddef_type(f) == UPB_TYPE_ENUM && + checksetdefault(f, UPB_TYPE_ENUM)) || + checksetdefault(f, UPB_TYPE_INT32)) { + f->defaultval.sint = value; + } +} + +void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t value) { + if (checksetdefault(f, UPB_TYPE_UINT64)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t value) { + if (checksetdefault(f, UPB_TYPE_UINT32)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultbool(upb_fielddef *f, bool value) { + if (checksetdefault(f, UPB_TYPE_BOOL)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultfloat(upb_fielddef *f, float value) { + if (checksetdefault(f, UPB_TYPE_FLOAT)) + f->defaultval.flt = value; +} + +void upb_fielddef_setdefaultdouble(upb_fielddef *f, double value) { + if (checksetdefault(f, UPB_TYPE_DOUBLE)) + f->defaultval.dbl = value; +} + +bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len, + upb_status *s) { + str_t *str2; + assert(upb_fielddef_isstring(f) || f->type_ == UPB_TYPE_ENUM); + if (f->type_ == UPB_TYPE_ENUM && !upb_isident(str, len, false, s)) + return false; + + if (f->default_is_string) { + str_t *s = f->defaultval.bytes; + assert(s || f->type_ == UPB_TYPE_ENUM); + if (s) freestr(s); + } else { + assert(f->type_ == UPB_TYPE_ENUM); + } + + str2 = newstr(str, len); + f->defaultval.bytes = str2; + f->default_is_string = true; + return true; +} + +void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str, + upb_status *s) { + assert(f->type_is_set_); + upb_fielddef_setdefaultstr(f, str, str ? strlen(str) : 0, s); +} + +bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f) { + int32_t val; + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + return enumdefaultint32(f, &val); +} + +bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f) { + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + return enumdefaultstr(f) != NULL; +} + +static bool upb_subdef_typecheck(upb_fielddef *f, const upb_def *subdef, + upb_status *s) { + if (f->type_ == UPB_TYPE_MESSAGE) { + if (upb_dyncast_msgdef(subdef)) return true; + upb_status_seterrmsg(s, "invalid subdef type for this submessage field"); + return false; + } else if (f->type_ == UPB_TYPE_ENUM) { + if (upb_dyncast_enumdef(subdef)) return true; + upb_status_seterrmsg(s, "invalid subdef type for this enum field"); + return false; + } else { + upb_status_seterrmsg(s, "only message and enum fields can have a subdef"); + return false; + } +} + +static void release_subdef(upb_fielddef *f) { + if (f->subdef_is_symbolic) { + upb_gfree(f->sub.name); + } else if (f->sub.def) { + upb_unref2(f->sub.def, f); + } +} + +bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef, + upb_status *s) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_hassubdef(f)); + if (subdef && !upb_subdef_typecheck(f, subdef, s)) return false; + release_subdef(f); + f->sub.def = subdef; + f->subdef_is_symbolic = false; + if (f->sub.def) upb_ref2(f->sub.def, f); + return true; +} + +bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef, + upb_status *s) { + return upb_fielddef_setsubdef(f, upb_msgdef_upcast(subdef), s); +} + +bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef, + upb_status *s) { + return upb_fielddef_setsubdef(f, upb_enumdef_upcast(subdef), s); +} + +bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name, + upb_status *s) { + char *name_copy; + assert(!upb_fielddef_isfrozen(f)); + if (!upb_fielddef_hassubdef(f)) { + upb_status_seterrmsg(s, "field type does not accept a subdef"); + return false; + } + + name_copy = upb_gstrdup(name); + if (!name_copy) { + upb_upberr_setoom(s); + return false; + } + + /* TODO: validate name (upb_isident() doesn't quite work atm because this name + * may have a leading "."). */ + release_subdef(f); + f->sub.name = name_copy; + f->subdef_is_symbolic = true; + return true; +} + +bool upb_fielddef_issubmsg(const upb_fielddef *f) { + return upb_fielddef_type(f) == UPB_TYPE_MESSAGE; +} + +bool upb_fielddef_isstring(const upb_fielddef *f) { + return upb_fielddef_type(f) == UPB_TYPE_STRING || + upb_fielddef_type(f) == UPB_TYPE_BYTES; +} + +bool upb_fielddef_isseq(const upb_fielddef *f) { + return upb_fielddef_label(f) == UPB_LABEL_REPEATED; +} + +bool upb_fielddef_isprimitive(const upb_fielddef *f) { + return !upb_fielddef_isstring(f) && !upb_fielddef_issubmsg(f); +} + +bool upb_fielddef_ismap(const upb_fielddef *f) { + return upb_fielddef_isseq(f) && upb_fielddef_issubmsg(f) && + upb_msgdef_mapentry(upb_fielddef_msgsubdef(f)); +} + +bool upb_fielddef_haspresence(const upb_fielddef *f) { + if (upb_fielddef_isseq(f)) return false; + if (upb_fielddef_issubmsg(f)) return true; + + /* Primitive field: return true unless there is a message that specifies + * presence should not exist. */ + if (f->msg_is_symbolic || !f->msg.def) return true; + return f->msg.def->syntax == UPB_SYNTAX_PROTO2; +} + +bool upb_fielddef_hassubdef(const upb_fielddef *f) { + return upb_fielddef_issubmsg(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM; +} + +static bool between(int32_t x, int32_t low, int32_t high) { + return x >= low && x <= high; +} + +bool upb_fielddef_checklabel(int32_t label) { return between(label, 1, 3); } +bool upb_fielddef_checktype(int32_t type) { return between(type, 1, 11); } +bool upb_fielddef_checkintfmt(int32_t fmt) { return between(fmt, 1, 3); } + +bool upb_fielddef_checkdescriptortype(int32_t type) { + return between(type, 1, 18); +} + +/* upb_msgdef *****************************************************************/ + +static void visitmsg(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + upb_msg_oneof_iter o; + const upb_msgdef *m = (const upb_msgdef*)r; + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + visit(r, upb_fielddef_upcast2(f), closure); + } + for(upb_msg_oneof_begin(&o, m); + !upb_msg_oneof_done(&o); + upb_msg_oneof_next(&o)) { + upb_oneofdef *f = upb_msg_iter_oneof(&o); + visit(r, upb_oneofdef_upcast(f), closure); + } +} + +static void freemsg(upb_refcounted *r) { + upb_msgdef *m = (upb_msgdef*)r; + upb_strtable_uninit(&m->ntof); + upb_inttable_uninit(&m->itof); + upb_def_uninit(upb_msgdef_upcast_mutable(m)); + upb_gfree(m); +} + +const struct upb_refcounted_vtbl upb_msgdef_vtbl = {visitmsg, freemsg}; + +upb_msgdef *upb_msgdef_new(const void *owner) { + upb_msgdef *m = upb_gmalloc(sizeof(*m)); + if (!m) return NULL; + + if (!upb_def_init(upb_msgdef_upcast_mutable(m), UPB_DEF_MSG, &upb_msgdef_vtbl, + owner)) { + goto err2; + } + + if (!upb_inttable_init(&m->itof, UPB_CTYPE_PTR)) goto err2; + if (!upb_strtable_init(&m->ntof, UPB_CTYPE_PTR)) goto err1; + m->map_entry = false; + m->syntax = UPB_SYNTAX_PROTO2; + return m; + +err1: + upb_inttable_uninit(&m->itof); +err2: + upb_gfree(m); + return NULL; +} + +upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner) { + bool ok; + upb_msg_field_iter i; + upb_msg_oneof_iter o; + + upb_msgdef *newm = upb_msgdef_new(owner); + if (!newm) return NULL; + ok = upb_def_setfullname(upb_msgdef_upcast_mutable(newm), + upb_def_fullname(upb_msgdef_upcast(m)), + NULL); + newm->map_entry = m->map_entry; + newm->syntax = m->syntax; + UPB_ASSERT_VAR(ok, ok); + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_fielddef_dup(upb_msg_iter_field(&i), &f); + /* Fields in oneofs are dup'd below. */ + if (upb_fielddef_containingoneof(f)) continue; + if (!f || !upb_msgdef_addfield(newm, f, &f, NULL)) { + upb_msgdef_unref(newm, owner); + return NULL; + } + } + for(upb_msg_oneof_begin(&o, m); + !upb_msg_oneof_done(&o); + upb_msg_oneof_next(&o)) { + upb_oneofdef *f = upb_oneofdef_dup(upb_msg_iter_oneof(&o), &f); + if (!f || !upb_msgdef_addoneof(newm, f, &f, NULL)) { + upb_msgdef_unref(newm, owner); + return NULL; + } + } + return newm; +} + +bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status) { + upb_def *d = upb_msgdef_upcast_mutable(m); + return upb_def_freeze(&d, 1, status); +} + +const char *upb_msgdef_fullname(const upb_msgdef *m) { + return upb_def_fullname(upb_msgdef_upcast(m)); +} + +const char *upb_msgdef_name(const upb_msgdef *m) { + return upb_def_name(upb_msgdef_upcast(m)); +} + +bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname, + upb_status *s) { + return upb_def_setfullname(upb_msgdef_upcast_mutable(m), fullname, s); +} + +bool upb_msgdef_setsyntax(upb_msgdef *m, upb_syntax_t syntax) { + if (syntax != UPB_SYNTAX_PROTO2 && syntax != UPB_SYNTAX_PROTO3) { + return false; + } + + m->syntax = syntax; + return true; +} + +upb_syntax_t upb_msgdef_syntax(const upb_msgdef *m) { + return m->syntax; +} + +/* Helper: check that the field |f| is safe to add to msgdef |m|. Set an error + * on status |s| and return false if not. */ +static bool check_field_add(const upb_msgdef *m, const upb_fielddef *f, + upb_status *s) { + if (upb_fielddef_containingtype(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a message"); + return false; + } else if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "field name or number were not set"); + return false; + } else if (upb_msgdef_itof(m, upb_fielddef_number(f))) { + upb_status_seterrmsg(s, "duplicate field number"); + return false; + } else if (upb_strtable_lookup(&m->ntof, upb_fielddef_name(f), NULL)) { + upb_status_seterrmsg(s, "name conflicts with existing field or oneof"); + return false; + } + return true; +} + +static void add_field(upb_msgdef *m, upb_fielddef *f, const void *ref_donor) { + release_containingtype(f); + f->msg.def = m; + f->msg_is_symbolic = false; + upb_inttable_insert(&m->itof, upb_fielddef_number(f), upb_value_ptr(f)); + upb_strtable_insert(&m->ntof, upb_fielddef_name(f), upb_value_ptr(f)); + upb_ref2(f, m); + upb_ref2(m, f); + if (ref_donor) upb_fielddef_unref(f, ref_donor); +} + +bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor, + upb_status *s) { + /* TODO: extensions need to have a separate namespace, because proto2 allows a + * top-level extension (ie. one not in any package) to have the same name as a + * field from the message. + * + * This also implies that there needs to be a separate lookup-by-name method + * for extensions. It seems desirable for iteration to return both extensions + * and non-extensions though. + * + * We also need to validate that the field number is in an extension range iff + * it is an extension. + * + * This method is idempotent. Check if |f| is already part of this msgdef and + * return immediately if so. */ + if (upb_fielddef_containingtype(f) == m) { + return true; + } + + /* Check constraints for all fields before performing any action. */ + if (!check_field_add(m, f, s)) { + return false; + } else if (upb_fielddef_containingoneof(f) != NULL) { + /* Fields in a oneof can only be added by adding the oneof to the msgdef. */ + upb_status_seterrmsg(s, "fielddef is part of a oneof"); + return false; + } + + /* Constraint checks ok, perform the action. */ + add_field(m, f, ref_donor); + return true; +} + +bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor, + upb_status *s) { + upb_oneof_iter it; + + /* Check various conditions that would prevent this oneof from being added. */ + if (upb_oneofdef_containingtype(o)) { + upb_status_seterrmsg(s, "oneofdef already belongs to a message"); + return false; + } else if (upb_oneofdef_name(o) == NULL) { + upb_status_seterrmsg(s, "oneofdef name was not set"); + return false; + } else if (upb_strtable_lookup(&m->ntof, upb_oneofdef_name(o), NULL)) { + upb_status_seterrmsg(s, "name conflicts with existing field or oneof"); + return false; + } + + /* Check that all of the oneof's fields do not conflict with names or numbers + * of fields already in the message. */ + for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) { + const upb_fielddef *f = upb_oneof_iter_field(&it); + if (!check_field_add(m, f, s)) { + return false; + } + } + + /* Everything checks out -- commit now. */ + + /* Add oneof itself first. */ + o->parent = m; + upb_strtable_insert(&m->ntof, upb_oneofdef_name(o), upb_value_ptr(o)); + upb_ref2(o, m); + upb_ref2(m, o); + + /* Add each field of the oneof directly to the msgdef. */ + for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) { + upb_fielddef *f = upb_oneof_iter_field(&it); + add_field(m, f, NULL); + } + + if (ref_donor) upb_oneofdef_unref(o, ref_donor); + + return true; +} + +const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i) { + upb_value val; + return upb_inttable_lookup32(&m->itof, i, &val) ? + upb_value_getptr(val) : NULL; +} + +const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name, + size_t len) { + upb_value val; + + if (!upb_strtable_lookup2(&m->ntof, name, len, &val)) { + return NULL; + } + + return upb_trygetfield(upb_value_getptr(val)); +} + +const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name, + size_t len) { + upb_value val; + + if (!upb_strtable_lookup2(&m->ntof, name, len, &val)) { + return NULL; + } + + return upb_trygetoneof(upb_value_getptr(val)); +} + +bool upb_msgdef_lookupname(const upb_msgdef *m, const char *name, size_t len, + const upb_fielddef **f, const upb_oneofdef **o) { + upb_value val; + + if (!upb_strtable_lookup2(&m->ntof, name, len, &val)) { + return false; + } + + *o = upb_trygetoneof(upb_value_getptr(val)); + *f = upb_trygetfield(upb_value_getptr(val)); + assert((*o != NULL) ^ (*f != NULL)); /* Exactly one of the two should be set. */ + return true; +} + +int upb_msgdef_numfields(const upb_msgdef *m) { + /* The number table contains only fields. */ + return upb_inttable_count(&m->itof); +} + +int upb_msgdef_numoneofs(const upb_msgdef *m) { + /* The name table includes oneofs, and the number table does not. */ + return upb_strtable_count(&m->ntof) - upb_inttable_count(&m->itof); +} + +void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry) { + assert(!upb_msgdef_isfrozen(m)); + m->map_entry = map_entry; +} + +bool upb_msgdef_mapentry(const upb_msgdef *m) { + return m->map_entry; +} + +void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m) { + upb_inttable_begin(iter, &m->itof); +} + +void upb_msg_field_next(upb_msg_field_iter *iter) { upb_inttable_next(iter); } + +bool upb_msg_field_done(const upb_msg_field_iter *iter) { + return upb_inttable_done(iter); +} + +upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter) { + return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter)); +} + +void upb_msg_field_iter_setdone(upb_msg_field_iter *iter) { + upb_inttable_iter_setdone(iter); +} + +void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m) { + upb_strtable_begin(iter, &m->ntof); + /* We need to skip past any initial fields. */ + while (!upb_strtable_done(iter) && + !upb_isoneof(upb_value_getptr(upb_strtable_iter_value(iter)))) { + upb_strtable_next(iter); + } +} + +void upb_msg_oneof_next(upb_msg_oneof_iter *iter) { + /* We need to skip past fields to return only oneofs. */ + do { + upb_strtable_next(iter); + } while (!upb_strtable_done(iter) && + !upb_isoneof(upb_value_getptr(upb_strtable_iter_value(iter)))); +} + +bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter) { + return upb_strtable_done(iter); +} + +upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter) { + return (upb_oneofdef*)upb_value_getptr(upb_strtable_iter_value(iter)); +} + +void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter) { + upb_strtable_iter_setdone(iter); +} + +/* upb_oneofdef ***************************************************************/ + +static void visitoneof(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_oneofdef *o = (const upb_oneofdef*)r; + upb_oneof_iter i; + for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) { + const upb_fielddef *f = upb_oneof_iter_field(&i); + visit(r, upb_fielddef_upcast2(f), closure); + } + if (o->parent) { + visit(r, upb_msgdef_upcast2(o->parent), closure); + } +} + +static void freeoneof(upb_refcounted *r) { + upb_oneofdef *o = (upb_oneofdef*)r; + upb_strtable_uninit(&o->ntof); + upb_inttable_uninit(&o->itof); + upb_gfree((void*)o->name); + upb_gfree(o); +} + +const struct upb_refcounted_vtbl upb_oneofdef_vtbl = {visitoneof, freeoneof}; + +upb_oneofdef *upb_oneofdef_new(const void *owner) { + upb_oneofdef *o = upb_gmalloc(sizeof(*o)); + + if (!o) { + return NULL; + } + + o->parent = NULL; + o->name = NULL; + + if (!upb_refcounted_init(upb_oneofdef_upcast_mutable(o), &upb_oneofdef_vtbl, + owner)) { + goto err2; + } + + if (!upb_inttable_init(&o->itof, UPB_CTYPE_PTR)) goto err2; + if (!upb_strtable_init(&o->ntof, UPB_CTYPE_PTR)) goto err1; + + return o; + +err1: + upb_inttable_uninit(&o->itof); +err2: + upb_gfree(o); + return NULL; +} + +upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner) { + bool ok; + upb_oneof_iter i; + upb_oneofdef *newo = upb_oneofdef_new(owner); + if (!newo) return NULL; + ok = upb_oneofdef_setname(newo, upb_oneofdef_name(o), NULL); + UPB_ASSERT_VAR(ok, ok); + for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) { + upb_fielddef *f = upb_fielddef_dup(upb_oneof_iter_field(&i), &f); + if (!f || !upb_oneofdef_addfield(newo, f, &f, NULL)) { + upb_oneofdef_unref(newo, owner); + return NULL; + } + } + return newo; +} + +const char *upb_oneofdef_name(const upb_oneofdef *o) { return o->name; } + +bool upb_oneofdef_setname(upb_oneofdef *o, const char *name, upb_status *s) { + assert(!upb_oneofdef_isfrozen(o)); + if (upb_oneofdef_containingtype(o)) { + upb_status_seterrmsg(s, "oneof already added to a message"); + return false; + } + + if (!upb_isident(name, strlen(name), true, s)) { + return false; + } + + name = upb_gstrdup(name); + if (!name) { + upb_status_seterrmsg(s, "One of memory"); + return false; + } + + upb_gfree((void*)o->name); + o->name = name; + return true; +} + +const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o) { + return o->parent; +} + +int upb_oneofdef_numfields(const upb_oneofdef *o) { + return upb_strtable_count(&o->ntof); +} + +bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f, + const void *ref_donor, + upb_status *s) { + assert(!upb_oneofdef_isfrozen(o)); + assert(!o->parent || !upb_msgdef_isfrozen(o->parent)); + + /* This method is idempotent. Check if |f| is already part of this oneofdef + * and return immediately if so. */ + if (upb_fielddef_containingoneof(f) == o) { + return true; + } + + /* The field must have an OPTIONAL label. */ + if (upb_fielddef_label(f) != UPB_LABEL_OPTIONAL) { + upb_status_seterrmsg(s, "fields in oneof must have OPTIONAL label"); + return false; + } + + /* Check that no field with this name or number exists already in the oneof. + * Also check that the field is not already part of a oneof. */ + if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "field name or number were not set"); + return false; + } else if (upb_oneofdef_itof(o, upb_fielddef_number(f)) || + upb_oneofdef_ntofz(o, upb_fielddef_name(f))) { + upb_status_seterrmsg(s, "duplicate field name or number"); + return false; + } else if (upb_fielddef_containingoneof(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a oneof"); + return false; + } + + /* We allow adding a field to the oneof either if the field is not part of a + * msgdef, or if it is and we are also part of the same msgdef. */ + if (o->parent == NULL) { + /* If we're not in a msgdef, the field cannot be either. Otherwise we would + * need to magically add this oneof to a msgdef to remain consistent, which + * is surprising behavior. */ + if (upb_fielddef_containingtype(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a message, but " + "oneof does not"); + return false; + } + } else { + /* If we're in a msgdef, the user can add fields that either aren't in any + * msgdef (in which case they're added to our msgdef) or already a part of + * our msgdef. */ + if (upb_fielddef_containingtype(f) != NULL && + upb_fielddef_containingtype(f) != o->parent) { + upb_status_seterrmsg(s, "fielddef belongs to a different message " + "than oneof"); + return false; + } + } + + /* Commit phase. First add the field to our parent msgdef, if any, because + * that may fail; then add the field to our own tables. */ + + if (o->parent != NULL && upb_fielddef_containingtype(f) == NULL) { + if (!upb_msgdef_addfield((upb_msgdef*)o->parent, f, NULL, s)) { + return false; + } + } + + release_containingtype(f); + f->oneof = o; + upb_inttable_insert(&o->itof, upb_fielddef_number(f), upb_value_ptr(f)); + upb_strtable_insert(&o->ntof, upb_fielddef_name(f), upb_value_ptr(f)); + upb_ref2(f, o); + upb_ref2(o, f); + if (ref_donor) upb_fielddef_unref(f, ref_donor); + + return true; +} + +const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o, + const char *name, size_t length) { + upb_value val; + return upb_strtable_lookup2(&o->ntof, name, length, &val) ? + upb_value_getptr(val) : NULL; +} + +const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num) { + upb_value val; + return upb_inttable_lookup32(&o->itof, num, &val) ? + upb_value_getptr(val) : NULL; +} + +void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o) { + upb_inttable_begin(iter, &o->itof); +} + +void upb_oneof_next(upb_oneof_iter *iter) { + upb_inttable_next(iter); +} + +bool upb_oneof_done(upb_oneof_iter *iter) { + return upb_inttable_done(iter); +} + +upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter) { + return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter)); +} + +void upb_oneof_iter_setdone(upb_oneof_iter *iter) { + upb_inttable_iter_setdone(iter); +} + +/* upb_filedef ****************************************************************/ + +static void visitfiledef(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_filedef *f = (const upb_filedef*)r; + size_t i; + + for(i = 0; i < upb_filedef_defcount(f); i++) { + visit(r, upb_def_upcast(upb_filedef_def(f, i)), closure); + } +} + +static void freefiledef(upb_refcounted *r) { + upb_filedef *f = (upb_filedef*)r; + size_t i; + + for(i = 0; i < upb_filedef_depcount(f); i++) { + upb_filedef_unref(upb_filedef_dep(f, i), f); + } + + upb_inttable_uninit(&f->defs); + upb_inttable_uninit(&f->deps); + upb_gfree((void*)f->name); + upb_gfree((void*)f->package); + upb_gfree(f); +} + +const struct upb_refcounted_vtbl upb_filedef_vtbl = {visitfiledef, freefiledef}; + +upb_filedef *upb_filedef_new(const void *owner) { + upb_filedef *f = upb_gmalloc(sizeof(*f)); + + if (!f) { + return NULL; + } + + f->package = NULL; + f->name = NULL; + f->syntax = UPB_SYNTAX_PROTO2; + + if (!upb_refcounted_init(upb_filedef_upcast_mutable(f), &upb_filedef_vtbl, + owner)) { + goto err; + } + + if (!upb_inttable_init(&f->defs, UPB_CTYPE_CONSTPTR)) { + goto err; + } + + if (!upb_inttable_init(&f->deps, UPB_CTYPE_CONSTPTR)) { + goto err2; + } + + return f; + + +err2: + upb_inttable_uninit(&f->defs); + +err: + upb_gfree(f); + return NULL; +} + +const char *upb_filedef_name(const upb_filedef *f) { + return f->name; +} + +const char *upb_filedef_package(const upb_filedef *f) { + return f->package; +} + +upb_syntax_t upb_filedef_syntax(const upb_filedef *f) { + return f->syntax; +} + +size_t upb_filedef_defcount(const upb_filedef *f) { + return upb_inttable_count(&f->defs); +} + +size_t upb_filedef_depcount(const upb_filedef *f) { + return upb_inttable_count(&f->deps); +} + +const upb_def *upb_filedef_def(const upb_filedef *f, size_t i) { + upb_value v; + + if (upb_inttable_lookup32(&f->defs, i, &v)) { + return upb_value_getconstptr(v); + } else { + return NULL; + } +} + +const upb_filedef *upb_filedef_dep(const upb_filedef *f, size_t i) { + upb_value v; + + if (upb_inttable_lookup32(&f->deps, i, &v)) { + return upb_value_getconstptr(v); + } else { + return NULL; + } +} + +bool upb_filedef_setname(upb_filedef *f, const char *name, upb_status *s) { + name = upb_gstrdup(name); + if (!name) { + upb_upberr_setoom(s); + return false; + } + upb_gfree((void*)f->name); + f->name = name; + return true; +} + +bool upb_filedef_setpackage(upb_filedef *f, const char *package, + upb_status *s) { + if (!upb_isident(package, strlen(package), true, s)) return false; + package = upb_gstrdup(package); + if (!package) { + upb_upberr_setoom(s); + return false; + } + upb_gfree((void*)f->package); + f->package = package; + return true; +} + +bool upb_filedef_setsyntax(upb_filedef *f, upb_syntax_t syntax, + upb_status *s) { + UPB_UNUSED(s); + if (syntax != UPB_SYNTAX_PROTO2 && + syntax != UPB_SYNTAX_PROTO3) { + upb_status_seterrmsg(s, "Unknown syntax value."); + return false; + } + f->syntax = syntax; + + { + /* Set all messages in this file to match. */ + size_t i; + for (i = 0; i < upb_filedef_defcount(f); i++) { + /* Casting const away is safe since all defs in mutable filedef must + * also be mutable. */ + upb_def *def = (upb_def*)upb_filedef_def(f, i); + + upb_msgdef *m = upb_dyncast_msgdef_mutable(def); + if (m) { + m->syntax = syntax; + } + } + } + + return true; +} + +bool upb_filedef_adddef(upb_filedef *f, upb_def *def, const void *ref_donor, + upb_status *s) { + if (def->file) { + upb_status_seterrmsg(s, "Def is already part of another filedef."); + return false; + } + + if (upb_inttable_push(&f->defs, upb_value_constptr(def))) { + def->file = f; + upb_ref2(def, f); + if (ref_donor) upb_def_unref(def, ref_donor); + if (def->type == UPB_DEF_MSG) { + upb_downcast_msgdef_mutable(def)->syntax = f->syntax; + } + return true; + } else { + upb_upberr_setoom(s); + return false; + } +} + +bool upb_filedef_adddep(upb_filedef *f, const upb_filedef *dep) { + if (upb_inttable_push(&f->deps, upb_value_constptr(dep))) { + /* Regular ref instead of ref2 because files can't form cycles. */ + upb_filedef_ref(dep, f); + return true; + } else { + return false; + } +} +/* +** TODO(haberman): it's unclear whether a lot of the consistency checks should +** assert() or return false. +*/ + + +#include <string.h> + + +static void *upb_calloc(size_t size) { + void *mem = upb_gmalloc(size); + if (mem) { + memset(mem, 0, size); + } + return mem; +} + +/* Defined for the sole purpose of having a unique pointer value for + * UPB_NO_CLOSURE. */ +char _upb_noclosure; + +static void freehandlers(upb_refcounted *r) { + upb_handlers *h = (upb_handlers*)r; + + upb_inttable_iter i; + upb_inttable_begin(&i, &h->cleanup_); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + void *val = (void*)upb_inttable_iter_key(&i); + upb_value func_val = upb_inttable_iter_value(&i); + upb_handlerfree *func = upb_value_getfptr(func_val); + func(val); + } + + upb_inttable_uninit(&h->cleanup_); + upb_msgdef_unref(h->msg, h); + upb_gfree(h->sub); + upb_gfree(h); +} + +static void visithandlers(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_handlers *h = (const upb_handlers*)r; + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, h->msg); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + const upb_handlers *sub; + if (!upb_fielddef_issubmsg(f)) continue; + sub = upb_handlers_getsubhandlers(h, f); + if (sub) visit(r, upb_handlers_upcast(sub), closure); + } +} + +static const struct upb_refcounted_vtbl vtbl = {visithandlers, freehandlers}; + +typedef struct { + upb_inttable tab; /* maps upb_msgdef* -> upb_handlers*. */ + upb_handlers_callback *callback; + const void *closure; +} dfs_state; + +/* TODO(haberman): discard upb_handlers* objects that do not actually have any + * handlers set and cannot reach any upb_handlers* object that does. This is + * slightly tricky to do correctly. */ +static upb_handlers *newformsg(const upb_msgdef *m, const void *owner, + dfs_state *s) { + upb_msg_field_iter i; + upb_handlers *h = upb_handlers_new(m, owner); + if (!h) return NULL; + if (!upb_inttable_insertptr(&s->tab, m, upb_value_ptr(h))) goto oom; + + s->callback(s->closure, h); + + /* For each submessage field, get or create a handlers object and set it as + * the subhandlers. */ + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + const upb_msgdef *subdef; + upb_value subm_ent; + + if (!upb_fielddef_issubmsg(f)) continue; + + subdef = upb_downcast_msgdef(upb_fielddef_subdef(f)); + if (upb_inttable_lookupptr(&s->tab, subdef, &subm_ent)) { + upb_handlers_setsubhandlers(h, f, upb_value_getptr(subm_ent)); + } else { + upb_handlers *sub_mh = newformsg(subdef, &sub_mh, s); + if (!sub_mh) goto oom; + upb_handlers_setsubhandlers(h, f, sub_mh); + upb_handlers_unref(sub_mh, &sub_mh); + } + } + return h; + +oom: + upb_handlers_unref(h, owner); + return NULL; +} + +/* Given a selector for a STARTSUBMSG handler, resolves to a pointer to the + * subhandlers for this submessage field. */ +#define SUBH(h, selector) (h->sub[selector]) + +/* The selector for a submessage field is the field index. */ +#define SUBH_F(h, f) SUBH(h, f->index_) + +static int32_t trygetsel(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + upb_selector_t sel; + assert(!upb_handlers_isfrozen(h)); + if (upb_handlers_msgdef(h) != upb_fielddef_containingtype(f)) { + upb_status_seterrf( + &h->status_, "type mismatch: field %s does not belong to message %s", + upb_fielddef_name(f), upb_msgdef_fullname(upb_handlers_msgdef(h))); + return -1; + } + if (!upb_handlers_getselector(f, type, &sel)) { + upb_status_seterrf( + &h->status_, + "type mismatch: cannot register handler type %d for field %s", + type, upb_fielddef_name(f)); + return -1; + } + return sel; +} + +static upb_selector_t handlers_getsel(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + int32_t sel = trygetsel(h, f, type); + assert(sel >= 0); + return sel; +} + +static const void **returntype(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + return &h->table[handlers_getsel(h, f, type)].attr.return_closure_type_; +} + +static bool doset(upb_handlers *h, int32_t sel, const upb_fielddef *f, + upb_handlertype_t type, upb_func *func, + upb_handlerattr *attr) { + upb_handlerattr set_attr = UPB_HANDLERATTR_INITIALIZER; + const void *closure_type; + const void **context_closure_type; + + assert(!upb_handlers_isfrozen(h)); + + if (sel < 0) { + upb_status_seterrmsg(&h->status_, + "incorrect handler type for this field."); + return false; + } + + if (h->table[sel].func) { + upb_status_seterrmsg(&h->status_, + "cannot change handler once it has been set."); + return false; + } + + if (attr) { + set_attr = *attr; + } + + /* Check that the given closure type matches the closure type that has been + * established for this context (if any). */ + closure_type = upb_handlerattr_closuretype(&set_attr); + + if (type == UPB_HANDLER_STRING) { + context_closure_type = returntype(h, f, UPB_HANDLER_STARTSTR); + } else if (f && upb_fielddef_isseq(f) && + type != UPB_HANDLER_STARTSEQ && + type != UPB_HANDLER_ENDSEQ) { + context_closure_type = returntype(h, f, UPB_HANDLER_STARTSEQ); + } else { + context_closure_type = &h->top_closure_type; + } + + if (closure_type && *context_closure_type && + closure_type != *context_closure_type) { + /* TODO(haberman): better message for debugging. */ + if (f) { + upb_status_seterrf(&h->status_, + "closure type does not match for field %s", + upb_fielddef_name(f)); + } else { + upb_status_seterrmsg( + &h->status_, "closure type does not match for message-level handler"); + } + return false; + } + + if (closure_type) + *context_closure_type = closure_type; + + /* If this is a STARTSEQ or STARTSTR handler, check that the returned pointer + * matches any pre-existing expectations about what type is expected. */ + if (type == UPB_HANDLER_STARTSEQ || type == UPB_HANDLER_STARTSTR) { + const void *return_type = upb_handlerattr_returnclosuretype(&set_attr); + const void *table_return_type = + upb_handlerattr_returnclosuretype(&h->table[sel].attr); + if (return_type && table_return_type && return_type != table_return_type) { + upb_status_seterrmsg(&h->status_, "closure return type does not match"); + return false; + } + + if (table_return_type && !return_type) + upb_handlerattr_setreturnclosuretype(&set_attr, table_return_type); + } + + h->table[sel].func = (upb_func*)func; + h->table[sel].attr = set_attr; + return true; +} + +/* Returns the effective closure type for this handler (which will propagate + * from outer frames if this frame has no START* handler). Not implemented for + * UPB_HANDLER_STRING at the moment since this is not needed. Returns NULL is + * the effective closure type is unspecified (either no handler was registered + * to specify it or the handler that was registered did not specify the closure + * type). */ +const void *effective_closure_type(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + const void *ret; + upb_selector_t sel; + + assert(type != UPB_HANDLER_STRING); + ret = h->top_closure_type; + + if (upb_fielddef_isseq(f) && + type != UPB_HANDLER_STARTSEQ && + type != UPB_HANDLER_ENDSEQ && + h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)].func) { + ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + } + + if (type == UPB_HANDLER_STRING && + h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSTR)].func) { + ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + } + + /* The effective type of the submessage; not used yet. + * if (type == SUBMESSAGE && + * h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)].func) { + * ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + * } */ + + return ret; +} + +/* Checks whether the START* handler specified by f & type is missing even + * though it is required to convert the established type of an outer frame + * ("closure_type") into the established type of an inner frame (represented in + * the return closure type of this handler's attr. */ +bool checkstart(upb_handlers *h, const upb_fielddef *f, upb_handlertype_t type, + upb_status *status) { + const void *closure_type; + const upb_handlerattr *attr; + const void *return_closure_type; + + upb_selector_t sel = handlers_getsel(h, f, type); + if (h->table[sel].func) return true; + closure_type = effective_closure_type(h, f, type); + attr = &h->table[sel].attr; + return_closure_type = upb_handlerattr_returnclosuretype(attr); + if (closure_type && return_closure_type && + closure_type != return_closure_type) { + upb_status_seterrf(status, + "expected start handler to return sub type for field %f", + upb_fielddef_name(f)); + return false; + } + return true; +} + +/* Public interface ***********************************************************/ + +upb_handlers *upb_handlers_new(const upb_msgdef *md, const void *owner) { + int extra; + upb_handlers *h; + + assert(upb_msgdef_isfrozen(md)); + + extra = sizeof(upb_handlers_tabent) * (md->selector_count - 1); + h = upb_calloc(sizeof(*h) + extra); + if (!h) return NULL; + + h->msg = md; + upb_msgdef_ref(h->msg, h); + upb_status_clear(&h->status_); + + if (md->submsg_field_count > 0) { + h->sub = upb_calloc(md->submsg_field_count * sizeof(*h->sub)); + if (!h->sub) goto oom; + } else { + h->sub = 0; + } + + if (!upb_refcounted_init(upb_handlers_upcast_mutable(h), &vtbl, owner)) + goto oom; + if (!upb_inttable_init(&h->cleanup_, UPB_CTYPE_FPTR)) goto oom; + + /* calloc() above initialized all handlers to NULL. */ + return h; + +oom: + freehandlers(upb_handlers_upcast_mutable(h)); + return NULL; +} + +const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m, + const void *owner, + upb_handlers_callback *callback, + const void *closure) { + dfs_state state; + upb_handlers *ret; + bool ok; + upb_refcounted *r; + + state.callback = callback; + state.closure = closure; + if (!upb_inttable_init(&state.tab, UPB_CTYPE_PTR)) return NULL; + + ret = newformsg(m, owner, &state); + + upb_inttable_uninit(&state.tab); + if (!ret) return NULL; + + r = upb_handlers_upcast_mutable(ret); + ok = upb_refcounted_freeze(&r, 1, NULL, UPB_MAX_HANDLER_DEPTH); + UPB_ASSERT_VAR(ok, ok); + + return ret; +} + +const upb_status *upb_handlers_status(upb_handlers *h) { + assert(!upb_handlers_isfrozen(h)); + return &h->status_; +} + +void upb_handlers_clearerr(upb_handlers *h) { + assert(!upb_handlers_isfrozen(h)); + upb_status_clear(&h->status_); +} + +#define SETTER(name, handlerctype, handlertype) \ + bool upb_handlers_set ## name(upb_handlers *h, const upb_fielddef *f, \ + handlerctype func, upb_handlerattr *attr) { \ + int32_t sel = trygetsel(h, f, handlertype); \ + return doset(h, sel, f, handlertype, (upb_func*)func, attr); \ + } + +SETTER(int32, upb_int32_handlerfunc*, UPB_HANDLER_INT32) +SETTER(int64, upb_int64_handlerfunc*, UPB_HANDLER_INT64) +SETTER(uint32, upb_uint32_handlerfunc*, UPB_HANDLER_UINT32) +SETTER(uint64, upb_uint64_handlerfunc*, UPB_HANDLER_UINT64) +SETTER(float, upb_float_handlerfunc*, UPB_HANDLER_FLOAT) +SETTER(double, upb_double_handlerfunc*, UPB_HANDLER_DOUBLE) +SETTER(bool, upb_bool_handlerfunc*, UPB_HANDLER_BOOL) +SETTER(startstr, upb_startstr_handlerfunc*, UPB_HANDLER_STARTSTR) +SETTER(string, upb_string_handlerfunc*, UPB_HANDLER_STRING) +SETTER(endstr, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSTR) +SETTER(startseq, upb_startfield_handlerfunc*, UPB_HANDLER_STARTSEQ) +SETTER(startsubmsg, upb_startfield_handlerfunc*, UPB_HANDLER_STARTSUBMSG) +SETTER(endsubmsg, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSUBMSG) +SETTER(endseq, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSEQ) + +#undef SETTER + +bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func, + upb_handlerattr *attr) { + return doset(h, UPB_STARTMSG_SELECTOR, NULL, UPB_HANDLER_INT32, + (upb_func *)func, attr); +} + +bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func, + upb_handlerattr *attr) { + assert(!upb_handlers_isfrozen(h)); + return doset(h, UPB_ENDMSG_SELECTOR, NULL, UPB_HANDLER_INT32, + (upb_func *)func, attr); +} + +bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f, + const upb_handlers *sub) { + assert(sub); + assert(!upb_handlers_isfrozen(h)); + assert(upb_fielddef_issubmsg(f)); + if (SUBH_F(h, f)) return false; /* Can't reset. */ + if (upb_msgdef_upcast(upb_handlers_msgdef(sub)) != upb_fielddef_subdef(f)) { + return false; + } + SUBH_F(h, f) = sub; + upb_ref2(sub, h); + return true; +} + +const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h, + const upb_fielddef *f) { + assert(upb_fielddef_issubmsg(f)); + return SUBH_F(h, f); +} + +bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t sel, + upb_handlerattr *attr) { + if (!upb_handlers_gethandler(h, sel)) + return false; + *attr = h->table[sel].attr; + return true; +} + +const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h, + upb_selector_t sel) { + /* STARTSUBMSG selector in sel is the field's selector base. */ + return SUBH(h, sel - UPB_STATIC_SELECTOR_COUNT); +} + +const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h) { return h->msg; } + +bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *func) { + bool ok; + if (upb_inttable_lookupptr(&h->cleanup_, p, NULL)) { + return false; + } + ok = upb_inttable_insertptr(&h->cleanup_, p, upb_value_fptr(func)); + UPB_ASSERT_VAR(ok, ok); + return true; +} + + +/* "Static" methods ***********************************************************/ + +bool upb_handlers_freeze(upb_handlers *const*handlers, int n, upb_status *s) { + /* TODO: verify we have a transitive closure. */ + int i; + for (i = 0; i < n; i++) { + upb_msg_field_iter j; + upb_handlers *h = handlers[i]; + + if (!upb_ok(&h->status_)) { + upb_status_seterrf(s, "handlers for message %s had error status: %s", + upb_msgdef_fullname(upb_handlers_msgdef(h)), + upb_status_errmsg(&h->status_)); + return false; + } + + /* Check that there are no closure mismatches due to missing Start* handlers + * or subhandlers with different type-level types. */ + for(upb_msg_field_begin(&j, h->msg); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + + const upb_fielddef *f = upb_msg_iter_field(&j); + if (upb_fielddef_isseq(f)) { + if (!checkstart(h, f, UPB_HANDLER_STARTSEQ, s)) + return false; + } + + if (upb_fielddef_isstring(f)) { + if (!checkstart(h, f, UPB_HANDLER_STARTSTR, s)) + return false; + } + + if (upb_fielddef_issubmsg(f)) { + bool hashandler = false; + if (upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)) || + upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_ENDSUBMSG))) { + hashandler = true; + } + + if (upb_fielddef_isseq(f) && + (upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)) || + upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_ENDSEQ)))) { + hashandler = true; + } + + if (hashandler && !upb_handlers_getsubhandlers(h, f)) { + /* For now we add an empty subhandlers in this case. It makes the + * decoder code generator simpler, because it only has to handle two + * cases (submessage has handlers or not) as opposed to three + * (submessage has handlers in enclosing message but no subhandlers). + * + * This makes parsing less efficient in the case that we want to + * notice a submessage but skip its contents (like if we're testing + * for submessage presence or counting the number of repeated + * submessages). In this case we will end up parsing the submessage + * field by field and throwing away the results for each, instead of + * skipping the whole delimited thing at once. If this is an issue we + * can revisit it, but do remember that this only arises when you have + * handlers (startseq/startsubmsg/endsubmsg/endseq) set for the + * submessage but no subhandlers. The uses cases for this are + * limited. */ + upb_handlers *sub = upb_handlers_new(upb_fielddef_msgsubdef(f), &sub); + upb_handlers_setsubhandlers(h, f, sub); + upb_handlers_unref(sub, &sub); + } + + /* TODO(haberman): check type of submessage. + * This is slightly tricky; also consider whether we should check that + * they match at setsubhandlers time. */ + } + } + } + + if (!upb_refcounted_freeze((upb_refcounted*const*)handlers, n, s, + UPB_MAX_HANDLER_DEPTH)) { + return false; + } + + return true; +} + +upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + case UPB_TYPE_ENUM: return UPB_HANDLER_INT32; + case UPB_TYPE_INT64: return UPB_HANDLER_INT64; + case UPB_TYPE_UINT32: return UPB_HANDLER_UINT32; + case UPB_TYPE_UINT64: return UPB_HANDLER_UINT64; + case UPB_TYPE_FLOAT: return UPB_HANDLER_FLOAT; + case UPB_TYPE_DOUBLE: return UPB_HANDLER_DOUBLE; + case UPB_TYPE_BOOL: return UPB_HANDLER_BOOL; + default: assert(false); return -1; /* Invalid input. */ + } +} + +bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type, + upb_selector_t *s) { + switch (type) { + case UPB_HANDLER_INT32: + case UPB_HANDLER_INT64: + case UPB_HANDLER_UINT32: + case UPB_HANDLER_UINT64: + case UPB_HANDLER_FLOAT: + case UPB_HANDLER_DOUBLE: + case UPB_HANDLER_BOOL: + if (!upb_fielddef_isprimitive(f) || + upb_handlers_getprimitivehandlertype(f) != type) + return false; + *s = f->selector_base; + break; + case UPB_HANDLER_STRING: + if (upb_fielddef_isstring(f)) { + *s = f->selector_base; + } else if (upb_fielddef_lazy(f)) { + *s = f->selector_base + 3; + } else { + return false; + } + break; + case UPB_HANDLER_STARTSTR: + if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) { + *s = f->selector_base + 1; + } else { + return false; + } + break; + case UPB_HANDLER_ENDSTR: + if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) { + *s = f->selector_base + 2; + } else { + return false; + } + break; + case UPB_HANDLER_STARTSEQ: + if (!upb_fielddef_isseq(f)) return false; + *s = f->selector_base - 2; + break; + case UPB_HANDLER_ENDSEQ: + if (!upb_fielddef_isseq(f)) return false; + *s = f->selector_base - 1; + break; + case UPB_HANDLER_STARTSUBMSG: + if (!upb_fielddef_issubmsg(f)) return false; + /* Selectors for STARTSUBMSG are at the beginning of the table so that the + * selector can also be used as an index into the "sub" array of + * subhandlers. The indexes for the two into these two tables are the + * same, except that in the handler table the static selectors come first. */ + *s = f->index_ + UPB_STATIC_SELECTOR_COUNT; + break; + case UPB_HANDLER_ENDSUBMSG: + if (!upb_fielddef_issubmsg(f)) return false; + *s = f->selector_base; + break; + } + assert((size_t)*s < upb_fielddef_containingtype(f)->selector_count); + return true; +} + +uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f) { + return upb_fielddef_isseq(f) ? 2 : 0; +} + +uint32_t upb_handlers_selectorcount(const upb_fielddef *f) { + uint32_t ret = 1; + if (upb_fielddef_isseq(f)) ret += 2; /* STARTSEQ/ENDSEQ */ + if (upb_fielddef_isstring(f)) ret += 2; /* [STRING]/STARTSTR/ENDSTR */ + if (upb_fielddef_issubmsg(f)) { + /* ENDSUBMSG (STARTSUBMSG is at table beginning) */ + ret += 0; + if (upb_fielddef_lazy(f)) { + /* STARTSTR/ENDSTR/STRING (for lazy) */ + ret += 3; + } + } + return ret; +} + + +/* upb_handlerattr ************************************************************/ + +void upb_handlerattr_init(upb_handlerattr *attr) { + upb_handlerattr from = UPB_HANDLERATTR_INITIALIZER; + memcpy(attr, &from, sizeof(*attr)); +} + +void upb_handlerattr_uninit(upb_handlerattr *attr) { + UPB_UNUSED(attr); +} + +bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd) { + attr->handler_data_ = hd; + return true; +} + +bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type) { + attr->closure_type_ = type; + return true; +} + +const void *upb_handlerattr_closuretype(const upb_handlerattr *attr) { + return attr->closure_type_; +} + +bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr, + const void *type) { + attr->return_closure_type_ = type; + return true; +} + +const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr) { + return attr->return_closure_type_; +} + +bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok) { + attr->alwaysok_ = alwaysok; + return true; +} + +bool upb_handlerattr_alwaysok(const upb_handlerattr *attr) { + return attr->alwaysok_; +} + +/* upb_bufhandle **************************************************************/ + +size_t upb_bufhandle_objofs(const upb_bufhandle *h) { + return h->objofs_; +} + +/* upb_byteshandler ***********************************************************/ + +void upb_byteshandler_init(upb_byteshandler* h) { + memset(h, 0, sizeof(*h)); +} + +/* For when we support handlerfree callbacks. */ +void upb_byteshandler_uninit(upb_byteshandler* h) { + UPB_UNUSED(h); +} + +bool upb_byteshandler_setstartstr(upb_byteshandler *h, + upb_startstr_handlerfunc *func, void *d) { + h->table[UPB_STARTSTR_SELECTOR].func = (upb_func*)func; + h->table[UPB_STARTSTR_SELECTOR].attr.handler_data_ = d; + return true; +} + +bool upb_byteshandler_setstring(upb_byteshandler *h, + upb_string_handlerfunc *func, void *d) { + h->table[UPB_STRING_SELECTOR].func = (upb_func*)func; + h->table[UPB_STRING_SELECTOR].attr.handler_data_ = d; + return true; +} + +bool upb_byteshandler_setendstr(upb_byteshandler *h, + upb_endfield_handlerfunc *func, void *d) { + h->table[UPB_ENDSTR_SELECTOR].func = (upb_func*)func; + h->table[UPB_ENDSTR_SELECTOR].attr.handler_data_ = d; + return true; +} +/* +** upb::RefCounted Implementation +** +** Our key invariants are: +** 1. reference cycles never span groups +** 2. for ref2(to, from), we increment to's count iff group(from) != group(to) +** +** The previous two are how we avoid leaking cycles. Other important +** invariants are: +** 3. for mutable objects "from" and "to", if there exists a ref2(to, from) +** this implies group(from) == group(to). (In practice, what we implement +** is even stronger; "from" and "to" will share a group if there has *ever* +** been a ref2(to, from), but all that is necessary for correctness is the +** weaker one). +** 4. mutable and immutable objects are never in the same group. +*/ + + +#include <setjmp.h> + +static void freeobj(upb_refcounted *o); + +const char untracked_val; +const void *UPB_UNTRACKED_REF = &untracked_val; + +/* arch-specific atomic primitives *******************************************/ + +#ifdef UPB_THREAD_UNSAFE /*---------------------------------------------------*/ + +static void atomic_inc(uint32_t *a) { (*a)++; } +static bool atomic_dec(uint32_t *a) { return --(*a) == 0; } + +#elif defined(__GNUC__) || defined(__clang__) /*------------------------------*/ + +static void atomic_inc(uint32_t *a) { __sync_fetch_and_add(a, 1); } +static bool atomic_dec(uint32_t *a) { return __sync_sub_and_fetch(a, 1) == 0; } + +#elif defined(WIN32) /*-------------------------------------------------------*/ + +#include <Windows.h> + +static void atomic_inc(upb_atomic_t *a) { InterlockedIncrement(&a->val); } +static bool atomic_dec(upb_atomic_t *a) { + return InterlockedDecrement(&a->val) == 0; +} + +#else +#error Atomic primitives not defined for your platform/CPU. \ + Implement them or compile with UPB_THREAD_UNSAFE. +#endif + +/* All static objects point to this refcount. + * It is special-cased in ref/unref below. */ +uint32_t static_refcount = -1; + +/* We can avoid atomic ops for statically-declared objects. + * This is a minor optimization but nice since we can avoid degrading under + * contention in this case. */ + +static void refgroup(uint32_t *group) { + if (group != &static_refcount) + atomic_inc(group); +} + +static bool unrefgroup(uint32_t *group) { + if (group == &static_refcount) { + return false; + } else { + return atomic_dec(group); + } +} + + +/* Reference tracking (debug only) ********************************************/ + +#ifdef UPB_DEBUG_REFS + +#ifdef UPB_THREAD_UNSAFE + +static void upb_lock() {} +static void upb_unlock() {} + +#else + +/* User must define functions that lock/unlock a global mutex and link this + * file against them. */ +void upb_lock(); +void upb_unlock(); + +#endif + +/* UPB_DEBUG_REFS mode counts on being able to malloc() memory in some + * code-paths that can normally never fail, like upb_refcounted_ref(). Since + * we have no way to propagage out-of-memory errors back to the user, and since + * these errors can only occur in UPB_DEBUG_REFS mode, we use an allocator that + * immediately aborts on failure (avoiding the global allocator, which might + * inject failures). */ + +#include <stdlib.h> + +static void *upb_debugrefs_allocfunc(upb_alloc *alloc, void *ptr, + size_t oldsize, size_t size) { + UPB_UNUSED(alloc); + UPB_UNUSED(oldsize); + if (size == 0) { + free(ptr); + return NULL; + } else { + void *ret = realloc(ptr, size); + + if (!ret) { + abort(); + } + + return ret; + } +} + +upb_alloc upb_alloc_debugrefs = {&upb_debugrefs_allocfunc}; + +typedef struct { + int count; /* How many refs there are (duplicates only allowed for ref2). */ + bool is_ref2; +} trackedref; + +static trackedref *trackedref_new(bool is_ref2) { + trackedref *ret = upb_malloc(&upb_alloc_debugrefs, sizeof(*ret)); + ret->count = 1; + ret->is_ref2 = is_ref2; + return ret; +} + +static void track(const upb_refcounted *r, const void *owner, bool ref2) { + upb_value v; + + assert(owner); + if (owner == UPB_UNTRACKED_REF) return; + + upb_lock(); + if (upb_inttable_lookupptr(r->refs, owner, &v)) { + trackedref *ref = upb_value_getptr(v); + /* Since we allow multiple ref2's for the same to/from pair without + * allocating separate memory for each one, we lose the fine-grained + * tracking behavior we get with regular refs. Since ref2s only happen + * inside upb, we'll accept this limitation until/unless there is a really + * difficult upb-internal bug that can't be figured out without it. */ + assert(ref2); + assert(ref->is_ref2); + ref->count++; + } else { + trackedref *ref = trackedref_new(ref2); + upb_inttable_insertptr2(r->refs, owner, upb_value_ptr(ref), + &upb_alloc_debugrefs); + if (ref2) { + /* We know this cast is safe when it is a ref2, because it's coming from + * another refcounted object. */ + const upb_refcounted *from = owner; + assert(!upb_inttable_lookupptr(from->ref2s, r, NULL)); + upb_inttable_insertptr2(from->ref2s, r, upb_value_ptr(NULL), + &upb_alloc_debugrefs); + } + } + upb_unlock(); +} + +static void untrack(const upb_refcounted *r, const void *owner, bool ref2) { + upb_value v; + bool found; + trackedref *ref; + + assert(owner); + if (owner == UPB_UNTRACKED_REF) return; + + upb_lock(); + found = upb_inttable_lookupptr(r->refs, owner, &v); + /* This assert will fail if an owner attempts to release a ref it didn't have. */ + UPB_ASSERT_VAR(found, found); + ref = upb_value_getptr(v); + assert(ref->is_ref2 == ref2); + if (--ref->count == 0) { + free(ref); + upb_inttable_removeptr(r->refs, owner, NULL); + if (ref2) { + /* We know this cast is safe when it is a ref2, because it's coming from + * another refcounted object. */ + const upb_refcounted *from = owner; + bool removed = upb_inttable_removeptr(from->ref2s, r, NULL); + assert(removed); + } + } + upb_unlock(); +} + +static void checkref(const upb_refcounted *r, const void *owner, bool ref2) { + upb_value v; + bool found; + trackedref *ref; + + upb_lock(); + found = upb_inttable_lookupptr(r->refs, owner, &v); + UPB_ASSERT_VAR(found, found); + ref = upb_value_getptr(v); + assert(ref->is_ref2 == ref2); + upb_unlock(); +} + +/* Populates the given UPB_CTYPE_INT32 inttable with counts of ref2's that + * originate from the given owner. */ +static void getref2s(const upb_refcounted *owner, upb_inttable *tab) { + upb_inttable_iter i; + + upb_lock(); + upb_inttable_begin(&i, owner->ref2s); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_value v; + upb_value count; + trackedref *ref; + bool found; + + upb_refcounted *to = (upb_refcounted*)upb_inttable_iter_key(&i); + + /* To get the count we need to look in the target's table. */ + found = upb_inttable_lookupptr(to->refs, owner, &v); + assert(found); + ref = upb_value_getptr(v); + count = upb_value_int32(ref->count); + + upb_inttable_insertptr2(tab, to, count, &upb_alloc_debugrefs); + } + upb_unlock(); +} + +typedef struct { + upb_inttable ref2; + const upb_refcounted *obj; +} check_state; + +static void visit_check(const upb_refcounted *obj, const upb_refcounted *subobj, + void *closure) { + check_state *s = closure; + upb_inttable *ref2 = &s->ref2; + upb_value v; + bool removed; + int32_t newcount; + + assert(obj == s->obj); + assert(subobj); + removed = upb_inttable_removeptr(ref2, subobj, &v); + /* The following assertion will fail if the visit() function visits a subobj + * that it did not have a ref2 on, or visits the same subobj too many times. */ + assert(removed); + newcount = upb_value_getint32(v) - 1; + if (newcount > 0) { + upb_inttable_insert2(ref2, (uintptr_t)subobj, upb_value_int32(newcount), + &upb_alloc_debugrefs); + } +} + +static void visit(const upb_refcounted *r, upb_refcounted_visit *v, + void *closure) { + /* In DEBUG_REFS mode we know what existing ref2 refs there are, so we know + * exactly the set of nodes that visit() should visit. So we verify visit()'s + * correctness here. */ + check_state state; + state.obj = r; + upb_inttable_init2(&state.ref2, UPB_CTYPE_INT32, &upb_alloc_debugrefs); + getref2s(r, &state.ref2); + + /* This should visit any children in the ref2 table. */ + if (r->vtbl->visit) r->vtbl->visit(r, visit_check, &state); + + /* This assertion will fail if the visit() function missed any children. */ + assert(upb_inttable_count(&state.ref2) == 0); + upb_inttable_uninit2(&state.ref2, &upb_alloc_debugrefs); + if (r->vtbl->visit) r->vtbl->visit(r, v, closure); +} + +static void trackinit(upb_refcounted *r) { + r->refs = upb_malloc(&upb_alloc_debugrefs, sizeof(*r->refs)); + r->ref2s = upb_malloc(&upb_alloc_debugrefs, sizeof(*r->ref2s)); + upb_inttable_init2(r->refs, UPB_CTYPE_PTR, &upb_alloc_debugrefs); + upb_inttable_init2(r->ref2s, UPB_CTYPE_PTR, &upb_alloc_debugrefs); +} + +static void trackfree(const upb_refcounted *r) { + upb_inttable_uninit2(r->refs, &upb_alloc_debugrefs); + upb_inttable_uninit2(r->ref2s, &upb_alloc_debugrefs); + upb_free(&upb_alloc_debugrefs, r->refs); + upb_free(&upb_alloc_debugrefs, r->ref2s); +} + +#else + +static void track(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static void untrack(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static void checkref(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static void trackinit(upb_refcounted *r) { + UPB_UNUSED(r); +} + +static void trackfree(const upb_refcounted *r) { + UPB_UNUSED(r); +} + +static void visit(const upb_refcounted *r, upb_refcounted_visit *v, + void *closure) { + if (r->vtbl->visit) r->vtbl->visit(r, v, closure); +} + +#endif /* UPB_DEBUG_REFS */ + + +/* freeze() *******************************************************************/ + +/* The freeze() operation is by far the most complicated part of this scheme. + * We compute strongly-connected components and then mutate the graph such that + * we preserve the invariants documented at the top of this file. And we must + * handle out-of-memory errors gracefully (without leaving the graph + * inconsistent), which adds to the fun. */ + +/* The state used by the freeze operation (shared across many functions). */ +typedef struct { + int depth; + int maxdepth; + uint64_t index; + /* Maps upb_refcounted* -> attributes (color, etc). attr layout varies by + * color. */ + upb_inttable objattr; + upb_inttable stack; /* stack of upb_refcounted* for Tarjan's algorithm. */ + upb_inttable groups; /* array of uint32_t*, malloc'd refcounts for new groups */ + upb_status *status; + jmp_buf err; +} tarjan; + +static void release_ref2(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure); + +/* Node attributes -----------------------------------------------------------*/ + +/* After our analysis phase all nodes will be either GRAY or WHITE. */ + +typedef enum { + BLACK = 0, /* Object has not been seen. */ + GRAY, /* Object has been found via a refgroup but may not be reachable. */ + GREEN, /* Object is reachable and is currently on the Tarjan stack. */ + WHITE /* Object is reachable and has been assigned a group (SCC). */ +} color_t; + +UPB_NORETURN static void err(tarjan *t) { longjmp(t->err, 1); } +UPB_NORETURN static void oom(tarjan *t) { + upb_status_seterrmsg(t->status, "out of memory"); + err(t); +} + +static uint64_t trygetattr(const tarjan *t, const upb_refcounted *r) { + upb_value v; + return upb_inttable_lookupptr(&t->objattr, r, &v) ? + upb_value_getuint64(v) : 0; +} + +static uint64_t getattr(const tarjan *t, const upb_refcounted *r) { + upb_value v; + bool found = upb_inttable_lookupptr(&t->objattr, r, &v); + UPB_ASSERT_VAR(found, found); + return upb_value_getuint64(v); +} + +static void setattr(tarjan *t, const upb_refcounted *r, uint64_t attr) { + upb_inttable_removeptr(&t->objattr, r, NULL); + upb_inttable_insertptr(&t->objattr, r, upb_value_uint64(attr)); +} + +static color_t color(tarjan *t, const upb_refcounted *r) { + return trygetattr(t, r) & 0x3; /* Color is always stored in the low 2 bits. */ +} + +static void set_gray(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == BLACK); + setattr(t, r, GRAY); +} + +/* Pushes an obj onto the Tarjan stack and sets it to GREEN. */ +static void push(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == BLACK || color(t, r) == GRAY); + /* This defines the attr layout for the GREEN state. "index" and "lowlink" + * get 31 bits, which is plenty (limit of 2B objects frozen at a time). */ + setattr(t, r, GREEN | (t->index << 2) | (t->index << 33)); + if (++t->index == 0x80000000) { + upb_status_seterrmsg(t->status, "too many objects to freeze"); + err(t); + } + upb_inttable_push(&t->stack, upb_value_ptr((void*)r)); +} + +/* Pops an obj from the Tarjan stack and sets it to WHITE, with a ptr to its + * SCC group. */ +static upb_refcounted *pop(tarjan *t) { + upb_refcounted *r = upb_value_getptr(upb_inttable_pop(&t->stack)); + assert(color(t, r) == GREEN); + /* This defines the attr layout for nodes in the WHITE state. + * Top of group stack is [group, NULL]; we point at group. */ + setattr(t, r, WHITE | (upb_inttable_count(&t->groups) - 2) << 8); + return r; +} + +static void tarjan_newgroup(tarjan *t) { + uint32_t *group = upb_gmalloc(sizeof(*group)); + if (!group) oom(t); + /* Push group and empty group leader (we'll fill in leader later). */ + if (!upb_inttable_push(&t->groups, upb_value_ptr(group)) || + !upb_inttable_push(&t->groups, upb_value_ptr(NULL))) { + upb_gfree(group); + oom(t); + } + *group = 0; +} + +static uint32_t idx(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == GREEN); + return (getattr(t, r) >> 2) & 0x7FFFFFFF; +} + +static uint32_t lowlink(tarjan *t, const upb_refcounted *r) { + if (color(t, r) == GREEN) { + return getattr(t, r) >> 33; + } else { + return UINT32_MAX; + } +} + +static void set_lowlink(tarjan *t, const upb_refcounted *r, uint32_t lowlink) { + assert(color(t, r) == GREEN); + setattr(t, r, ((uint64_t)lowlink << 33) | (getattr(t, r) & 0x1FFFFFFFF)); +} + +static uint32_t *group(tarjan *t, upb_refcounted *r) { + uint64_t groupnum; + upb_value v; + bool found; + + assert(color(t, r) == WHITE); + groupnum = getattr(t, r) >> 8; + found = upb_inttable_lookup(&t->groups, groupnum, &v); + UPB_ASSERT_VAR(found, found); + return upb_value_getptr(v); +} + +/* If the group leader for this object's group has not previously been set, + * the given object is assigned to be its leader. */ +static upb_refcounted *groupleader(tarjan *t, upb_refcounted *r) { + uint64_t leader_slot; + upb_value v; + bool found; + + assert(color(t, r) == WHITE); + leader_slot = (getattr(t, r) >> 8) + 1; + found = upb_inttable_lookup(&t->groups, leader_slot, &v); + UPB_ASSERT_VAR(found, found); + if (upb_value_getptr(v)) { + return upb_value_getptr(v); + } else { + upb_inttable_remove(&t->groups, leader_slot, NULL); + upb_inttable_insert(&t->groups, leader_slot, upb_value_ptr(r)); + return r; + } +} + + +/* Tarjan's algorithm --------------------------------------------------------*/ + +/* See: + * http://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm */ +static void do_tarjan(const upb_refcounted *obj, tarjan *t); + +static void tarjan_visit(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure) { + tarjan *t = closure; + if (++t->depth > t->maxdepth) { + upb_status_seterrf(t->status, "graph too deep to freeze (%d)", t->maxdepth); + err(t); + } else if (subobj->is_frozen || color(t, subobj) == WHITE) { + /* Do nothing: we don't want to visit or color already-frozen nodes, + * and WHITE nodes have already been assigned a SCC. */ + } else if (color(t, subobj) < GREEN) { + /* Subdef has not yet been visited; recurse on it. */ + do_tarjan(subobj, t); + set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), lowlink(t, subobj))); + } else if (color(t, subobj) == GREEN) { + /* Subdef is in the stack and hence in the current SCC. */ + set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), idx(t, subobj))); + } + --t->depth; +} + +static void do_tarjan(const upb_refcounted *obj, tarjan *t) { + if (color(t, obj) == BLACK) { + /* We haven't seen this object's group; mark the whole group GRAY. */ + const upb_refcounted *o = obj; + do { set_gray(t, o); } while ((o = o->next) != obj); + } + + push(t, obj); + visit(obj, tarjan_visit, t); + if (lowlink(t, obj) == idx(t, obj)) { + tarjan_newgroup(t); + while (pop(t) != obj) + ; + } +} + + +/* freeze() ------------------------------------------------------------------*/ + +static void crossref(const upb_refcounted *r, const upb_refcounted *subobj, + void *_t) { + tarjan *t = _t; + assert(color(t, r) > BLACK); + if (color(t, subobj) > BLACK && r->group != subobj->group) { + /* Previously this ref was not reflected in subobj->group because they + * were in the same group; now that they are split a ref must be taken. */ + refgroup(subobj->group); + } +} + +static bool freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth) { + volatile bool ret = false; + int i; + upb_inttable_iter iter; + + /* We run in two passes so that we can allocate all memory before performing + * any mutation of the input -- this allows us to leave the input unchanged + * in the case of memory allocation failure. */ + tarjan t; + t.index = 0; + t.depth = 0; + t.maxdepth = maxdepth; + t.status = s; + if (!upb_inttable_init(&t.objattr, UPB_CTYPE_UINT64)) goto err1; + if (!upb_inttable_init(&t.stack, UPB_CTYPE_PTR)) goto err2; + if (!upb_inttable_init(&t.groups, UPB_CTYPE_PTR)) goto err3; + if (setjmp(t.err) != 0) goto err4; + + + for (i = 0; i < n; i++) { + if (color(&t, roots[i]) < GREEN) { + do_tarjan(roots[i], &t); + } + } + + /* If we've made it this far, no further errors are possible so it's safe to + * mutate the objects without risk of leaving them in an inconsistent state. */ + ret = true; + + /* The transformation that follows requires care. The preconditions are: + * - all objects in attr map are WHITE or GRAY, and are in mutable groups + * (groups of all mutable objs) + * - no ref2(to, from) refs have incremented count(to) if both "to" and + * "from" are in our attr map (this follows from invariants (2) and (3)) */ + + /* Pass 1: we remove WHITE objects from their mutable groups, and add them to + * new groups according to the SCC's we computed. These new groups will + * consist of only frozen objects. None will be immediately collectible, + * because WHITE objects are by definition reachable from one of "roots", + * which the caller must own refs on. */ + upb_inttable_begin(&iter, &t.objattr); + for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter); + /* Since removal from a singly-linked list requires access to the object's + * predecessor, we consider obj->next instead of obj for moving. With the + * while() loop we guarantee that we will visit every node's predecessor. + * Proof: + * 1. every node's predecessor is in our attr map. + * 2. though the loop body may change a node's predecessor, it will only + * change it to be the node we are currently operating on, so with a + * while() loop we guarantee ourselves the chance to remove each node. */ + while (color(&t, obj->next) == WHITE && + group(&t, obj->next) != obj->next->group) { + upb_refcounted *leader; + + /* Remove from old group. */ + upb_refcounted *move = obj->next; + if (obj == move) { + /* Removing the last object from a group. */ + assert(*obj->group == obj->individual_count); + upb_gfree(obj->group); + } else { + obj->next = move->next; + /* This may decrease to zero; we'll collect GRAY objects (if any) that + * remain in the group in the third pass. */ + assert(*move->group >= move->individual_count); + *move->group -= move->individual_count; + } + + /* Add to new group. */ + leader = groupleader(&t, move); + if (move == leader) { + /* First object added to new group is its leader. */ + move->group = group(&t, move); + move->next = move; + *move->group = move->individual_count; + } else { + /* Group already has at least one object in it. */ + assert(leader->group == group(&t, move)); + move->group = group(&t, move); + move->next = leader->next; + leader->next = move; + *move->group += move->individual_count; + } + + move->is_frozen = true; + } + } + + /* Pass 2: GRAY and WHITE objects "obj" with ref2(to, obj) references must + * increment count(to) if group(obj) != group(to) (which could now be the + * case if "to" was just frozen). */ + upb_inttable_begin(&iter, &t.objattr); + for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter); + visit(obj, crossref, &t); + } + + /* Pass 3: GRAY objects are collected if their group's refcount dropped to + * zero when we removed its white nodes. This can happen if they had only + * been kept alive by virtue of sharing a group with an object that was just + * frozen. + * + * It is important that we do this last, since the GRAY object's free() + * function could call unref2() on just-frozen objects, which will decrement + * refs that were added in pass 2. */ + upb_inttable_begin(&iter, &t.objattr); + for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&iter); + if (obj->group == NULL || *obj->group == 0) { + if (obj->group) { + upb_refcounted *o; + + /* We eagerly free() the group's count (since we can't easily determine + * the group's remaining size it's the easiest way to ensure it gets + * done). */ + upb_gfree(obj->group); + + /* Visit to release ref2's (done in a separate pass since release_ref2 + * depends on o->group being unmodified so it can test merged()). */ + o = obj; + do { visit(o, release_ref2, NULL); } while ((o = o->next) != obj); + + /* Mark "group" fields as NULL so we know to free the objects later in + * this loop, but also don't try to delete the group twice. */ + o = obj; + do { o->group = NULL; } while ((o = o->next) != obj); + } + freeobj(obj); + } + } + +err4: + if (!ret) { + upb_inttable_begin(&iter, &t.groups); + for(; !upb_inttable_done(&iter); upb_inttable_next(&iter)) + upb_gfree(upb_value_getptr(upb_inttable_iter_value(&iter))); + } + upb_inttable_uninit(&t.groups); +err3: + upb_inttable_uninit(&t.stack); +err2: + upb_inttable_uninit(&t.objattr); +err1: + return ret; +} + + +/* Misc internal functions ***************************************************/ + +static bool merged(const upb_refcounted *r, const upb_refcounted *r2) { + return r->group == r2->group; +} + +static void merge(upb_refcounted *r, upb_refcounted *from) { + upb_refcounted *base; + upb_refcounted *tmp; + + if (merged(r, from)) return; + *r->group += *from->group; + upb_gfree(from->group); + base = from; + + /* Set all refcount pointers in the "from" chain to the merged refcount. + * + * TODO(haberman): this linear algorithm can result in an overall O(n^2) bound + * if the user continuously extends a group by one object. Prevent this by + * using one of the techniques in this paper: + * ftp://www.ncedc.org/outgoing/geomorph/dino/orals/p245-tarjan.pdf */ + do { from->group = r->group; } while ((from = from->next) != base); + + /* Merge the two circularly linked lists by swapping their next pointers. */ + tmp = r->next; + r->next = base->next; + base->next = tmp; +} + +static void unref(const upb_refcounted *r); + +static void release_ref2(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure) { + UPB_UNUSED(closure); + untrack(subobj, obj, true); + if (!merged(obj, subobj)) { + assert(subobj->is_frozen); + unref(subobj); + } +} + +static void unref(const upb_refcounted *r) { + if (unrefgroup(r->group)) { + const upb_refcounted *o; + + upb_gfree(r->group); + + /* In two passes, since release_ref2 needs a guarantee that any subobjs + * are alive. */ + o = r; + do { visit(o, release_ref2, NULL); } while((o = o->next) != r); + + o = r; + do { + const upb_refcounted *next = o->next; + assert(o->is_frozen || o->individual_count == 0); + freeobj((upb_refcounted*)o); + o = next; + } while(o != r); + } +} + +static void freeobj(upb_refcounted *o) { + trackfree(o); + o->vtbl->free((upb_refcounted*)o); +} + + +/* Public interface ***********************************************************/ + +bool upb_refcounted_init(upb_refcounted *r, + const struct upb_refcounted_vtbl *vtbl, + const void *owner) { +#ifndef NDEBUG + /* Endianness check. This is unrelated to upb_refcounted, it's just a + * convenient place to put the check that we can be assured will run for + * basically every program using upb. */ + const int x = 1; +#ifdef UPB_BIG_ENDIAN + assert(*(char*)&x != 1); +#else + assert(*(char*)&x == 1); +#endif +#endif + + r->next = r; + r->vtbl = vtbl; + r->individual_count = 0; + r->is_frozen = false; + r->group = upb_gmalloc(sizeof(*r->group)); + if (!r->group) return false; + *r->group = 0; + trackinit(r); + upb_refcounted_ref(r, owner); + return true; +} + +bool upb_refcounted_isfrozen(const upb_refcounted *r) { + return r->is_frozen; +} + +void upb_refcounted_ref(const upb_refcounted *r, const void *owner) { + track(r, owner, false); + if (!r->is_frozen) + ((upb_refcounted*)r)->individual_count++; + refgroup(r->group); +} + +void upb_refcounted_unref(const upb_refcounted *r, const void *owner) { + untrack(r, owner, false); + if (!r->is_frozen) + ((upb_refcounted*)r)->individual_count--; + unref(r); +} + +void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from) { + assert(!from->is_frozen); /* Non-const pointer implies this. */ + track(r, from, true); + if (r->is_frozen) { + refgroup(r->group); + } else { + merge((upb_refcounted*)r, from); + } +} + +void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from) { + assert(!from->is_frozen); /* Non-const pointer implies this. */ + untrack(r, from, true); + if (r->is_frozen) { + unref(r); + } else { + assert(merged(r, from)); + } +} + +void upb_refcounted_donateref( + const upb_refcounted *r, const void *from, const void *to) { + assert(from != to); + if (to != NULL) + upb_refcounted_ref(r, to); + if (from != NULL) + upb_refcounted_unref(r, from); +} + +void upb_refcounted_checkref(const upb_refcounted *r, const void *owner) { + checkref(r, owner, false); +} + +bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth) { + int i; + bool ret; + for (i = 0; i < n; i++) { + assert(!roots[i]->is_frozen); + } + ret = freeze(roots, n, s, maxdepth); + assert(!s || ret == upb_ok(s)); + return ret; +} + + +/* Fallback implementation if the shim is not specialized by the JIT. */ +#define SHIM_WRITER(type, ctype) \ + bool upb_shim_set ## type (void *c, const void *hd, ctype val) { \ + uint8_t *m = c; \ + const upb_shim_data *d = hd; \ + if (d->hasbit > 0) \ + *(uint8_t*)&m[d->hasbit / 8] |= 1 << (d->hasbit % 8); \ + *(ctype*)&m[d->offset] = val; \ + return true; \ + } \ + +SHIM_WRITER(double, double) +SHIM_WRITER(float, float) +SHIM_WRITER(int32, int32_t) +SHIM_WRITER(int64, int64_t) +SHIM_WRITER(uint32, uint32_t) +SHIM_WRITER(uint64, uint64_t) +SHIM_WRITER(bool, bool) +#undef SHIM_WRITER + +bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset, + int32_t hasbit) { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + bool ok; + + upb_shim_data *d = upb_gmalloc(sizeof(*d)); + if (!d) return false; + d->offset = offset; + d->hasbit = hasbit; + + upb_handlerattr_sethandlerdata(&attr, d); + upb_handlerattr_setalwaysok(&attr, true); + upb_handlers_addcleanup(h, d, upb_gfree); + +#define TYPE(u, l) \ + case UPB_TYPE_##u: \ + ok = upb_handlers_set##l(h, f, upb_shim_set##l, &attr); break; + + ok = false; + + switch (upb_fielddef_type(f)) { + TYPE(INT64, int64); + TYPE(INT32, int32); + TYPE(ENUM, int32); + TYPE(UINT64, uint64); + TYPE(UINT32, uint32); + TYPE(DOUBLE, double); + TYPE(FLOAT, float); + TYPE(BOOL, bool); + default: assert(false); break; + } +#undef TYPE + + upb_handlerattr_uninit(&attr); + return ok; +} + +const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s, + upb_fieldtype_t *type) { + upb_func *f = upb_handlers_gethandler(h, s); + + if ((upb_int64_handlerfunc*)f == upb_shim_setint64) { + *type = UPB_TYPE_INT64; + } else if ((upb_int32_handlerfunc*)f == upb_shim_setint32) { + *type = UPB_TYPE_INT32; + } else if ((upb_uint64_handlerfunc*)f == upb_shim_setuint64) { + *type = UPB_TYPE_UINT64; + } else if ((upb_uint32_handlerfunc*)f == upb_shim_setuint32) { + *type = UPB_TYPE_UINT32; + } else if ((upb_double_handlerfunc*)f == upb_shim_setdouble) { + *type = UPB_TYPE_DOUBLE; + } else if ((upb_float_handlerfunc*)f == upb_shim_setfloat) { + *type = UPB_TYPE_FLOAT; + } else if ((upb_bool_handlerfunc*)f == upb_shim_setbool) { + *type = UPB_TYPE_BOOL; + } else { + return NULL; + } + + return (const upb_shim_data*)upb_handlers_gethandlerdata(h, s); +} + + +#include <string.h> + +static void upb_symtab_free(upb_refcounted *r) { + upb_symtab *s = (upb_symtab*)r; + upb_strtable_iter i; + upb_strtable_begin(&i, &s->symtab); + for (; !upb_strtable_done(&i); upb_strtable_next(&i)) { + const upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i)); + upb_def_unref(def, s); + } + upb_strtable_uninit(&s->symtab); + upb_gfree(s); +} + +upb_symtab *upb_symtab_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {NULL, &upb_symtab_free}; + + upb_symtab *s = upb_gmalloc(sizeof(*s)); + if (!s) { + return NULL; + } + + upb_refcounted_init(upb_symtab_upcast_mutable(s), &vtbl, owner); + upb_strtable_init(&s->symtab, UPB_CTYPE_PTR); + return s; +} + +void upb_symtab_freeze(upb_symtab *s) { + upb_refcounted *r; + bool ok; + + assert(!upb_symtab_isfrozen(s)); + r = upb_symtab_upcast_mutable(s); + /* The symtab does not take ref2's (see refcounted.h) on the defs, because + * defs cannot refer back to the table and therefore cannot create cycles. So + * 0 will suffice for maxdepth here. */ + ok = upb_refcounted_freeze(&r, 1, NULL, 0); + UPB_ASSERT_VAR(ok, ok); +} + +const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *ret = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return ret; +} + +const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return def ? upb_dyncast_msgdef(def) : NULL; +} + +const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return def ? upb_dyncast_enumdef(def) : NULL; +} + +/* Given a symbol and the base symbol inside which it is defined, find the + * symbol's definition in t. */ +static upb_def *upb_resolvename(const upb_strtable *t, + const char *base, const char *sym) { + if(strlen(sym) == 0) return NULL; + if(sym[0] == '.') { + /* Symbols starting with '.' are absolute, so we do a single lookup. + * Slice to omit the leading '.' */ + upb_value v; + return upb_strtable_lookup(t, sym + 1, &v) ? upb_value_getptr(v) : NULL; + } else { + /* Remove components from base until we find an entry or run out. + * TODO: This branch is totally broken, but currently not used. */ + (void)base; + assert(false); + return NULL; + } +} + +const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base, + const char *sym) { + upb_def *ret = upb_resolvename(&s->symtab, base, sym); + return ret; +} + +/* Starts a depth-first traversal at "def", recursing into any subdefs + * (ie. submessage types). Adds duplicates of existing defs to addtab + * wherever necessary, so that the resulting symtab will be consistent once + * addtab is added. + * + * More specifically, if any def D is found in the DFS that: + * + * 1. can reach a def that is being replaced by something in addtab, AND + * + * 2. is not itself being replaced already (ie. this name doesn't already + * exist in addtab) + * + * ...then a duplicate (new copy) of D will be added to addtab. + * + * Returns true if this happened for any def reachable from "def." + * + * It is slightly tricky to do this correctly in the presence of cycles. If we + * detect that our DFS has hit a cycle, we might not yet know if any SCCs on + * our stack can reach a def in addtab or not. Once we figure this out, that + * answer needs to apply to *all* defs in these SCCs, even if we visited them + * already. So a straight up one-pass cycle-detecting DFS won't work. + * + * To work around this problem, we traverse each SCC (which we already + * computed, since these defs are frozen) as a single node. We first compute + * whether the SCC as a whole can reach any def in addtab, then we dup (or not) + * the entire SCC. This requires breaking the encapsulation of upb_refcounted, + * since that is where we get the data about what SCC we are in. */ +static bool upb_resolve_dfs(const upb_def *def, upb_strtable *addtab, + const void *new_owner, upb_inttable *seen, + upb_status *s) { + upb_value v; + bool need_dup; + const upb_def *base; + const void* memoize_key; + + /* Memoize results of this function for efficiency (since we're traversing a + * DAG this is not needed to limit the depth of the search). + * + * We memoize by SCC instead of by individual def. */ + memoize_key = def->base.group; + + if (upb_inttable_lookupptr(seen, memoize_key, &v)) + return upb_value_getbool(v); + + /* Visit submessages for all messages in the SCC. */ + need_dup = false; + base = def; + do { + upb_value v; + const upb_msgdef *m; + + assert(upb_def_isfrozen(def)); + if (def->type == UPB_DEF_FIELD) continue; + if (upb_strtable_lookup(addtab, upb_def_fullname(def), &v)) { + need_dup = true; + } + + /* For messages, continue the recursion by visiting all subdefs, but only + * ones in different SCCs. */ + m = upb_dyncast_msgdef(def); + if (m) { + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + const upb_def *subdef; + + if (!upb_fielddef_hassubdef(f)) continue; + subdef = upb_fielddef_subdef(f); + + /* Skip subdefs in this SCC. */ + if (def->base.group == subdef->base.group) continue; + + /* |= to avoid short-circuit; we need its side-effects. */ + need_dup |= upb_resolve_dfs(subdef, addtab, new_owner, seen, s); + if (!upb_ok(s)) return false; + } + } + } while ((def = (upb_def*)def->base.next) != base); + + if (need_dup) { + /* Dup all defs in this SCC that don't already have entries in addtab. */ + def = base; + do { + const char *name; + + if (def->type == UPB_DEF_FIELD) continue; + name = upb_def_fullname(def); + if (!upb_strtable_lookup(addtab, name, NULL)) { + upb_def *newdef = upb_def_dup(def, new_owner); + if (!newdef) goto oom; + newdef->came_from_user = false; + if (!upb_strtable_insert(addtab, name, upb_value_ptr(newdef))) + goto oom; + } + } while ((def = (upb_def*)def->base.next) != base); + } + + upb_inttable_insertptr(seen, memoize_key, upb_value_bool(need_dup)); + return need_dup; + +oom: + upb_status_seterrmsg(s, "out of memory"); + return false; +} + +/* TODO(haberman): we need a lot more testing of error conditions. + * The came_from_user stuff in particular is not tested. */ +static bool symtab_add(upb_symtab *s, upb_def *const*defs, size_t n, + void *ref_donor, upb_refcounted *freeze_also, + upb_status *status) { + size_t i; + size_t add_n; + size_t freeze_n; + upb_strtable_iter iter; + upb_refcounted **add_objs = NULL; + upb_def **add_defs = NULL; + size_t add_objs_size; + upb_strtable addtab; + upb_inttable seen; + + if (n == 0 && !freeze_also) { + return true; + } + + assert(!upb_symtab_isfrozen(s)); + if (!upb_strtable_init(&addtab, UPB_CTYPE_PTR)) { + upb_status_seterrmsg(status, "out of memory"); + return false; + } + + /* Add new defs to our "add" set. */ + for (i = 0; i < n; i++) { + upb_def *def = defs[i]; + const char *fullname; + upb_fielddef *f; + + if (upb_def_isfrozen(def)) { + upb_status_seterrmsg(status, "added defs must be mutable"); + goto err; + } + assert(!upb_def_isfrozen(def)); + fullname = upb_def_fullname(def); + if (!fullname) { + upb_status_seterrmsg( + status, "Anonymous defs cannot be added to a symtab"); + goto err; + } + + f = upb_dyncast_fielddef_mutable(def); + + if (f) { + if (!upb_fielddef_containingtypename(f)) { + upb_status_seterrmsg(status, + "Standalone fielddefs must have a containing type " + "(extendee) name set"); + goto err; + } + } else { + if (upb_strtable_lookup(&addtab, fullname, NULL)) { + upb_status_seterrf(status, "Conflicting defs named '%s'", fullname); + goto err; + } + /* We need this to back out properly, because if there is a failure we + * need to donate the ref back to the caller. */ + def->came_from_user = true; + upb_def_donateref(def, ref_donor, s); + if (!upb_strtable_insert(&addtab, fullname, upb_value_ptr(def))) + goto oom_err; + } + } + + /* Add standalone fielddefs (ie. extensions) to the appropriate messages. + * If the appropriate message only exists in the existing symtab, duplicate + * it so we have a mutable copy we can add the fields to. */ + for (i = 0; i < n; i++) { + upb_def *def = defs[i]; + upb_fielddef *f = upb_dyncast_fielddef_mutable(def); + const char *msgname; + upb_value v; + upb_msgdef *m; + + if (!f) continue; + msgname = upb_fielddef_containingtypename(f); + /* We validated this earlier in this function. */ + assert(msgname); + + /* If the extendee name is absolutely qualified, move past the initial ".". + * TODO(haberman): it is not obvious what it would mean if this was not + * absolutely qualified. */ + if (msgname[0] == '.') { + msgname++; + } + + if (upb_strtable_lookup(&addtab, msgname, &v)) { + /* Extendee is in the set of defs the user asked us to add. */ + m = upb_value_getptr(v); + } else { + /* Need to find and dup the extendee from the existing symtab. */ + const upb_msgdef *frozen_m = upb_symtab_lookupmsg(s, msgname); + if (!frozen_m) { + upb_status_seterrf(status, + "Tried to extend message %s that does not exist " + "in this SymbolTable.", + msgname); + goto err; + } + m = upb_msgdef_dup(frozen_m, s); + if (!m) goto oom_err; + if (!upb_strtable_insert(&addtab, msgname, upb_value_ptr(m))) { + upb_msgdef_unref(m, s); + goto oom_err; + } + } + + if (!upb_msgdef_addfield(m, f, ref_donor, status)) { + goto err; + } + } + + /* Add dups of any existing def that can reach a def with the same name as + * anything in our "add" set. */ + if (!upb_inttable_init(&seen, UPB_CTYPE_BOOL)) goto oom_err; + upb_strtable_begin(&iter, &s->symtab); + for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) { + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter)); + upb_resolve_dfs(def, &addtab, s, &seen, status); + if (!upb_ok(status)) goto err; + } + upb_inttable_uninit(&seen); + + /* Now using the table, resolve symbolic references for subdefs. */ + upb_strtable_begin(&iter, &addtab); + for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) { + const char *base; + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter)); + upb_msgdef *m = upb_dyncast_msgdef_mutable(def); + upb_msg_field_iter j; + + if (!m) continue; + /* Type names are resolved relative to the message in which they appear. */ + base = upb_msgdef_fullname(m); + + for(upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + upb_fielddef *f = upb_msg_iter_field(&j); + const char *name = upb_fielddef_subdefname(f); + if (name && !upb_fielddef_subdef(f)) { + /* Try the lookup in the current set of to-be-added defs first. If not + * there, try existing defs. */ + upb_def *subdef = upb_resolvename(&addtab, base, name); + if (subdef == NULL) { + subdef = upb_resolvename(&s->symtab, base, name); + } + if (subdef == NULL) { + upb_status_seterrf( + status, "couldn't resolve name '%s' in message '%s'", name, base); + goto err; + } else if (!upb_fielddef_setsubdef(f, subdef, status)) { + goto err; + } + } + } + } + + /* We need an array of the defs in addtab, for passing to + * upb_refcounted_freeze(). */ + add_objs_size = upb_strtable_count(&addtab); + if (freeze_also) { + add_objs_size++; + } + + add_defs = upb_gmalloc(sizeof(void*) * add_objs_size); + if (add_defs == NULL) goto oom_err; + upb_strtable_begin(&iter, &addtab); + for (add_n = 0; !upb_strtable_done(&iter); upb_strtable_next(&iter)) { + add_defs[add_n++] = upb_value_getptr(upb_strtable_iter_value(&iter)); + } + + /* Validate defs. */ + if (!_upb_def_validate(add_defs, add_n, status)) { + goto err; + } + + /* Cheat a little and give the array a new type. + * This is probably undefined behavior, but this code will be deleted soon. */ + add_objs = (upb_refcounted**)add_defs; + + freeze_n = add_n; + if (freeze_also) { + add_objs[freeze_n++] = freeze_also; + } + + if (!upb_refcounted_freeze(add_objs, freeze_n, status, + UPB_MAX_MESSAGE_DEPTH * 2)) { + goto err; + } + + /* This must be delayed until all errors have been detected, since error + * recovery code uses this table to cleanup defs. */ + upb_strtable_uninit(&addtab); + + /* TODO(haberman) we don't properly handle errors after this point (like + * OOM in upb_strtable_insert() below). */ + for (i = 0; i < add_n; i++) { + upb_def *def = (upb_def*)add_objs[i]; + const char *name = upb_def_fullname(def); + upb_value v; + bool success; + + if (upb_strtable_remove(&s->symtab, name, &v)) { + const upb_def *def = upb_value_getptr(v); + upb_def_unref(def, s); + } + success = upb_strtable_insert(&s->symtab, name, upb_value_ptr(def)); + UPB_ASSERT_VAR(success, success == true); + } + upb_gfree(add_defs); + return true; + +oom_err: + upb_status_seterrmsg(status, "out of memory"); +err: { + /* For defs the user passed in, we need to donate the refs back. For defs + * we dup'd, we need to just unref them. */ + upb_strtable_begin(&iter, &addtab); + for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) { + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter)); + bool came_from_user = def->came_from_user; + def->came_from_user = false; + if (came_from_user) { + upb_def_donateref(def, s, ref_donor); + } else { + upb_def_unref(def, s); + } + } + } + upb_strtable_uninit(&addtab); + upb_gfree(add_defs); + assert(!upb_ok(status)); + return false; +} + +bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, size_t n, + void *ref_donor, upb_status *status) { + return symtab_add(s, defs, n, ref_donor, NULL, status); +} + +bool upb_symtab_addfile(upb_symtab *s, upb_filedef *file, upb_status *status) { + size_t n; + size_t i; + upb_def **defs; + bool ret; + + n = upb_filedef_defcount(file); + defs = upb_gmalloc(sizeof(*defs) * n); + + if (defs == NULL) { + upb_status_seterrmsg(status, "Out of memory"); + return false; + } + + for (i = 0; i < n; i++) { + defs[i] = upb_filedef_mutabledef(file, i); + } + + ret = symtab_add(s, defs, n, NULL, upb_filedef_upcast_mutable(file), status); + + upb_gfree(defs); + return ret; +} + +/* Iteration. */ + +static void advance_to_matching(upb_symtab_iter *iter) { + if (iter->type == UPB_DEF_ANY) + return; + + while (!upb_strtable_done(&iter->iter) && + iter->type != upb_symtab_iter_def(iter)->type) { + upb_strtable_next(&iter->iter); + } +} + +void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s, + upb_deftype_t type) { + upb_strtable_begin(&iter->iter, &s->symtab); + iter->type = type; + advance_to_matching(iter); +} + +void upb_symtab_next(upb_symtab_iter *iter) { + upb_strtable_next(&iter->iter); + advance_to_matching(iter); +} + +bool upb_symtab_done(const upb_symtab_iter *iter) { + return upb_strtable_done(&iter->iter); +} + +const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter) { + return upb_value_getptr(upb_strtable_iter_value(&iter->iter)); +} +/* +** upb_table Implementation +** +** Implementation is heavily inspired by Lua's ltable.c. +*/ + + +#include <string.h> + +#define UPB_MAXARRSIZE 16 /* 64k. */ + +/* From Chromium. */ +#define ARRAY_SIZE(x) \ + ((sizeof(x)/sizeof(0[x])) / ((size_t)(!(sizeof(x) % sizeof(0[x]))))) + +#ifdef NDEBUG +static void upb_check_alloc(upb_table *t, upb_alloc *a) { + UPB_UNUSED(t); + UPB_UNUSED(a); +} +#else +static void upb_check_alloc(upb_table *t, upb_alloc *a) { + assert(t->alloc == a); +} +#endif + +static const double MAX_LOAD = 0.85; + +/* The minimum utilization of the array part of a mixed hash/array table. This + * is a speed/memory-usage tradeoff (though it's not straightforward because of + * cache effects). The lower this is, the more memory we'll use. */ +static const double MIN_DENSITY = 0.1; + +bool is_pow2(uint64_t v) { return v == 0 || (v & (v - 1)) == 0; } + +int log2ceil(uint64_t v) { + int ret = 0; + bool pow2 = is_pow2(v); + while (v >>= 1) ret++; + ret = pow2 ? ret : ret + 1; /* Ceiling. */ + return UPB_MIN(UPB_MAXARRSIZE, ret); +} + +char *upb_strdup(const char *s, upb_alloc *a) { + return upb_strdup2(s, strlen(s), a); +} + +char *upb_strdup2(const char *s, size_t len, upb_alloc *a) { + size_t n; + char *p; + + /* Prevent overflow errors. */ + if (len == SIZE_MAX) return NULL; + /* Always null-terminate, even if binary data; but don't rely on the input to + * have a null-terminating byte since it may be a raw binary buffer. */ + n = len + 1; + p = upb_malloc(a, n); + if (p) { + memcpy(p, s, len); + p[len] = 0; + } + return p; +} + +/* A type to represent the lookup key of either a strtable or an inttable. */ +typedef union { + uintptr_t num; + struct { + const char *str; + size_t len; + } str; +} lookupkey_t; + +static lookupkey_t strkey2(const char *str, size_t len) { + lookupkey_t k; + k.str.str = str; + k.str.len = len; + return k; +} + +static lookupkey_t intkey(uintptr_t key) { + lookupkey_t k; + k.num = key; + return k; +} + +typedef uint32_t hashfunc_t(upb_tabkey key); +typedef bool eqlfunc_t(upb_tabkey k1, lookupkey_t k2); + +/* Base table (shared code) ***************************************************/ + +/* For when we need to cast away const. */ +static upb_tabent *mutable_entries(upb_table *t) { + return (upb_tabent*)t->entries; +} + +static bool isfull(upb_table *t) { + if (upb_table_size(t) == 0) { + return true; + } else { + return ((double)(t->count + 1) / upb_table_size(t)) > MAX_LOAD; + } +} + +static bool init(upb_table *t, upb_ctype_t ctype, uint8_t size_lg2, + upb_alloc *a) { + size_t bytes; + + t->count = 0; + t->ctype = ctype; + t->size_lg2 = size_lg2; + t->mask = upb_table_size(t) ? upb_table_size(t) - 1 : 0; +#ifndef NDEBUG + t->alloc = a; +#endif + bytes = upb_table_size(t) * sizeof(upb_tabent); + if (bytes > 0) { + t->entries = upb_malloc(a, bytes); + if (!t->entries) return false; + memset(mutable_entries(t), 0, bytes); + } else { + t->entries = NULL; + } + return true; +} + +static void uninit(upb_table *t, upb_alloc *a) { + upb_check_alloc(t, a); + upb_free(a, mutable_entries(t)); +} + +static upb_tabent *emptyent(upb_table *t) { + upb_tabent *e = mutable_entries(t) + upb_table_size(t); + while (1) { if (upb_tabent_isempty(--e)) return e; assert(e > t->entries); } +} + +static upb_tabent *getentry_mutable(upb_table *t, uint32_t hash) { + return (upb_tabent*)upb_getentry(t, hash); +} + +static const upb_tabent *findentry(const upb_table *t, lookupkey_t key, + uint32_t hash, eqlfunc_t *eql) { + const upb_tabent *e; + + if (t->size_lg2 == 0) return NULL; + e = upb_getentry(t, hash); + if (upb_tabent_isempty(e)) return NULL; + while (1) { + if (eql(e->key, key)) return e; + if ((e = e->next) == NULL) return NULL; + } +} + +static upb_tabent *findentry_mutable(upb_table *t, lookupkey_t key, + uint32_t hash, eqlfunc_t *eql) { + return (upb_tabent*)findentry(t, key, hash, eql); +} + +static bool lookup(const upb_table *t, lookupkey_t key, upb_value *v, + uint32_t hash, eqlfunc_t *eql) { + const upb_tabent *e = findentry(t, key, hash, eql); + if (e) { + if (v) { + _upb_value_setval(v, e->val.val, t->ctype); + } + return true; + } else { + return false; + } +} + +/* The given key must not already exist in the table. */ +static void insert(upb_table *t, lookupkey_t key, upb_tabkey tabkey, + upb_value val, uint32_t hash, + hashfunc_t *hashfunc, eqlfunc_t *eql) { + upb_tabent *mainpos_e; + upb_tabent *our_e; + + UPB_UNUSED(eql); + UPB_UNUSED(key); + assert(findentry(t, key, hash, eql) == NULL); + assert(val.ctype == t->ctype); + + t->count++; + mainpos_e = getentry_mutable(t, hash); + our_e = mainpos_e; + + if (upb_tabent_isempty(mainpos_e)) { + /* Our main position is empty; use it. */ + our_e->next = NULL; + } else { + /* Collision. */ + upb_tabent *new_e = emptyent(t); + /* Head of collider's chain. */ + upb_tabent *chain = getentry_mutable(t, hashfunc(mainpos_e->key)); + if (chain == mainpos_e) { + /* Existing ent is in its main posisiton (it has the same hash as us, and + * is the head of our chain). Insert to new ent and append to this chain. */ + new_e->next = mainpos_e->next; + mainpos_e->next = new_e; + our_e = new_e; + } else { + /* Existing ent is not in its main position (it is a node in some other + * chain). This implies that no existing ent in the table has our hash. + * Evict it (updating its chain) and use its ent for head of our chain. */ + *new_e = *mainpos_e; /* copies next. */ + while (chain->next != mainpos_e) { + chain = (upb_tabent*)chain->next; + assert(chain); + } + chain->next = new_e; + our_e = mainpos_e; + our_e->next = NULL; + } + } + our_e->key = tabkey; + our_e->val.val = val.val; + assert(findentry(t, key, hash, eql) == our_e); +} + +static bool rm(upb_table *t, lookupkey_t key, upb_value *val, + upb_tabkey *removed, uint32_t hash, eqlfunc_t *eql) { + upb_tabent *chain = getentry_mutable(t, hash); + if (upb_tabent_isempty(chain)) return false; + if (eql(chain->key, key)) { + /* Element to remove is at the head of its chain. */ + t->count--; + if (val) { + _upb_value_setval(val, chain->val.val, t->ctype); + } + if (chain->next) { + upb_tabent *move = (upb_tabent*)chain->next; + *chain = *move; + if (removed) *removed = move->key; + move->key = 0; /* Make the slot empty. */ + } else { + if (removed) *removed = chain->key; + chain->key = 0; /* Make the slot empty. */ + } + return true; + } else { + /* Element to remove is either in a non-head position or not in the + * table. */ + while (chain->next && !eql(chain->next->key, key)) + chain = (upb_tabent*)chain->next; + if (chain->next) { + /* Found element to remove. */ + upb_tabent *rm; + + if (val) { + _upb_value_setval(val, chain->next->val.val, t->ctype); + } + rm = (upb_tabent*)chain->next; + if (removed) *removed = rm->key; + rm->key = 0; + chain->next = rm->next; + t->count--; + return true; + } else { + return false; + } + } +} + +static size_t next(const upb_table *t, size_t i) { + do { + if (++i >= upb_table_size(t)) + return SIZE_MAX; + } while(upb_tabent_isempty(&t->entries[i])); + + return i; +} + +static size_t begin(const upb_table *t) { + return next(t, -1); +} + + +/* upb_strtable ***************************************************************/ + +/* A simple "subclass" of upb_table that only adds a hash function for strings. */ + +static upb_tabkey strcopy(lookupkey_t k2, upb_alloc *a) { + char *str = upb_malloc(a, k2.str.len + sizeof(uint32_t) + 1); + if (str == NULL) return 0; + memcpy(str, &k2.str.len, sizeof(uint32_t)); + memcpy(str + sizeof(uint32_t), k2.str.str, k2.str.len + 1); + return (uintptr_t)str; +} + +static uint32_t strhash(upb_tabkey key) { + uint32_t len; + char *str = upb_tabstr(key, &len); + return MurmurHash2(str, len, 0); +} + +static bool streql(upb_tabkey k1, lookupkey_t k2) { + uint32_t len; + char *str = upb_tabstr(k1, &len); + return len == k2.str.len && memcmp(str, k2.str.str, len) == 0; +} + +bool upb_strtable_init2(upb_strtable *t, upb_ctype_t ctype, upb_alloc *a) { + return init(&t->t, ctype, 2, a); +} + +void upb_strtable_uninit2(upb_strtable *t, upb_alloc *a) { + size_t i; + for (i = 0; i < upb_table_size(&t->t); i++) + upb_free(a, (void*)t->t.entries[i].key); + uninit(&t->t, a); +} + +bool upb_strtable_resize(upb_strtable *t, size_t size_lg2, upb_alloc *a) { + upb_strtable new_table; + upb_strtable_iter i; + + upb_check_alloc(&t->t, a); + + if (!init(&new_table.t, t->t.ctype, size_lg2, a)) + return false; + upb_strtable_begin(&i, t); + for ( ; !upb_strtable_done(&i); upb_strtable_next(&i)) { + upb_strtable_insert3( + &new_table, + upb_strtable_iter_key(&i), + upb_strtable_iter_keylength(&i), + upb_strtable_iter_value(&i), + a); + } + upb_strtable_uninit2(t, a); + *t = new_table; + return true; +} + +bool upb_strtable_insert3(upb_strtable *t, const char *k, size_t len, + upb_value v, upb_alloc *a) { + lookupkey_t key; + upb_tabkey tabkey; + uint32_t hash; + + upb_check_alloc(&t->t, a); + + if (isfull(&t->t)) { + /* Need to resize. New table of double the size, add old elements to it. */ + if (!upb_strtable_resize(t, t->t.size_lg2 + 1, a)) { + return false; + } + } + + key = strkey2(k, len); + tabkey = strcopy(key, a); + if (tabkey == 0) return false; + + hash = MurmurHash2(key.str.str, key.str.len, 0); + insert(&t->t, key, tabkey, v, hash, &strhash, &streql); + return true; +} + +bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len, + upb_value *v) { + uint32_t hash = MurmurHash2(key, len, 0); + return lookup(&t->t, strkey2(key, len), v, hash, &streql); +} + +bool upb_strtable_remove3(upb_strtable *t, const char *key, size_t len, + upb_value *val, upb_alloc *alloc) { + uint32_t hash = MurmurHash2(key, strlen(key), 0); + upb_tabkey tabkey; + if (rm(&t->t, strkey2(key, len), val, &tabkey, hash, &streql)) { + upb_free(alloc, (void*)tabkey); + return true; + } else { + return false; + } +} + +/* Iteration */ + +static const upb_tabent *str_tabent(const upb_strtable_iter *i) { + return &i->t->t.entries[i->index]; +} + +void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t) { + i->t = t; + i->index = begin(&t->t); +} + +void upb_strtable_next(upb_strtable_iter *i) { + i->index = next(&i->t->t, i->index); +} + +bool upb_strtable_done(const upb_strtable_iter *i) { + return i->index >= upb_table_size(&i->t->t) || + upb_tabent_isempty(str_tabent(i)); +} + +const char *upb_strtable_iter_key(const upb_strtable_iter *i) { + assert(!upb_strtable_done(i)); + return upb_tabstr(str_tabent(i)->key, NULL); +} + +size_t upb_strtable_iter_keylength(const upb_strtable_iter *i) { + uint32_t len; + assert(!upb_strtable_done(i)); + upb_tabstr(str_tabent(i)->key, &len); + return len; +} + +upb_value upb_strtable_iter_value(const upb_strtable_iter *i) { + assert(!upb_strtable_done(i)); + return _upb_value_val(str_tabent(i)->val.val, i->t->t.ctype); +} + +void upb_strtable_iter_setdone(upb_strtable_iter *i) { + i->index = SIZE_MAX; +} + +bool upb_strtable_iter_isequal(const upb_strtable_iter *i1, + const upb_strtable_iter *i2) { + if (upb_strtable_done(i1) && upb_strtable_done(i2)) + return true; + return i1->t == i2->t && i1->index == i2->index; +} + + +/* upb_inttable ***************************************************************/ + +/* For inttables we use a hybrid structure where small keys are kept in an + * array and large keys are put in the hash table. */ + +static uint32_t inthash(upb_tabkey key) { return upb_inthash(key); } + +static bool inteql(upb_tabkey k1, lookupkey_t k2) { + return k1 == k2.num; +} + +static upb_tabval *mutable_array(upb_inttable *t) { + return (upb_tabval*)t->array; +} + +static upb_tabval *inttable_val(upb_inttable *t, uintptr_t key) { + if (key < t->array_size) { + return upb_arrhas(t->array[key]) ? &(mutable_array(t)[key]) : NULL; + } else { + upb_tabent *e = + findentry_mutable(&t->t, intkey(key), upb_inthash(key), &inteql); + return e ? &e->val : NULL; + } +} + +static const upb_tabval *inttable_val_const(const upb_inttable *t, + uintptr_t key) { + return inttable_val((upb_inttable*)t, key); +} + +size_t upb_inttable_count(const upb_inttable *t) { + return t->t.count + t->array_count; +} + +static void check(upb_inttable *t) { + UPB_UNUSED(t); +#if defined(UPB_DEBUG_TABLE) && !defined(NDEBUG) + { + /* This check is very expensive (makes inserts/deletes O(N)). */ + size_t count = 0; + upb_inttable_iter i; + upb_inttable_begin(&i, t); + for(; !upb_inttable_done(&i); upb_inttable_next(&i), count++) { + assert(upb_inttable_lookup(t, upb_inttable_iter_key(&i), NULL)); + } + assert(count == upb_inttable_count(t)); + } +#endif +} + +bool upb_inttable_sizedinit(upb_inttable *t, upb_ctype_t ctype, + size_t asize, int hsize_lg2, upb_alloc *a) { + size_t array_bytes; + + if (!init(&t->t, ctype, hsize_lg2, a)) return false; + /* Always make the array part at least 1 long, so that we know key 0 + * won't be in the hash part, which simplifies things. */ + t->array_size = UPB_MAX(1, asize); + t->array_count = 0; + array_bytes = t->array_size * sizeof(upb_value); + t->array = upb_malloc(a, array_bytes); + if (!t->array) { + uninit(&t->t, a); + return false; + } + memset(mutable_array(t), 0xff, array_bytes); + check(t); + return true; +} + +bool upb_inttable_init2(upb_inttable *t, upb_ctype_t ctype, upb_alloc *a) { + return upb_inttable_sizedinit(t, ctype, 0, 4, a); +} + +void upb_inttable_uninit2(upb_inttable *t, upb_alloc *a) { + uninit(&t->t, a); + upb_free(a, mutable_array(t)); +} + +bool upb_inttable_insert2(upb_inttable *t, uintptr_t key, upb_value val, + upb_alloc *a) { + upb_tabval tabval; + tabval.val = val.val; + UPB_UNUSED(tabval); + assert(upb_arrhas(tabval)); /* This will reject (uint64_t)-1. Fix this. */ + + upb_check_alloc(&t->t, a); + + if (key < t->array_size) { + assert(!upb_arrhas(t->array[key])); + t->array_count++; + mutable_array(t)[key].val = val.val; + } else { + if (isfull(&t->t)) { + /* Need to resize the hash part, but we re-use the array part. */ + size_t i; + upb_table new_table; + + if (!init(&new_table, t->t.ctype, t->t.size_lg2 + 1, a)) { + return false; + } + + for (i = begin(&t->t); i < upb_table_size(&t->t); i = next(&t->t, i)) { + const upb_tabent *e = &t->t.entries[i]; + uint32_t hash; + upb_value v; + + _upb_value_setval(&v, e->val.val, t->t.ctype); + hash = upb_inthash(e->key); + insert(&new_table, intkey(e->key), e->key, v, hash, &inthash, &inteql); + } + + assert(t->t.count == new_table.count); + + uninit(&t->t, a); + t->t = new_table; + } + insert(&t->t, intkey(key), key, val, upb_inthash(key), &inthash, &inteql); + } + check(t); + return true; +} + +bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v) { + const upb_tabval *table_v = inttable_val_const(t, key); + if (!table_v) return false; + if (v) _upb_value_setval(v, table_v->val, t->t.ctype); + return true; +} + +bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val) { + upb_tabval *table_v = inttable_val(t, key); + if (!table_v) return false; + table_v->val = val.val; + return true; +} + +bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val) { + bool success; + if (key < t->array_size) { + if (upb_arrhas(t->array[key])) { + upb_tabval empty = UPB_TABVALUE_EMPTY_INIT; + t->array_count--; + if (val) { + _upb_value_setval(val, t->array[key].val, t->t.ctype); + } + mutable_array(t)[key] = empty; + success = true; + } else { + success = false; + } + } else { + upb_tabkey removed; + uint32_t hash = upb_inthash(key); + success = rm(&t->t, intkey(key), val, &removed, hash, &inteql); + } + check(t); + return success; +} + +bool upb_inttable_push2(upb_inttable *t, upb_value val, upb_alloc *a) { + upb_check_alloc(&t->t, a); + return upb_inttable_insert2(t, upb_inttable_count(t), val, a); +} + +upb_value upb_inttable_pop(upb_inttable *t) { + upb_value val; + bool ok = upb_inttable_remove(t, upb_inttable_count(t) - 1, &val); + UPB_ASSERT_VAR(ok, ok); + return val; +} + +bool upb_inttable_insertptr2(upb_inttable *t, const void *key, upb_value val, + upb_alloc *a) { + upb_check_alloc(&t->t, a); + return upb_inttable_insert2(t, (uintptr_t)key, val, a); +} + +bool upb_inttable_lookupptr(const upb_inttable *t, const void *key, + upb_value *v) { + return upb_inttable_lookup(t, (uintptr_t)key, v); +} + +bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val) { + return upb_inttable_remove(t, (uintptr_t)key, val); +} + +void upb_inttable_compact2(upb_inttable *t, upb_alloc *a) { + /* A power-of-two histogram of the table keys. */ + size_t counts[UPB_MAXARRSIZE + 1] = {0}; + + /* The max key in each bucket. */ + uintptr_t max[UPB_MAXARRSIZE + 1] = {0}; + + upb_inttable_iter i; + size_t arr_count; + int size_lg2; + upb_inttable new_t; + + upb_check_alloc(&t->t, a); + + upb_inttable_begin(&i, t); + for (; !upb_inttable_done(&i); upb_inttable_next(&i)) { + uintptr_t key = upb_inttable_iter_key(&i); + int bucket = log2ceil(key); + max[bucket] = UPB_MAX(max[bucket], key); + counts[bucket]++; + } + + /* Find the largest power of two that satisfies the MIN_DENSITY + * definition (while actually having some keys). */ + arr_count = upb_inttable_count(t); + + for (size_lg2 = ARRAY_SIZE(counts) - 1; size_lg2 > 0; size_lg2--) { + if (counts[size_lg2] == 0) { + /* We can halve again without losing any entries. */ + continue; + } else if (arr_count >= (1 << size_lg2) * MIN_DENSITY) { + break; + } + + arr_count -= counts[size_lg2]; + } + + assert(arr_count <= upb_inttable_count(t)); + + { + /* Insert all elements into new, perfectly-sized table. */ + size_t arr_size = max[size_lg2] + 1; /* +1 so arr[max] will fit. */ + size_t hash_count = upb_inttable_count(t) - arr_count; + size_t hash_size = hash_count ? (hash_count / MAX_LOAD) + 1 : 0; + size_t hashsize_lg2 = log2ceil(hash_size); + + upb_inttable_sizedinit(&new_t, t->t.ctype, arr_size, hashsize_lg2, a); + upb_inttable_begin(&i, t); + for (; !upb_inttable_done(&i); upb_inttable_next(&i)) { + uintptr_t k = upb_inttable_iter_key(&i); + upb_inttable_insert2(&new_t, k, upb_inttable_iter_value(&i), a); + } + assert(new_t.array_size == arr_size); + assert(new_t.t.size_lg2 == hashsize_lg2); + } + upb_inttable_uninit2(t, a); + *t = new_t; +} + +/* Iteration. */ + +static const upb_tabent *int_tabent(const upb_inttable_iter *i) { + assert(!i->array_part); + return &i->t->t.entries[i->index]; +} + +static upb_tabval int_arrent(const upb_inttable_iter *i) { + assert(i->array_part); + return i->t->array[i->index]; +} + +void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t) { + i->t = t; + i->index = -1; + i->array_part = true; + upb_inttable_next(i); +} + +void upb_inttable_next(upb_inttable_iter *iter) { + const upb_inttable *t = iter->t; + if (iter->array_part) { + while (++iter->index < t->array_size) { + if (upb_arrhas(int_arrent(iter))) { + return; + } + } + iter->array_part = false; + iter->index = begin(&t->t); + } else { + iter->index = next(&t->t, iter->index); + } +} + +bool upb_inttable_done(const upb_inttable_iter *i) { + if (i->array_part) { + return i->index >= i->t->array_size || + !upb_arrhas(int_arrent(i)); + } else { + return i->index >= upb_table_size(&i->t->t) || + upb_tabent_isempty(int_tabent(i)); + } +} + +uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i) { + assert(!upb_inttable_done(i)); + return i->array_part ? i->index : int_tabent(i)->key; +} + +upb_value upb_inttable_iter_value(const upb_inttable_iter *i) { + assert(!upb_inttable_done(i)); + return _upb_value_val( + i->array_part ? i->t->array[i->index].val : int_tabent(i)->val.val, + i->t->t.ctype); +} + +void upb_inttable_iter_setdone(upb_inttable_iter *i) { + i->index = SIZE_MAX; + i->array_part = false; +} + +bool upb_inttable_iter_isequal(const upb_inttable_iter *i1, + const upb_inttable_iter *i2) { + if (upb_inttable_done(i1) && upb_inttable_done(i2)) + return true; + return i1->t == i2->t && i1->index == i2->index && + i1->array_part == i2->array_part; +} + +#ifdef UPB_UNALIGNED_READS_OK +/* ----------------------------------------------------------------------------- + * MurmurHash2, by Austin Appleby (released as public domain). + * Reformatted and C99-ified by Joshua Haberman. + * Note - This code makes a few assumptions about how your machine behaves - + * 1. We can read a 4-byte value from any address without crashing + * 2. sizeof(int) == 4 (in upb this limitation is removed by using uint32_t + * And it has a few limitations - + * 1. It will not work incrementally. + * 2. It will not produce the same results on little-endian and big-endian + * machines. */ +uint32_t MurmurHash2(const void *key, size_t len, uint32_t seed) { + /* 'm' and 'r' are mixing constants generated offline. + * They're not really 'magic', they just happen to work well. */ + const uint32_t m = 0x5bd1e995; + const int32_t r = 24; + + /* Initialize the hash to a 'random' value */ + uint32_t h = seed ^ len; + + /* Mix 4 bytes at a time into the hash */ + const uint8_t * data = (const uint8_t *)key; + while(len >= 4) { + uint32_t k = *(uint32_t *)data; + + k *= m; + k ^= k >> r; + k *= m; + + h *= m; + h ^= k; + + data += 4; + len -= 4; + } + + /* Handle the last few bytes of the input array */ + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + + /* Do a few final mixes of the hash to ensure the last few + * bytes are well-incorporated. */ + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; +} + +#else /* !UPB_UNALIGNED_READS_OK */ + +/* ----------------------------------------------------------------------------- + * MurmurHashAligned2, by Austin Appleby + * Same algorithm as MurmurHash2, but only does aligned reads - should be safer + * on certain platforms. + * Performance will be lower than MurmurHash2 */ + +#define MIX(h,k,m) { k *= m; k ^= k >> r; k *= m; h *= m; h ^= k; } + +uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed) { + const uint32_t m = 0x5bd1e995; + const int32_t r = 24; + const uint8_t * data = (const uint8_t *)key; + uint32_t h = seed ^ len; + uint8_t align = (uintptr_t)data & 3; + + if(align && (len >= 4)) { + /* Pre-load the temp registers */ + uint32_t t = 0, d = 0; + int32_t sl; + int32_t sr; + + switch(align) { + case 1: t |= data[2] << 16; + case 2: t |= data[1] << 8; + case 3: t |= data[0]; + } + + t <<= (8 * align); + + data += 4-align; + len -= 4-align; + + sl = 8 * (4-align); + sr = 8 * align; + + /* Mix */ + + while(len >= 4) { + uint32_t k; + + d = *(uint32_t *)data; + t = (t >> sr) | (d << sl); + + k = t; + + MIX(h,k,m); + + t = d; + + data += 4; + len -= 4; + } + + /* Handle leftover data in temp registers */ + + d = 0; + + if(len >= align) { + uint32_t k; + + switch(align) { + case 3: d |= data[2] << 16; + case 2: d |= data[1] << 8; + case 1: d |= data[0]; + } + + k = (t >> sr) | (d << sl); + MIX(h,k,m); + + data += align; + len -= align; + + /* ---------- + * Handle tail bytes */ + + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + } else { + switch(len) { + case 3: d |= data[2] << 16; + case 2: d |= data[1] << 8; + case 1: d |= data[0]; + case 0: h ^= (t >> sr) | (d << sl); h *= m; + } + } + + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; + } else { + while(len >= 4) { + uint32_t k = *(uint32_t *)data; + + MIX(h,k,m); + + data += 4; + len -= 4; + } + + /* ---------- + * Handle tail bytes */ + + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; + } +} +#undef MIX + +#endif /* UPB_UNALIGNED_READS_OK */ + +#include <errno.h> +#include <stdarg.h> +#include <stddef.h> +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +bool upb_dumptostderr(void *closure, const upb_status* status) { + UPB_UNUSED(closure); + fprintf(stderr, "%s\n", upb_status_errmsg(status)); + return false; +} + +/* Guarantee null-termination and provide ellipsis truncation. + * It may be tempting to "optimize" this by initializing these final + * four bytes up-front and then being careful never to overwrite them, + * this is safer and simpler. */ +static void nullz(upb_status *status) { + const char *ellipsis = "..."; + size_t len = strlen(ellipsis); + assert(sizeof(status->msg) > len); + memcpy(status->msg + sizeof(status->msg) - len, ellipsis, len); +} + + +/* upb_upberr *****************************************************************/ + +upb_errorspace upb_upberr = {"upb error"}; + +void upb_upberr_setoom(upb_status *status) { + status->error_space_ = &upb_upberr; + upb_status_seterrmsg(status, "Out of memory"); +} + + +/* upb_status *****************************************************************/ + +void upb_status_clear(upb_status *status) { + if (!status) return; + status->ok_ = true; + status->code_ = 0; + status->msg[0] = '\0'; +} + +bool upb_ok(const upb_status *status) { return status->ok_; } + +upb_errorspace *upb_status_errspace(const upb_status *status) { + return status->error_space_; +} + +int upb_status_errcode(const upb_status *status) { return status->code_; } + +const char *upb_status_errmsg(const upb_status *status) { return status->msg; } + +void upb_status_seterrmsg(upb_status *status, const char *msg) { + if (!status) return; + status->ok_ = false; + strncpy(status->msg, msg, sizeof(status->msg)); + nullz(status); +} + +void upb_status_seterrf(upb_status *status, const char *fmt, ...) { + va_list args; + va_start(args, fmt); + upb_status_vseterrf(status, fmt, args); + va_end(args); +} + +void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args) { + if (!status) return; + status->ok_ = false; + _upb_vsnprintf(status->msg, sizeof(status->msg), fmt, args); + nullz(status); +} + +void upb_status_copy(upb_status *to, const upb_status *from) { + if (!to) return; + *to = *from; +} + + +/* upb_alloc ******************************************************************/ + +static void *upb_global_allocfunc(upb_alloc *alloc, void *ptr, size_t oldsize, + size_t size) { + UPB_UNUSED(alloc); + UPB_UNUSED(oldsize); + if (size == 0) { + free(ptr); + return NULL; + } else { + return realloc(ptr, size); + } +} + +upb_alloc upb_alloc_global = {&upb_global_allocfunc}; + + +/* upb_arena ******************************************************************/ + +/* Be conservative and choose 16 in case anyone is using SSE. */ +static const size_t maxalign = 16; + +static size_t align_up(size_t size) { + return ((size + maxalign - 1) / maxalign) * maxalign; +} + +typedef struct mem_block { + struct mem_block *next; + size_t size; + size_t used; + bool owned; + /* Data follows. */ +} mem_block; + +typedef struct cleanup_ent { + struct cleanup_ent *next; + upb_cleanup_func *cleanup; + void *ud; +} cleanup_ent; + +static void upb_arena_addblock(upb_arena *a, void *ptr, size_t size, + bool owned) { + mem_block *block = ptr; + + block->next = a->block_head; + block->size = size; + block->used = align_up(sizeof(mem_block)); + block->owned = owned; + + a->block_head = block; + + /* TODO(haberman): ASAN poison. */ +} + + +static mem_block *upb_arena_allocblock(upb_arena *a, size_t size) { + size_t block_size = UPB_MAX(size, a->next_block_size) + sizeof(mem_block); + mem_block *block = upb_malloc(a->block_alloc, block_size); + + if (!block) { + return NULL; + } + + upb_arena_addblock(a, block, block_size, true); + a->next_block_size = UPB_MIN(block_size * 2, a->max_block_size); + + return block; +} + +static void *upb_arena_doalloc(upb_alloc *alloc, void *ptr, size_t oldsize, + size_t size) { + upb_arena *a = (upb_arena*)alloc; /* upb_alloc is initial member. */ + mem_block *block = a->block_head; + void *ret; + + if (size == 0) { + return NULL; /* We are an arena, don't need individual frees. */ + } + + size = align_up(size); + + /* TODO(haberman): special-case if this is a realloc of the last alloc? */ + + if (!block || block->size - block->used < size) { + /* Slow path: have to allocate a new block. */ + block = upb_arena_allocblock(a, size); + + if (!block) { + return NULL; /* Out of memory. */ + } + } + + ret = (char*)block + block->used; + block->used += size; + + if (oldsize > 0) { + memcpy(ret, ptr, oldsize); /* Preserve existing data. */ + } + + /* TODO(haberman): ASAN unpoison. */ + + a->bytes_allocated += size; + return ret; +} + +/* Public Arena API ***********************************************************/ + +void upb_arena_init(upb_arena *a) { + a->alloc.func = &upb_arena_doalloc; + a->block_alloc = &upb_alloc_global; + a->bytes_allocated = 0; + a->next_block_size = 256; + a->max_block_size = 16384; + a->cleanup_head = NULL; + a->block_head = NULL; +} + +void upb_arena_init2(upb_arena *a, void *mem, size_t size, upb_alloc *alloc) { + upb_arena_init(a); + + if (size > sizeof(mem_block)) { + upb_arena_addblock(a, mem, size, false); + } + + if (alloc) { + a->block_alloc = alloc; + } +} + +void upb_arena_uninit(upb_arena *a) { + cleanup_ent *ent = a->cleanup_head; + mem_block *block = a->block_head; + + while (ent) { + ent->cleanup(ent->ud); + ent = ent->next; + } + + /* Must do this after running cleanup functions, because this will delete + * the memory we store our cleanup entries in! */ + while (block) { + mem_block *next = block->next; + + if (block->owned) { + upb_free(a->block_alloc, block); + } + + block = next; + } +} + +bool upb_arena_addcleanup(upb_arena *a, upb_cleanup_func *func, void *ud) { + cleanup_ent *ent = upb_malloc(&a->alloc, sizeof(cleanup_ent)); + if (!ent) { + return false; /* Out of memory. */ + } + + ent->cleanup = func; + ent->ud = ud; + ent->next = a->cleanup_head; + a->cleanup_head = ent; + + return true; +} + +size_t upb_arena_bytesallocated(const upb_arena *a) { + return a->bytes_allocated; +} + + +/* Standard error functions ***************************************************/ + +static bool default_err(void *ud, const upb_status *status) { + UPB_UNUSED(ud); + UPB_UNUSED(status); + return false; +} + +static bool write_err_to(void *ud, const upb_status *status) { + upb_status *copy_to = ud; + upb_status_copy(copy_to, status); + return false; +} + + +/* upb_env ********************************************************************/ + +void upb_env_initonly(upb_env *e) { + e->ok_ = true; + e->error_func_ = &default_err; + e->error_ud_ = NULL; +} + +void upb_env_init(upb_env *e) { + upb_arena_init(&e->arena_); + upb_env_initonly(e); +} + +void upb_env_init2(upb_env *e, void *mem, size_t n, upb_alloc *alloc) { + upb_arena_init2(&e->arena_, mem, n, alloc); + upb_env_initonly(e); +} + +void upb_env_uninit(upb_env *e) { + upb_arena_uninit(&e->arena_); +} + +void upb_env_seterrorfunc(upb_env *e, upb_error_func *func, void *ud) { + e->error_func_ = func; + e->error_ud_ = ud; +} + +void upb_env_reporterrorsto(upb_env *e, upb_status *s) { + e->error_func_ = &write_err_to; + e->error_ud_ = s; +} + +bool upb_env_reporterror(upb_env *e, const upb_status *status) { + e->ok_ = false; + return e->error_func_(e->error_ud_, status); +} + +void *upb_env_malloc(upb_env *e, size_t size) { + return upb_malloc(&e->arena_.alloc, size); +} + +void *upb_env_realloc(upb_env *e, void *ptr, size_t oldsize, size_t size) { + return upb_realloc(&e->arena_.alloc, ptr, oldsize, size); +} + +void upb_env_free(upb_env *e, void *ptr) { + upb_free(&e->arena_.alloc, ptr); +} + +bool upb_env_addcleanup(upb_env *e, upb_cleanup_func *func, void *ud) { + return upb_arena_addcleanup(&e->arena_, func, ud); +} + +size_t upb_env_bytesallocated(const upb_env *e) { + return upb_arena_bytesallocated(&e->arena_); +} +/* This file was generated by upbc (the upb compiler) from the input + * file: + * + * upb/descriptor/descriptor.proto + * + * Do not edit -- your changes will be discarded when the file is + * regenerated. */ + +#include <assert.h> + + +static const upb_msgdef msgs[22]; +static const upb_fielddef fields[105]; +static const upb_enumdef enums[5]; +static const upb_tabent strentries[236]; +static const upb_tabent intentries[18]; +static const upb_tabval arrays[184]; + +#ifdef UPB_DEBUG_REFS +static upb_inttable reftables[264]; +#endif + +static const upb_msgdef msgs[22] = { + UPB_MSGDEF_INIT("google.protobuf.DescriptorProto", 40, 8, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[0], 11, 10), UPB_STRTABLE_INIT(10, 15, UPB_CTYPE_PTR, 4, &strentries[0]), false, UPB_SYNTAX_PROTO2, &reftables[0], &reftables[1]), + UPB_MSGDEF_INIT("google.protobuf.DescriptorProto.ExtensionRange", 4, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[11], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[16]), false, UPB_SYNTAX_PROTO2, &reftables[2], &reftables[3]), + UPB_MSGDEF_INIT("google.protobuf.DescriptorProto.ReservedRange", 4, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[14], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[20]), false, UPB_SYNTAX_PROTO2, &reftables[4], &reftables[5]), + UPB_MSGDEF_INIT("google.protobuf.EnumDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[17], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[24]), false, UPB_SYNTAX_PROTO2, &reftables[6], &reftables[7]), + UPB_MSGDEF_INIT("google.protobuf.EnumOptions", 8, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[0], &arrays[21], 4, 2), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[28]), false, UPB_SYNTAX_PROTO2, &reftables[8], &reftables[9]), + UPB_MSGDEF_INIT("google.protobuf.EnumValueDescriptorProto", 8, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[25], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[32]), false, UPB_SYNTAX_PROTO2, &reftables[10], &reftables[11]), + UPB_MSGDEF_INIT("google.protobuf.EnumValueOptions", 7, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[2], &arrays[29], 2, 1), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[36]), false, UPB_SYNTAX_PROTO2, &reftables[12], &reftables[13]), + UPB_MSGDEF_INIT("google.protobuf.FieldDescriptorProto", 23, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[31], 11, 10), UPB_STRTABLE_INIT(10, 15, UPB_CTYPE_PTR, 4, &strentries[40]), false, UPB_SYNTAX_PROTO2, &reftables[14], &reftables[15]), + UPB_MSGDEF_INIT("google.protobuf.FieldOptions", 12, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[4], &arrays[42], 11, 6), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[56]), false, UPB_SYNTAX_PROTO2, &reftables[16], &reftables[17]), + UPB_MSGDEF_INIT("google.protobuf.FileDescriptorProto", 42, 6, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[53], 13, 12), UPB_STRTABLE_INIT(12, 15, UPB_CTYPE_PTR, 4, &strentries[72]), false, UPB_SYNTAX_PROTO2, &reftables[18], &reftables[19]), + UPB_MSGDEF_INIT("google.protobuf.FileDescriptorSet", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[66], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[88]), false, UPB_SYNTAX_PROTO2, &reftables[20], &reftables[21]), + UPB_MSGDEF_INIT("google.protobuf.FileOptions", 31, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[6], &arrays[68], 39, 15), UPB_STRTABLE_INIT(16, 31, UPB_CTYPE_PTR, 5, &strentries[92]), false, UPB_SYNTAX_PROTO2, &reftables[22], &reftables[23]), + UPB_MSGDEF_INIT("google.protobuf.MessageOptions", 10, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[8], &arrays[107], 8, 4), UPB_STRTABLE_INIT(5, 7, UPB_CTYPE_PTR, 3, &strentries[124]), false, UPB_SYNTAX_PROTO2, &reftables[24], &reftables[25]), + UPB_MSGDEF_INIT("google.protobuf.MethodDescriptorProto", 15, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[115], 7, 6), UPB_STRTABLE_INIT(6, 7, UPB_CTYPE_PTR, 3, &strentries[132]), false, UPB_SYNTAX_PROTO2, &reftables[26], &reftables[27]), + UPB_MSGDEF_INIT("google.protobuf.MethodOptions", 7, 1, UPB_INTTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &intentries[10], &arrays[122], 1, 0), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[140]), false, UPB_SYNTAX_PROTO2, &reftables[28], &reftables[29]), + UPB_MSGDEF_INIT("google.protobuf.OneofDescriptorProto", 5, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[123], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[144]), false, UPB_SYNTAX_PROTO2, &reftables[30], &reftables[31]), + UPB_MSGDEF_INIT("google.protobuf.ServiceDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[125], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[148]), false, UPB_SYNTAX_PROTO2, &reftables[32], &reftables[33]), + UPB_MSGDEF_INIT("google.protobuf.ServiceOptions", 7, 1, UPB_INTTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &intentries[14], &arrays[129], 1, 0), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[152]), false, UPB_SYNTAX_PROTO2, &reftables[34], &reftables[35]), + UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[130], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[156]), false, UPB_SYNTAX_PROTO2, &reftables[36], &reftables[37]), + UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo.Location", 19, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[132], 7, 5), UPB_STRTABLE_INIT(5, 7, UPB_CTYPE_PTR, 3, &strentries[160]), false, UPB_SYNTAX_PROTO2, &reftables[38], &reftables[39]), + UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption", 18, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[139], 9, 7), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[168]), false, UPB_SYNTAX_PROTO2, &reftables[40], &reftables[41]), + UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption.NamePart", 6, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[148], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[184]), false, UPB_SYNTAX_PROTO2, &reftables[42], &reftables[43]), +}; + +static const upb_fielddef fields[105] = { + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "aggregate_value", 8, &msgs[20], NULL, 15, 6, {0},&reftables[44], &reftables[45]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "allow_alias", 2, &msgs[4], NULL, 6, 1, {0},&reftables[46], &reftables[47]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "cc_enable_arenas", 31, &msgs[11], NULL, 23, 12, {0},&reftables[48], &reftables[49]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "cc_generic_services", 16, &msgs[11], NULL, 17, 6, {0},&reftables[50], &reftables[51]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "client_streaming", 5, &msgs[13], NULL, 13, 4, {0},&reftables[52], &reftables[53]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "csharp_namespace", 37, &msgs[11], NULL, 27, 14, {0},&reftables[54], &reftables[55]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "ctype", 1, &msgs[8], (const upb_def*)(&enums[2]), 6, 1, {0},&reftables[56], &reftables[57]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "default_value", 7, &msgs[7], NULL, 16, 7, {0},&reftables[58], &reftables[59]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_STRING, 0, false, false, false, false, "dependency", 3, &msgs[9], NULL, 30, 8, {0},&reftables[60], &reftables[61]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 3, &msgs[12], NULL, 8, 3, {0},&reftables[62], &reftables[63]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 3, &msgs[8], NULL, 8, 3, {0},&reftables[64], &reftables[65]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 33, &msgs[14], NULL, 6, 1, {0},&reftables[66], &reftables[67]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 23, &msgs[11], NULL, 21, 10, {0},&reftables[68], &reftables[69]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 3, &msgs[4], NULL, 7, 2, {0},&reftables[70], &reftables[71]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 33, &msgs[17], NULL, 6, 1, {0},&reftables[72], &reftables[73]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 1, &msgs[6], NULL, 6, 1, {0},&reftables[74], &reftables[75]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_DOUBLE, 0, false, false, false, false, "double_value", 6, &msgs[20], NULL, 11, 4, {0},&reftables[76], &reftables[77]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "end", 2, &msgs[2], NULL, 3, 1, {0},&reftables[78], &reftables[79]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "end", 2, &msgs[1], NULL, 3, 1, {0},&reftables[80], &reftables[81]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 4, &msgs[0], (const upb_def*)(&msgs[3]), 18, 2, {0},&reftables[82], &reftables[83]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 5, &msgs[9], (const upb_def*)(&msgs[3]), 13, 1, {0},&reftables[84], &reftables[85]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "extendee", 2, &msgs[7], NULL, 7, 2, {0},&reftables[86], &reftables[87]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 6, &msgs[0], (const upb_def*)(&msgs[7]), 24, 4, {0},&reftables[88], &reftables[89]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 7, &msgs[9], (const upb_def*)(&msgs[7]), 19, 3, {0},&reftables[90], &reftables[91]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension_range", 5, &msgs[0], (const upb_def*)(&msgs[1]), 21, 3, {0},&reftables[92], &reftables[93]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "field", 2, &msgs[0], (const upb_def*)(&msgs[7]), 12, 0, {0},&reftables[94], &reftables[95]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "file", 1, &msgs[10], (const upb_def*)(&msgs[9]), 5, 0, {0},&reftables[96], &reftables[97]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "go_package", 11, &msgs[11], NULL, 14, 5, {0},&reftables[98], &reftables[99]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "identifier_value", 3, &msgs[20], NULL, 6, 1, {0},&reftables[100], &reftables[101]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "input_type", 2, &msgs[13], NULL, 7, 2, {0},&reftables[102], &reftables[103]), + UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_BOOL, 0, false, false, false, false, "is_extension", 2, &msgs[21], NULL, 5, 1, {0},&reftables[104], &reftables[105]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generate_equals_and_hash", 20, &msgs[11], NULL, 20, 9, {0},&reftables[106], &reftables[107]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generic_services", 17, &msgs[11], NULL, 18, 7, {0},&reftables[108], &reftables[109]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_multiple_files", 10, &msgs[11], NULL, 13, 4, {0},&reftables[110], &reftables[111]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_outer_classname", 8, &msgs[11], NULL, 9, 2, {0},&reftables[112], &reftables[113]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_package", 1, &msgs[11], NULL, 6, 1, {0},&reftables[114], &reftables[115]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_string_check_utf8", 27, &msgs[11], NULL, 22, 11, {0},&reftables[116], &reftables[117]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "javanano_use_deprecated_package", 38, &msgs[11], NULL, 30, 15, {0},&reftables[118], &reftables[119]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "json_name", 10, &msgs[7], NULL, 20, 9, {0},&reftables[120], &reftables[121]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "jstype", 6, &msgs[8], (const upb_def*)(&enums[3]), 10, 5, {0},&reftables[122], &reftables[123]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "label", 4, &msgs[7], (const upb_def*)(&enums[0]), 11, 4, {0},&reftables[124], &reftables[125]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "lazy", 5, &msgs[8], NULL, 9, 4, {0},&reftables[126], &reftables[127]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "leading_comments", 3, &msgs[19], NULL, 8, 2, {0},&reftables[128], &reftables[129]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_STRING, 0, false, false, false, false, "leading_detached_comments", 6, &msgs[19], NULL, 16, 4, {0},&reftables[130], &reftables[131]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "location", 1, &msgs[18], (const upb_def*)(&msgs[19]), 5, 0, {0},&reftables[132], &reftables[133]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "map_entry", 7, &msgs[12], NULL, 9, 4, {0},&reftables[134], &reftables[135]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "message_set_wire_format", 1, &msgs[12], NULL, 6, 1, {0},&reftables[136], &reftables[137]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "message_type", 4, &msgs[9], (const upb_def*)(&msgs[0]), 10, 0, {0},&reftables[138], &reftables[139]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "method", 2, &msgs[16], (const upb_def*)(&msgs[13]), 6, 0, {0},&reftables[140], &reftables[141]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[3], NULL, 8, 2, {0},&reftables[142], &reftables[143]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[15], NULL, 2, 0, {0},&reftables[144], &reftables[145]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "name", 2, &msgs[20], (const upb_def*)(&msgs[21]), 5, 0, {0},&reftables[146], &reftables[147]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[0], NULL, 32, 8, {0},&reftables[148], &reftables[149]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[5], NULL, 4, 1, {0},&reftables[150], &reftables[151]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[9], NULL, 22, 6, {0},&reftables[152], &reftables[153]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[7], NULL, 4, 1, {0},&reftables[154], &reftables[155]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[13], NULL, 4, 1, {0},&reftables[156], &reftables[157]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[16], NULL, 8, 2, {0},&reftables[158], &reftables[159]), + UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_STRING, 0, false, false, false, false, "name_part", 1, &msgs[21], NULL, 2, 0, {0},&reftables[160], &reftables[161]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT64, UPB_INTFMT_VARIABLE, false, false, false, false, "negative_int_value", 5, &msgs[20], NULL, 10, 3, {0},&reftables[162], &reftables[163]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "nested_type", 3, &msgs[0], (const upb_def*)(&msgs[0]), 15, 1, {0},&reftables[164], &reftables[165]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "no_standard_descriptor_accessor", 2, &msgs[12], NULL, 7, 2, {0},&reftables[166], &reftables[167]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 2, &msgs[5], NULL, 7, 2, {0},&reftables[168], &reftables[169]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 3, &msgs[7], NULL, 10, 3, {0},&reftables[170], &reftables[171]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "objc_class_prefix", 36, &msgs[11], NULL, 24, 13, {0},&reftables[172], &reftables[173]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "oneof_decl", 8, &msgs[0], (const upb_def*)(&msgs[15]), 28, 6, {0},&reftables[174], &reftables[175]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "oneof_index", 9, &msgs[7], NULL, 19, 8, {0},&reftables[176], &reftables[177]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "optimize_for", 9, &msgs[11], (const upb_def*)(&enums[4]), 12, 3, {0},&reftables[178], &reftables[179]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 7, &msgs[0], (const upb_def*)(&msgs[12]), 25, 5, {0},&reftables[180], &reftables[181]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[9], (const upb_def*)(&msgs[11]), 20, 4, {0},&reftables[182], &reftables[183]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 4, &msgs[13], (const upb_def*)(&msgs[14]), 3, 0, {0},&reftables[184], &reftables[185]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[7], (const upb_def*)(&msgs[8]), 3, 0, {0},&reftables[186], &reftables[187]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[16], (const upb_def*)(&msgs[17]), 7, 1, {0},&reftables[188], &reftables[189]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[5], (const upb_def*)(&msgs[6]), 3, 0, {0},&reftables[190], &reftables[191]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[3], (const upb_def*)(&msgs[4]), 7, 1, {0},&reftables[192], &reftables[193]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "output_type", 3, &msgs[13], NULL, 10, 3, {0},&reftables[194], &reftables[195]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "package", 2, &msgs[9], NULL, 25, 7, {0},&reftables[196], &reftables[197]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "packed", 2, &msgs[8], NULL, 7, 2, {0},&reftables[198], &reftables[199]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "path", 1, &msgs[19], NULL, 4, 0, {0},&reftables[200], &reftables[201]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_UINT64, UPB_INTFMT_VARIABLE, false, false, false, false, "positive_int_value", 4, &msgs[20], NULL, 9, 2, {0},&reftables[202], &reftables[203]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "public_dependency", 10, &msgs[9], NULL, 35, 9, {0},&reftables[204], &reftables[205]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "py_generic_services", 18, &msgs[11], NULL, 19, 8, {0},&reftables[206], &reftables[207]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_STRING, 0, false, false, false, false, "reserved_name", 10, &msgs[0], NULL, 37, 9, {0},&reftables[208], &reftables[209]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "reserved_range", 9, &msgs[0], (const upb_def*)(&msgs[2]), 31, 7, {0},&reftables[210], &reftables[211]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "server_streaming", 6, &msgs[13], NULL, 14, 5, {0},&reftables[212], &reftables[213]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "service", 6, &msgs[9], (const upb_def*)(&msgs[16]), 16, 2, {0},&reftables[214], &reftables[215]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "source_code_info", 9, &msgs[9], (const upb_def*)(&msgs[18]), 21, 5, {0},&reftables[216], &reftables[217]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "span", 2, &msgs[19], NULL, 7, 1, {0},&reftables[218], &reftables[219]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "start", 1, &msgs[2], NULL, 2, 0, {0},&reftables[220], &reftables[221]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "start", 1, &msgs[1], NULL, 2, 0, {0},&reftables[222], &reftables[223]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BYTES, 0, false, false, false, false, "string_value", 7, &msgs[20], NULL, 12, 5, {0},&reftables[224], &reftables[225]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "syntax", 12, &msgs[9], NULL, 39, 11, {0},&reftables[226], &reftables[227]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "trailing_comments", 4, &msgs[19], NULL, 11, 3, {0},&reftables[228], &reftables[229]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "type", 5, &msgs[7], (const upb_def*)(&enums[1]), 12, 5, {0},&reftables[230], &reftables[231]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "type_name", 6, &msgs[7], NULL, 13, 6, {0},&reftables[232], &reftables[233]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[11], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[234], &reftables[235]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[12], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[236], &reftables[237]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[6], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[238], &reftables[239]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[4], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[240], &reftables[241]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[8], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[242], &reftables[243]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[14], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[244], &reftables[245]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[17], (const upb_def*)(&msgs[20]), 5, 0, {0},&reftables[246], &reftables[247]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "value", 2, &msgs[3], (const upb_def*)(&msgs[5]), 6, 0, {0},&reftables[248], &reftables[249]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "weak", 10, &msgs[8], NULL, 11, 6, {0},&reftables[250], &reftables[251]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "weak_dependency", 11, &msgs[9], NULL, 38, 10, {0},&reftables[252], &reftables[253]), +}; + +static const upb_enumdef enums[5] = { + UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Label", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[188]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[151], 4, 3), 0, &reftables[254], &reftables[255]), + UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Type", UPB_STRTABLE_INIT(18, 31, UPB_CTYPE_INT32, 5, &strentries[192]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[155], 19, 18), 0, &reftables[256], &reftables[257]), + UPB_ENUMDEF_INIT("google.protobuf.FieldOptions.CType", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[224]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[174], 3, 3), 0, &reftables[258], &reftables[259]), + UPB_ENUMDEF_INIT("google.protobuf.FieldOptions.JSType", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[228]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[177], 3, 3), 0, &reftables[260], &reftables[261]), + UPB_ENUMDEF_INIT("google.protobuf.FileOptions.OptimizeMode", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[232]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[180], 4, 3), 0, &reftables[262], &reftables[263]), +}; + +static const upb_tabent strentries[236] = { + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "extension"), UPB_TABVALUE_PTR_INIT(&fields[22]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "reserved_name"), UPB_TABVALUE_PTR_INIT(&fields[82]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[52]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "field"), UPB_TABVALUE_PTR_INIT(&fields[25]), &strentries[12]}, + {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "extension_range"), UPB_TABVALUE_PTR_INIT(&fields[24]), &strentries[14]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "nested_type"), UPB_TABVALUE_PTR_INIT(&fields[60]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "reserved_range"), UPB_TABVALUE_PTR_INIT(&fields[83]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[68]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "oneof_decl"), UPB_TABVALUE_PTR_INIT(&fields[65]), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "enum_type"), UPB_TABVALUE_PTR_INIT(&fields[19]), &strentries[13]}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "start"), UPB_TABVALUE_PTR_INIT(&fields[89]), NULL}, + {UPB_TABKEY_STR("\003", "\000", "\000", "\000", "end"), UPB_TABVALUE_PTR_INIT(&fields[18]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "start"), UPB_TABVALUE_PTR_INIT(&fields[88]), NULL}, + {UPB_TABKEY_STR("\003", "\000", "\000", "\000", "end"), UPB_TABVALUE_PTR_INIT(&fields[17]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "value"), UPB_TABVALUE_PTR_INIT(&fields[102]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[74]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[49]), &strentries[26]}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[98]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[13]), NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "allow_alias"), UPB_TABVALUE_PTR_INIT(&fields[1]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "number"), UPB_TABVALUE_PTR_INIT(&fields[62]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[73]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[53]), &strentries[34]}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[97]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[15]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "oneof_index"), UPB_TABVALUE_PTR_INIT(&fields[66]), NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "label"), UPB_TABVALUE_PTR_INIT(&fields[40]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[55]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "number"), UPB_TABVALUE_PTR_INIT(&fields[63]), &strentries[53]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\010", "\000", "\000", "\000", "extendee"), UPB_TABVALUE_PTR_INIT(&fields[21]), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "type_name"), UPB_TABVALUE_PTR_INIT(&fields[94]), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "json_name"), UPB_TABVALUE_PTR_INIT(&fields[38]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "type"), UPB_TABVALUE_PTR_INIT(&fields[93]), &strentries[50]}, + {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "default_value"), UPB_TABVALUE_PTR_INIT(&fields[7]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[71]), NULL}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[99]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "weak"), UPB_TABVALUE_PTR_INIT(&fields[103]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "packed"), UPB_TABVALUE_PTR_INIT(&fields[77]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "lazy"), UPB_TABVALUE_PTR_INIT(&fields[41]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "ctype"), UPB_TABVALUE_PTR_INIT(&fields[6]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "jstype"), UPB_TABVALUE_PTR_INIT(&fields[39]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[10]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "extension"), UPB_TABVALUE_PTR_INIT(&fields[23]), NULL}, + {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "weak_dependency"), UPB_TABVALUE_PTR_INIT(&fields[104]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[54]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "service"), UPB_TABVALUE_PTR_INIT(&fields[85]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "source_code_info"), UPB_TABVALUE_PTR_INIT(&fields[86]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "syntax"), UPB_TABVALUE_PTR_INIT(&fields[91]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "dependency"), UPB_TABVALUE_PTR_INIT(&fields[8]), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "message_type"), UPB_TABVALUE_PTR_INIT(&fields[47]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "package"), UPB_TABVALUE_PTR_INIT(&fields[76]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[69]), &strentries[86]}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "enum_type"), UPB_TABVALUE_PTR_INIT(&fields[20]), NULL}, + {UPB_TABKEY_STR("\021", "\000", "\000", "\000", "public_dependency"), UPB_TABVALUE_PTR_INIT(&fields[80]), &strentries[85]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "file"), UPB_TABVALUE_PTR_INIT(&fields[26]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "cc_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[3]), NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "csharp_namespace"), UPB_TABVALUE_PTR_INIT(&fields[5]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "go_package"), UPB_TABVALUE_PTR_INIT(&fields[27]), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "java_package"), UPB_TABVALUE_PTR_INIT(&fields[35]), &strentries[120]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "java_outer_classname"), UPB_TABVALUE_PTR_INIT(&fields[34]), NULL}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[95]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "java_multiple_files"), UPB_TABVALUE_PTR_INIT(&fields[33]), &strentries[117]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\025", "\000", "\000", "\000", "java_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[32]), &strentries[118]}, + {UPB_TABKEY_STR("\035", "\000", "\000", "\000", "java_generate_equals_and_hash"), UPB_TABVALUE_PTR_INIT(&fields[31]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\037", "\000", "\000", "\000", "javanano_use_deprecated_package"), UPB_TABVALUE_PTR_INIT(&fields[37]), &strentries[123]}, + {UPB_TABKEY_STR("\023", "\000", "\000", "\000", "py_generic_services"), UPB_TABVALUE_PTR_INIT(&fields[81]), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "optimize_for"), UPB_TABVALUE_PTR_INIT(&fields[67]), NULL}, + {UPB_TABKEY_STR("\026", "\000", "\000", "\000", "java_string_check_utf8"), UPB_TABVALUE_PTR_INIT(&fields[36]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[12]), &strentries[119]}, + {UPB_TABKEY_STR("\021", "\000", "\000", "\000", "objc_class_prefix"), UPB_TABVALUE_PTR_INIT(&fields[64]), NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "cc_enable_arenas"), UPB_TABVALUE_PTR_INIT(&fields[2]), NULL}, + {UPB_TABKEY_STR("\027", "\000", "\000", "\000", "message_set_wire_format"), UPB_TABVALUE_PTR_INIT(&fields[46]), &strentries[128]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[96]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[9]), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "map_entry"), UPB_TABVALUE_PTR_INIT(&fields[45]), NULL}, + {UPB_TABKEY_STR("\037", "\000", "\000", "\000", "no_standard_descriptor_accessor"), UPB_TABVALUE_PTR_INIT(&fields[61]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "client_streaming"), UPB_TABVALUE_PTR_INIT(&fields[4]), NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "server_streaming"), UPB_TABVALUE_PTR_INIT(&fields[84]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[56]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "input_type"), UPB_TABVALUE_PTR_INIT(&fields[29]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "output_type"), UPB_TABVALUE_PTR_INIT(&fields[75]), NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[70]), NULL}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[100]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[11]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[50]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\007", "\000", "\000", "\000", "options"), UPB_TABVALUE_PTR_INIT(&fields[72]), &strentries[150]}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "method"), UPB_TABVALUE_PTR_INIT(&fields[48]), NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[57]), &strentries[149]}, + {UPB_TABKEY_STR("\024", "\000", "\000", "\000", "uninterpreted_option"), UPB_TABVALUE_PTR_INIT(&fields[101]), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "deprecated"), UPB_TABVALUE_PTR_INIT(&fields[14]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\010", "\000", "\000", "\000", "location"), UPB_TABVALUE_PTR_INIT(&fields[44]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "span"), UPB_TABVALUE_PTR_INIT(&fields[87]), &strentries[167]}, + {UPB_TABKEY_STR("\031", "\000", "\000", "\000", "leading_detached_comments"), UPB_TABVALUE_PTR_INIT(&fields[43]), &strentries[165]}, + {UPB_TABKEY_STR("\021", "\000", "\000", "\000", "trailing_comments"), UPB_TABVALUE_PTR_INIT(&fields[92]), NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "leading_comments"), UPB_TABVALUE_PTR_INIT(&fields[42]), &strentries[164]}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "path"), UPB_TABVALUE_PTR_INIT(&fields[78]), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "double_value"), UPB_TABVALUE_PTR_INIT(&fields[16]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "name"), UPB_TABVALUE_PTR_INIT(&fields[51]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\022", "\000", "\000", "\000", "negative_int_value"), UPB_TABVALUE_PTR_INIT(&fields[59]), NULL}, + {UPB_TABKEY_STR("\017", "\000", "\000", "\000", "aggregate_value"), UPB_TABVALUE_PTR_INIT(&fields[0]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\022", "\000", "\000", "\000", "positive_int_value"), UPB_TABVALUE_PTR_INIT(&fields[79]), NULL}, + {UPB_TABKEY_STR("\020", "\000", "\000", "\000", "identifier_value"), UPB_TABVALUE_PTR_INIT(&fields[28]), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "string_value"), UPB_TABVALUE_PTR_INIT(&fields[90]), &strentries[182]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "is_extension"), UPB_TABVALUE_PTR_INIT(&fields[30]), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "name_part"), UPB_TABVALUE_PTR_INIT(&fields[58]), NULL}, + {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_REQUIRED"), UPB_TABVALUE_INT_INIT(2), &strentries[190]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_REPEATED"), UPB_TABVALUE_INT_INIT(3), NULL}, + {UPB_TABKEY_STR("\016", "\000", "\000", "\000", "LABEL_OPTIONAL"), UPB_TABVALUE_INT_INIT(1), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_FIXED64"), UPB_TABVALUE_INT_INIT(6), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_STRING"), UPB_TABVALUE_INT_INIT(9), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_FLOAT"), UPB_TABVALUE_INT_INIT(2), &strentries[221]}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_DOUBLE"), UPB_TABVALUE_INT_INIT(1), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_INT32"), UPB_TABVALUE_INT_INIT(5), NULL}, + {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "TYPE_SFIXED32"), UPB_TABVALUE_INT_INIT(15), NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_FIXED32"), UPB_TABVALUE_INT_INIT(7), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "TYPE_MESSAGE"), UPB_TABVALUE_INT_INIT(11), &strentries[222]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_INT64"), UPB_TABVALUE_INT_INIT(3), &strentries[219]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "TYPE_ENUM"), UPB_TABVALUE_INT_INIT(14), NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_UINT32"), UPB_TABVALUE_INT_INIT(13), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_UINT64"), UPB_TABVALUE_INT_INIT(4), &strentries[218]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\015", "\000", "\000", "\000", "TYPE_SFIXED64"), UPB_TABVALUE_INT_INIT(16), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_BYTES"), UPB_TABVALUE_INT_INIT(12), NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_SINT64"), UPB_TABVALUE_INT_INIT(18), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "TYPE_BOOL"), UPB_TABVALUE_INT_INIT(8), NULL}, + {UPB_TABKEY_STR("\012", "\000", "\000", "\000", "TYPE_GROUP"), UPB_TABVALUE_INT_INIT(10), NULL}, + {UPB_TABKEY_STR("\013", "\000", "\000", "\000", "TYPE_SINT32"), UPB_TABVALUE_INT_INIT(17), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\004", "\000", "\000", "\000", "CORD"), UPB_TABVALUE_INT_INIT(1), NULL}, + {UPB_TABKEY_STR("\006", "\000", "\000", "\000", "STRING"), UPB_TABVALUE_INT_INIT(0), &strentries[225]}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "STRING_PIECE"), UPB_TABVALUE_INT_INIT(2), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "JS_NORMAL"), UPB_TABVALUE_INT_INIT(0), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "JS_NUMBER"), UPB_TABVALUE_INT_INIT(2), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "JS_STRING"), UPB_TABVALUE_INT_INIT(1), NULL}, + {UPB_TABKEY_STR("\011", "\000", "\000", "\000", "CODE_SIZE"), UPB_TABVALUE_INT_INIT(2), NULL}, + {UPB_TABKEY_STR("\005", "\000", "\000", "\000", "SPEED"), UPB_TABVALUE_INT_INIT(1), &strentries[235]}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_STR("\014", "\000", "\000", "\000", "LITE_RUNTIME"), UPB_TABVALUE_INT_INIT(3), NULL}, +}; + +static const upb_tabent intentries[18] = { + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[98]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[97]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[99]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[95]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[96]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(33), UPB_TABVALUE_PTR_INIT(&fields[11]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[100]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(33), UPB_TABVALUE_PTR_INIT(&fields[14]), NULL}, + {UPB_TABKEY_NONE, UPB_TABVALUE_EMPTY_INIT, NULL}, + {UPB_TABKEY_NUM(999), UPB_TABVALUE_PTR_INIT(&fields[101]), NULL}, +}; + +static const upb_tabval arrays[184] = { + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[52]), + UPB_TABVALUE_PTR_INIT(&fields[25]), + UPB_TABVALUE_PTR_INIT(&fields[60]), + UPB_TABVALUE_PTR_INIT(&fields[19]), + UPB_TABVALUE_PTR_INIT(&fields[24]), + UPB_TABVALUE_PTR_INIT(&fields[22]), + UPB_TABVALUE_PTR_INIT(&fields[68]), + UPB_TABVALUE_PTR_INIT(&fields[65]), + UPB_TABVALUE_PTR_INIT(&fields[83]), + UPB_TABVALUE_PTR_INIT(&fields[82]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[89]), + UPB_TABVALUE_PTR_INIT(&fields[18]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[88]), + UPB_TABVALUE_PTR_INIT(&fields[17]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[49]), + UPB_TABVALUE_PTR_INIT(&fields[102]), + UPB_TABVALUE_PTR_INIT(&fields[74]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[1]), + UPB_TABVALUE_PTR_INIT(&fields[13]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[53]), + UPB_TABVALUE_PTR_INIT(&fields[62]), + UPB_TABVALUE_PTR_INIT(&fields[73]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[15]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[55]), + UPB_TABVALUE_PTR_INIT(&fields[21]), + UPB_TABVALUE_PTR_INIT(&fields[63]), + UPB_TABVALUE_PTR_INIT(&fields[40]), + UPB_TABVALUE_PTR_INIT(&fields[93]), + UPB_TABVALUE_PTR_INIT(&fields[94]), + UPB_TABVALUE_PTR_INIT(&fields[7]), + UPB_TABVALUE_PTR_INIT(&fields[71]), + UPB_TABVALUE_PTR_INIT(&fields[66]), + UPB_TABVALUE_PTR_INIT(&fields[38]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[6]), + UPB_TABVALUE_PTR_INIT(&fields[77]), + UPB_TABVALUE_PTR_INIT(&fields[10]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[41]), + UPB_TABVALUE_PTR_INIT(&fields[39]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[103]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[54]), + UPB_TABVALUE_PTR_INIT(&fields[76]), + UPB_TABVALUE_PTR_INIT(&fields[8]), + UPB_TABVALUE_PTR_INIT(&fields[47]), + UPB_TABVALUE_PTR_INIT(&fields[20]), + UPB_TABVALUE_PTR_INIT(&fields[85]), + UPB_TABVALUE_PTR_INIT(&fields[23]), + UPB_TABVALUE_PTR_INIT(&fields[69]), + UPB_TABVALUE_PTR_INIT(&fields[86]), + UPB_TABVALUE_PTR_INIT(&fields[80]), + UPB_TABVALUE_PTR_INIT(&fields[104]), + UPB_TABVALUE_PTR_INIT(&fields[91]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[26]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[35]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[34]), + UPB_TABVALUE_PTR_INIT(&fields[67]), + UPB_TABVALUE_PTR_INIT(&fields[33]), + UPB_TABVALUE_PTR_INIT(&fields[27]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[3]), + UPB_TABVALUE_PTR_INIT(&fields[32]), + UPB_TABVALUE_PTR_INIT(&fields[81]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[31]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[12]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[36]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[2]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[64]), + UPB_TABVALUE_PTR_INIT(&fields[5]), + UPB_TABVALUE_PTR_INIT(&fields[37]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[46]), + UPB_TABVALUE_PTR_INIT(&fields[61]), + UPB_TABVALUE_PTR_INIT(&fields[9]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[45]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[56]), + UPB_TABVALUE_PTR_INIT(&fields[29]), + UPB_TABVALUE_PTR_INIT(&fields[75]), + UPB_TABVALUE_PTR_INIT(&fields[70]), + UPB_TABVALUE_PTR_INIT(&fields[4]), + UPB_TABVALUE_PTR_INIT(&fields[84]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[50]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[57]), + UPB_TABVALUE_PTR_INIT(&fields[48]), + UPB_TABVALUE_PTR_INIT(&fields[72]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[44]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[78]), + UPB_TABVALUE_PTR_INIT(&fields[87]), + UPB_TABVALUE_PTR_INIT(&fields[42]), + UPB_TABVALUE_PTR_INIT(&fields[92]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[43]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[51]), + UPB_TABVALUE_PTR_INIT(&fields[28]), + UPB_TABVALUE_PTR_INIT(&fields[79]), + UPB_TABVALUE_PTR_INIT(&fields[59]), + UPB_TABVALUE_PTR_INIT(&fields[16]), + UPB_TABVALUE_PTR_INIT(&fields[90]), + UPB_TABVALUE_PTR_INIT(&fields[0]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT(&fields[58]), + UPB_TABVALUE_PTR_INIT(&fields[30]), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT("LABEL_OPTIONAL"), + UPB_TABVALUE_PTR_INIT("LABEL_REQUIRED"), + UPB_TABVALUE_PTR_INIT("LABEL_REPEATED"), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT("TYPE_DOUBLE"), + UPB_TABVALUE_PTR_INIT("TYPE_FLOAT"), + UPB_TABVALUE_PTR_INIT("TYPE_INT64"), + UPB_TABVALUE_PTR_INIT("TYPE_UINT64"), + UPB_TABVALUE_PTR_INIT("TYPE_INT32"), + UPB_TABVALUE_PTR_INIT("TYPE_FIXED64"), + UPB_TABVALUE_PTR_INIT("TYPE_FIXED32"), + UPB_TABVALUE_PTR_INIT("TYPE_BOOL"), + UPB_TABVALUE_PTR_INIT("TYPE_STRING"), + UPB_TABVALUE_PTR_INIT("TYPE_GROUP"), + UPB_TABVALUE_PTR_INIT("TYPE_MESSAGE"), + UPB_TABVALUE_PTR_INIT("TYPE_BYTES"), + UPB_TABVALUE_PTR_INIT("TYPE_UINT32"), + UPB_TABVALUE_PTR_INIT("TYPE_ENUM"), + UPB_TABVALUE_PTR_INIT("TYPE_SFIXED32"), + UPB_TABVALUE_PTR_INIT("TYPE_SFIXED64"), + UPB_TABVALUE_PTR_INIT("TYPE_SINT32"), + UPB_TABVALUE_PTR_INIT("TYPE_SINT64"), + UPB_TABVALUE_PTR_INIT("STRING"), + UPB_TABVALUE_PTR_INIT("CORD"), + UPB_TABVALUE_PTR_INIT("STRING_PIECE"), + UPB_TABVALUE_PTR_INIT("JS_NORMAL"), + UPB_TABVALUE_PTR_INIT("JS_STRING"), + UPB_TABVALUE_PTR_INIT("JS_NUMBER"), + UPB_TABVALUE_EMPTY_INIT, + UPB_TABVALUE_PTR_INIT("SPEED"), + UPB_TABVALUE_PTR_INIT("CODE_SIZE"), + UPB_TABVALUE_PTR_INIT("LITE_RUNTIME"), +}; + +#ifdef UPB_DEBUG_REFS +static upb_inttable reftables[264] = { + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), +}; +#endif + +static const upb_msgdef *refm(const upb_msgdef *m, const void *owner) { + upb_msgdef_ref(m, owner); + return m; +} + +static const upb_enumdef *refe(const upb_enumdef *e, const void *owner) { + upb_enumdef_ref(e, owner); + return e; +} + +/* Public API. */ +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_get(const void *owner) { return refm(&msgs[0], owner); } +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_get(const void *owner) { return refm(&msgs[1], owner); } +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ReservedRange_get(const void *owner) { return refm(&msgs[2], owner); } +const upb_msgdef *upbdefs_google_protobuf_EnumDescriptorProto_get(const void *owner) { return refm(&msgs[3], owner); } +const upb_msgdef *upbdefs_google_protobuf_EnumOptions_get(const void *owner) { return refm(&msgs[4], owner); } +const upb_msgdef *upbdefs_google_protobuf_EnumValueDescriptorProto_get(const void *owner) { return refm(&msgs[5], owner); } +const upb_msgdef *upbdefs_google_protobuf_EnumValueOptions_get(const void *owner) { return refm(&msgs[6], owner); } +const upb_msgdef *upbdefs_google_protobuf_FieldDescriptorProto_get(const void *owner) { return refm(&msgs[7], owner); } +const upb_msgdef *upbdefs_google_protobuf_FieldOptions_get(const void *owner) { return refm(&msgs[8], owner); } +const upb_msgdef *upbdefs_google_protobuf_FileDescriptorProto_get(const void *owner) { return refm(&msgs[9], owner); } +const upb_msgdef *upbdefs_google_protobuf_FileDescriptorSet_get(const void *owner) { return refm(&msgs[10], owner); } +const upb_msgdef *upbdefs_google_protobuf_FileOptions_get(const void *owner) { return refm(&msgs[11], owner); } +const upb_msgdef *upbdefs_google_protobuf_MessageOptions_get(const void *owner) { return refm(&msgs[12], owner); } +const upb_msgdef *upbdefs_google_protobuf_MethodDescriptorProto_get(const void *owner) { return refm(&msgs[13], owner); } +const upb_msgdef *upbdefs_google_protobuf_MethodOptions_get(const void *owner) { return refm(&msgs[14], owner); } +const upb_msgdef *upbdefs_google_protobuf_OneofDescriptorProto_get(const void *owner) { return refm(&msgs[15], owner); } +const upb_msgdef *upbdefs_google_protobuf_ServiceDescriptorProto_get(const void *owner) { return refm(&msgs[16], owner); } +const upb_msgdef *upbdefs_google_protobuf_ServiceOptions_get(const void *owner) { return refm(&msgs[17], owner); } +const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_get(const void *owner) { return refm(&msgs[18], owner); } +const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_Location_get(const void *owner) { return refm(&msgs[19], owner); } +const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_get(const void *owner) { return refm(&msgs[20], owner); } +const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_NamePart_get(const void *owner) { return refm(&msgs[21], owner); } + +const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Label_get(const void *owner) { return refe(&enums[0], owner); } +const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Type_get(const void *owner) { return refe(&enums[1], owner); } +const upb_enumdef *upbdefs_google_protobuf_FieldOptions_CType_get(const void *owner) { return refe(&enums[2], owner); } +const upb_enumdef *upbdefs_google_protobuf_FieldOptions_JSType_get(const void *owner) { return refe(&enums[3], owner); } +const upb_enumdef *upbdefs_google_protobuf_FileOptions_OptimizeMode_get(const void *owner) { return refe(&enums[4], owner); } +/* +** XXX: The routines in this file that consume a string do not currently +** support having the string span buffers. In the future, as upb_sink and +** its buffering/sharing functionality evolve there should be an easy and +** idiomatic way of correctly handling this case. For now, we accept this +** limitation since we currently only parse descriptors from single strings. +*/ + + +#include <errno.h> +#include <stdlib.h> +#include <string.h> + +/* Compares a NULL-terminated string with a non-NULL-terminated string. */ +static bool upb_streq(const char *str, const char *buf, size_t n) { + return strlen(str) == n && memcmp(str, buf, n) == 0; +} + +/* We keep a stack of all the messages scopes we are currently in, as well as + * the top-level file scope. This is necessary to correctly qualify the + * definitions that are contained inside. "name" tracks the name of the + * message or package (a bare name -- not qualified by any enclosing scopes). */ +typedef struct { + char *name; + /* Index of the first def that is under this scope. For msgdefs, the + * msgdef itself is at start-1. */ + int start; +} upb_descreader_frame; + +/* The maximum number of nested declarations that are allowed, ie. + * message Foo { + * message Bar { + * message Baz { + * } + * } + * } + * + * This is a resource limit that affects how big our runtime stack can grow. + * TODO: make this a runtime-settable property of the Reader instance. */ +#define UPB_MAX_MESSAGE_NESTING 64 + +struct upb_descreader { + upb_sink sink; + upb_inttable files; + upb_filedef *file; /* The last file in files. */ + upb_descreader_frame stack[UPB_MAX_MESSAGE_NESTING]; + int stack_len; + + uint32_t number; + char *name; + bool saw_number; + bool saw_name; + + char *default_string; + + upb_fielddef *f; +}; + +static char *upb_strndup(const char *buf, size_t n) { + char *ret = upb_gmalloc(n + 1); + if (!ret) return NULL; + memcpy(ret, buf, n); + ret[n] = '\0'; + return ret; +} + +/* Returns a newly allocated string that joins input strings together, for + * example: + * join("Foo.Bar", "Baz") -> "Foo.Bar.Baz" + * join("", "Baz") -> "Baz" + * Caller owns a ref on the returned string. */ +static char *upb_join(const char *base, const char *name) { + if (!base || strlen(base) == 0) { + return upb_gstrdup(name); + } else { + char *ret = upb_gmalloc(strlen(base) + strlen(name) + 2); + if (!ret) { + return NULL; + } + ret[0] = '\0'; + strcat(ret, base); + strcat(ret, "."); + strcat(ret, name); + return ret; + } +} + +/* Qualify the defname for all defs starting with offset "start" with "str". */ +static bool upb_descreader_qualify(upb_filedef *f, char *str, int32_t start) { + size_t i; + for (i = start; i < upb_filedef_defcount(f); i++) { + upb_def *def = upb_filedef_mutabledef(f, i); + char *name = upb_join(str, upb_def_fullname(def)); + if (!name) { + /* Need better logic here; at this point we've qualified some names but + * not others. */ + return false; + } + upb_def_setfullname(def, name, NULL); + upb_gfree(name); + } + return true; +} + + +/* upb_descreader ************************************************************/ + +static upb_msgdef *upb_descreader_top(upb_descreader *r) { + int index; + assert(r->stack_len > 1); + index = r->stack[r->stack_len-1].start - 1; + assert(index >= 0); + return upb_downcast_msgdef_mutable(upb_filedef_mutabledef(r->file, index)); +} + +static upb_def *upb_descreader_last(upb_descreader *r) { + return upb_filedef_mutabledef(r->file, upb_filedef_defcount(r->file) - 1); +} + +/* Start/end handlers for FileDescriptorProto and DescriptorProto (the two + * entities that have names and can contain sub-definitions. */ +void upb_descreader_startcontainer(upb_descreader *r) { + upb_descreader_frame *f = &r->stack[r->stack_len++]; + f->start = upb_filedef_defcount(r->file); + f->name = NULL; +} + +bool upb_descreader_endcontainer(upb_descreader *r) { + upb_descreader_frame *f = &r->stack[--r->stack_len]; + if (!upb_descreader_qualify(r->file, f->name, f->start)) { + return false; + } + upb_gfree(f->name); + f->name = NULL; + return true; +} + +void upb_descreader_setscopename(upb_descreader *r, char *str) { + upb_descreader_frame *f = &r->stack[r->stack_len-1]; + upb_gfree(f->name); + f->name = str; +} + +/** Handlers for google.protobuf.FileDescriptorSet. ***************************/ + +static void *fileset_startfile(void *closure, const void *hd) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + r->file = upb_filedef_new(&r->files); + upb_inttable_push(&r->files, upb_value_ptr(r->file)); + return r; +} + +/** Handlers for google.protobuf.FileDescriptorProto. *************************/ + +static bool file_start(void *closure, const void *hd) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + upb_descreader_startcontainer(r); + return true; +} + +static bool file_end(void *closure, const void *hd, upb_status *status) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + UPB_UNUSED(status); + return upb_descreader_endcontainer(r); +} + +static size_t file_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *name; + bool ok; + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + name = upb_strndup(buf, n); + /* XXX: see comment at the top of the file. */ + ok = upb_filedef_setname(r->file, name, NULL); + upb_gfree(name); + UPB_ASSERT_VAR(ok, ok); + return n; +} + +static size_t file_onpackage(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *package; + bool ok; + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + package = upb_strndup(buf, n); + /* XXX: see comment at the top of the file. */ + upb_descreader_setscopename(r, package); + ok = upb_filedef_setpackage(r->file, package, NULL); + UPB_ASSERT_VAR(ok, ok); + return n; +} + +static size_t file_onsyntax(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + bool ok; + UPB_UNUSED(hd); + UPB_UNUSED(handle); + /* XXX: see comment at the top of the file. */ + if (upb_streq("proto2", buf, n)) { + ok = upb_filedef_setsyntax(r->file, UPB_SYNTAX_PROTO2, NULL); + } else if (upb_streq("proto3", buf, n)) { + ok = upb_filedef_setsyntax(r->file, UPB_SYNTAX_PROTO3, NULL); + } else { + ok = false; + } + + UPB_ASSERT_VAR(ok, ok); + return n; +} + +static void *file_startmsg(void *closure, const void *hd) { + upb_descreader *r = closure; + upb_msgdef *m = upb_msgdef_new(&m); + bool ok = upb_filedef_addmsg(r->file, m, &m, NULL); + UPB_UNUSED(hd); + UPB_ASSERT_VAR(ok, ok); + return r; +} + +static void *file_startenum(void *closure, const void *hd) { + upb_descreader *r = closure; + upb_enumdef *e = upb_enumdef_new(&e); + bool ok = upb_filedef_addenum(r->file, e, &e, NULL); + UPB_UNUSED(hd); + UPB_ASSERT_VAR(ok, ok); + return r; +} + +static void *file_startext(void *closure, const void *hd) { + upb_descreader *r = closure; + bool ok; + r->f = upb_fielddef_new(r); + ok = upb_filedef_addext(r->file, r->f, r, NULL); + UPB_UNUSED(hd); + UPB_ASSERT_VAR(ok, ok); + return r; +} + +/** Handlers for google.protobuf.EnumValueDescriptorProto. *********************/ + +static bool enumval_startmsg(void *closure, const void *hd) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + r->saw_number = false; + r->saw_name = false; + return true; +} + +static size_t enumval_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + UPB_UNUSED(handle); + /* XXX: see comment at the top of the file. */ + upb_gfree(r->name); + r->name = upb_strndup(buf, n); + r->saw_name = true; + return n; +} + +static bool enumval_onnumber(void *closure, const void *hd, int32_t val) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + r->number = val; + r->saw_number = true; + return true; +} + +static bool enumval_endmsg(void *closure, const void *hd, upb_status *status) { + upb_descreader *r = closure; + upb_enumdef *e; + UPB_UNUSED(hd); + + if(!r->saw_number || !r->saw_name) { + upb_status_seterrmsg(status, "Enum value missing name or number."); + return false; + } + e = upb_downcast_enumdef_mutable(upb_descreader_last(r)); + upb_enumdef_addval(e, r->name, r->number, status); + upb_gfree(r->name); + r->name = NULL; + return true; +} + +/** Handlers for google.protobuf.EnumDescriptorProto. *************************/ + +static bool enum_endmsg(void *closure, const void *hd, upb_status *status) { + upb_descreader *r = closure; + upb_enumdef *e; + UPB_UNUSED(hd); + + e = upb_downcast_enumdef_mutable(upb_descreader_last(r)); + if (upb_def_fullname(upb_descreader_last(r)) == NULL) { + upb_status_seterrmsg(status, "Enum had no name."); + return false; + } + if (upb_enumdef_numvals(e) == 0) { + upb_status_seterrmsg(status, "Enum had no values."); + return false; + } + return true; +} + +static size_t enum_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *fullname = upb_strndup(buf, n); + UPB_UNUSED(hd); + UPB_UNUSED(handle); + /* XXX: see comment at the top of the file. */ + upb_def_setfullname(upb_descreader_last(r), fullname, NULL); + upb_gfree(fullname); + return n; +} + +/** Handlers for google.protobuf.FieldDescriptorProto *************************/ + +static bool field_startmsg(void *closure, const void *hd) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + assert(r->f); + upb_gfree(r->default_string); + r->default_string = NULL; + + /* fielddefs default to packed, but descriptors default to non-packed. */ + upb_fielddef_setpacked(r->f, false); + return true; +} + +/* Converts the default value in string "str" into "d". Passes a ref on str. + * Returns true on success. */ +static bool parse_default(char *str, upb_fielddef *f) { + bool success = true; + char *end; + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: { + long val = strtol(str, &end, 0); + if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultint32(f, val); + break; + } + case UPB_TYPE_INT64: { + /* XXX: Need to write our own strtoll, since it's not available in c89. */ + long long val = strtol(str, &end, 0); + if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultint64(f, val); + break; + } + case UPB_TYPE_UINT32: { + unsigned long val = strtoul(str, &end, 0); + if (val > UINT32_MAX || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultuint32(f, val); + break; + } + case UPB_TYPE_UINT64: { + /* XXX: Need to write our own strtoull, since it's not available in c89. */ + unsigned long long val = strtoul(str, &end, 0); + if (val > UINT64_MAX || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultuint64(f, val); + break; + } + case UPB_TYPE_DOUBLE: { + double val = strtod(str, &end); + if (errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultdouble(f, val); + break; + } + case UPB_TYPE_FLOAT: { + /* XXX: Need to write our own strtof, since it's not available in c89. */ + float val = strtod(str, &end); + if (errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultfloat(f, val); + break; + } + case UPB_TYPE_BOOL: { + if (strcmp(str, "false") == 0) + upb_fielddef_setdefaultbool(f, false); + else if (strcmp(str, "true") == 0) + upb_fielddef_setdefaultbool(f, true); + else + success = false; + break; + } + default: abort(); + } + return success; +} + +static bool field_endmsg(void *closure, const void *hd, upb_status *status) { + upb_descreader *r = closure; + upb_fielddef *f = r->f; + UPB_UNUSED(hd); + + /* TODO: verify that all required fields were present. */ + assert(upb_fielddef_number(f) != 0); + assert(upb_fielddef_name(f) != NULL); + assert((upb_fielddef_subdefname(f) != NULL) == upb_fielddef_hassubdef(f)); + + if (r->default_string) { + if (upb_fielddef_issubmsg(f)) { + upb_status_seterrmsg(status, "Submessages cannot have defaults."); + return false; + } + if (upb_fielddef_isstring(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM) { + upb_fielddef_setdefaultcstr(f, r->default_string, NULL); + } else { + if (r->default_string && !parse_default(r->default_string, f)) { + /* We don't worry too much about giving a great error message since the + * compiler should have ensured this was correct. */ + upb_status_seterrmsg(status, "Error converting default value."); + return false; + } + } + } + return true; +} + +static bool field_onlazy(void *closure, const void *hd, bool val) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + + upb_fielddef_setlazy(r->f, val); + return true; +} + +static bool field_onpacked(void *closure, const void *hd, bool val) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + + upb_fielddef_setpacked(r->f, val); + return true; +} + +static bool field_ontype(void *closure, const void *hd, int32_t val) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + + upb_fielddef_setdescriptortype(r->f, val); + return true; +} + +static bool field_onlabel(void *closure, const void *hd, int32_t val) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + + upb_fielddef_setlabel(r->f, val); + return true; +} + +static bool field_onnumber(void *closure, const void *hd, int32_t val) { + upb_descreader *r = closure; + bool ok; + UPB_UNUSED(hd); + + ok = upb_fielddef_setnumber(r->f, val, NULL); + UPB_ASSERT_VAR(ok, ok); + return true; +} + +static size_t field_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *name = upb_strndup(buf, n); + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + /* XXX: see comment at the top of the file. */ + upb_fielddef_setname(r->f, name, NULL); + upb_gfree(name); + return n; +} + +static size_t field_ontypename(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *name = upb_strndup(buf, n); + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + /* XXX: see comment at the top of the file. */ + upb_fielddef_setsubdefname(r->f, name, NULL); + upb_gfree(name); + return n; +} + +static size_t field_onextendee(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + char *name = upb_strndup(buf, n); + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + /* XXX: see comment at the top of the file. */ + upb_fielddef_setcontainingtypename(r->f, name, NULL); + upb_gfree(name); + return n; +} + +static size_t field_ondefaultval(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + /* Have to convert from string to the correct type, but we might not know the + * type yet, so we save it as a string until the end of the field. + * XXX: see comment at the top of the file. */ + upb_gfree(r->default_string); + r->default_string = upb_strndup(buf, n); + return n; +} + +/** Handlers for google.protobuf.DescriptorProto ******************************/ + +static bool msg_start(void *closure, const void *hd) { + upb_descreader *r = closure; + UPB_UNUSED(hd); + + upb_descreader_startcontainer(r); + return true; +} + +static bool msg_end(void *closure, const void *hd, upb_status *status) { + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + UPB_UNUSED(hd); + + if(!upb_def_fullname(upb_msgdef_upcast_mutable(m))) { + upb_status_seterrmsg(status, "Encountered message with no name."); + return false; + } + return upb_descreader_endcontainer(r); +} + +static size_t msg_name(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + /* XXX: see comment at the top of the file. */ + char *name = upb_strndup(buf, n); + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + upb_def_setfullname(upb_msgdef_upcast_mutable(m), name, NULL); + upb_descreader_setscopename(r, name); /* Passes ownership of name. */ + return n; +} + +static void *msg_startmsg(void *closure, const void *hd) { + upb_descreader *r = closure; + upb_msgdef *m = upb_msgdef_new(&m); + bool ok = upb_filedef_addmsg(r->file, m, &m, NULL); + UPB_UNUSED(hd); + UPB_ASSERT_VAR(ok, ok); + return r; +} + +static void *msg_startext(void *closure, const void *hd) { + upb_descreader *r = closure; + upb_fielddef *f = upb_fielddef_new(&f); + bool ok = upb_filedef_addext(r->file, f, &f, NULL); + UPB_UNUSED(hd); + UPB_ASSERT_VAR(ok, ok); + return r; +} + +static void *msg_startfield(void *closure, const void *hd) { + upb_descreader *r = closure; + r->f = upb_fielddef_new(&r->f); + /* We can't add the new field to the message until its name/number are + * filled in. */ + UPB_UNUSED(hd); + return r; +} + +static bool msg_endfield(void *closure, const void *hd) { + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + UPB_UNUSED(hd); + + upb_msgdef_addfield(m, r->f, &r->f, NULL); + r->f = NULL; + return true; +} + +static bool msg_onmapentry(void *closure, const void *hd, bool mapentry) { + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + UPB_UNUSED(hd); + + upb_msgdef_setmapentry(m, mapentry); + r->f = NULL; + return true; +} + + + +/** Code to register handlers *************************************************/ + +#define F(msg, field) upbdefs_google_protobuf_ ## msg ## _f_ ## field(m) + +static void reghandlers(const void *closure, upb_handlers *h) { + const upb_msgdef *m = upb_handlers_msgdef(h); + UPB_UNUSED(closure); + + if (upbdefs_google_protobuf_FileDescriptorSet_is(m)) { + upb_handlers_setstartsubmsg(h, F(FileDescriptorSet, file), + &fileset_startfile, NULL); + } else if (upbdefs_google_protobuf_DescriptorProto_is(m)) { + upb_handlers_setstartmsg(h, &msg_start, NULL); + upb_handlers_setendmsg(h, &msg_end, NULL); + upb_handlers_setstring(h, F(DescriptorProto, name), &msg_name, NULL); + upb_handlers_setstartsubmsg(h, F(DescriptorProto, extension), &msg_startext, + NULL); + upb_handlers_setstartsubmsg(h, F(DescriptorProto, nested_type), + &msg_startmsg, NULL); + upb_handlers_setstartsubmsg(h, F(DescriptorProto, field), + &msg_startfield, NULL); + upb_handlers_setendsubmsg(h, F(DescriptorProto, field), + &msg_endfield, NULL); + upb_handlers_setstartsubmsg(h, F(DescriptorProto, enum_type), + &file_startenum, NULL); + } else if (upbdefs_google_protobuf_FileDescriptorProto_is(m)) { + upb_handlers_setstartmsg(h, &file_start, NULL); + upb_handlers_setendmsg(h, &file_end, NULL); + upb_handlers_setstring(h, F(FileDescriptorProto, name), &file_onname, + NULL); + upb_handlers_setstring(h, F(FileDescriptorProto, package), &file_onpackage, + NULL); + upb_handlers_setstring(h, F(FileDescriptorProto, syntax), &file_onsyntax, + NULL); + upb_handlers_setstartsubmsg(h, F(FileDescriptorProto, message_type), + &file_startmsg, NULL); + upb_handlers_setstartsubmsg(h, F(FileDescriptorProto, enum_type), + &file_startenum, NULL); + upb_handlers_setstartsubmsg(h, F(FileDescriptorProto, extension), + &file_startext, NULL); + } else if (upbdefs_google_protobuf_EnumValueDescriptorProto_is(m)) { + upb_handlers_setstartmsg(h, &enumval_startmsg, NULL); + upb_handlers_setendmsg(h, &enumval_endmsg, NULL); + upb_handlers_setstring(h, F(EnumValueDescriptorProto, name), &enumval_onname, NULL); + upb_handlers_setint32(h, F(EnumValueDescriptorProto, number), &enumval_onnumber, + NULL); + } else if (upbdefs_google_protobuf_EnumDescriptorProto_is(m)) { + upb_handlers_setendmsg(h, &enum_endmsg, NULL); + upb_handlers_setstring(h, F(EnumDescriptorProto, name), &enum_onname, NULL); + } else if (upbdefs_google_protobuf_FieldDescriptorProto_is(m)) { + upb_handlers_setstartmsg(h, &field_startmsg, NULL); + upb_handlers_setendmsg(h, &field_endmsg, NULL); + upb_handlers_setint32(h, F(FieldDescriptorProto, type), &field_ontype, + NULL); + upb_handlers_setint32(h, F(FieldDescriptorProto, label), &field_onlabel, + NULL); + upb_handlers_setint32(h, F(FieldDescriptorProto, number), &field_onnumber, + NULL); + upb_handlers_setstring(h, F(FieldDescriptorProto, name), &field_onname, + NULL); + upb_handlers_setstring(h, F(FieldDescriptorProto, type_name), + &field_ontypename, NULL); + upb_handlers_setstring(h, F(FieldDescriptorProto, extendee), + &field_onextendee, NULL); + upb_handlers_setstring(h, F(FieldDescriptorProto, default_value), + &field_ondefaultval, NULL); + } else if (upbdefs_google_protobuf_FieldOptions_is(m)) { + upb_handlers_setbool(h, F(FieldOptions, lazy), &field_onlazy, NULL); + upb_handlers_setbool(h, F(FieldOptions, packed), &field_onpacked, NULL); + } else if (upbdefs_google_protobuf_MessageOptions_is(m)) { + upb_handlers_setbool(h, F(MessageOptions, map_entry), &msg_onmapentry, NULL); + } + + assert(upb_ok(upb_handlers_status(h))); +} + +#undef F + +void descreader_cleanup(void *_r) { + upb_descreader *r = _r; + size_t i; + + for (i = 0; i < upb_descreader_filecount(r); i++) { + upb_filedef_unref(upb_descreader_file(r, i), &r->files); + } + + upb_gfree(r->name); + upb_inttable_uninit(&r->files); + upb_gfree(r->default_string); + while (r->stack_len > 0) { + upb_descreader_frame *f = &r->stack[--r->stack_len]; + upb_gfree(f->name); + } +} + + +/* Public API ****************************************************************/ + +upb_descreader *upb_descreader_create(upb_env *e, const upb_handlers *h) { + upb_descreader *r = upb_env_malloc(e, sizeof(upb_descreader)); + if (!r || !upb_env_addcleanup(e, descreader_cleanup, r)) { + return NULL; + } + + upb_inttable_init(&r->files, UPB_CTYPE_PTR); + upb_sink_reset(upb_descreader_input(r), h, r); + r->stack_len = 0; + r->name = NULL; + r->default_string = NULL; + + return r; +} + +size_t upb_descreader_filecount(const upb_descreader *r) { + return upb_inttable_count(&r->files); +} + +upb_filedef *upb_descreader_file(const upb_descreader *r, size_t i) { + upb_value v; + if (upb_inttable_lookup(&r->files, i, &v)) { + return upb_value_getptr(v); + } else { + return NULL; + } +} + +upb_sink *upb_descreader_input(upb_descreader *r) { + return &r->sink; +} + +const upb_handlers *upb_descreader_newhandlers(const void *owner) { + const upb_msgdef *m = upbdefs_google_protobuf_FileDescriptorSet_get(&m); + const upb_handlers *h = upb_handlers_newfrozen(m, owner, reghandlers, NULL); + upb_msgdef_unref(m, &m); + return h; +} +/* +** protobuf decoder bytecode compiler +** +** Code to compile a upb::Handlers into bytecode for decoding a protobuf +** according to that specific schema and destination handlers. +** +** Compiling to bytecode is always the first step. If we are using the +** interpreted decoder we leave it as bytecode and interpret that. If we are +** using a JIT decoder we use a code generator to turn the bytecode into native +** code, LLVM IR, etc. +** +** Bytecode definition is in decoder.int.h. +*/ + +#include <stdarg.h> + +#ifdef UPB_DUMP_BYTECODE +#include <stdio.h> +#endif + +#define MAXLABEL 5 +#define EMPTYLABEL -1 + +/* mgroup *********************************************************************/ + +static void freegroup(upb_refcounted *r) { + mgroup *g = (mgroup*)r; + upb_inttable_uninit(&g->methods); +#ifdef UPB_USE_JIT_X64 + upb_pbdecoder_freejit(g); +#endif + upb_gfree(g->bytecode); + upb_gfree(g); +} + +static void visitgroup(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const mgroup *g = (const mgroup*)r; + upb_inttable_iter i; + upb_inttable_begin(&i, &g->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i)); + visit(r, upb_pbdecodermethod_upcast(method), closure); + } +} + +mgroup *newgroup(const void *owner) { + mgroup *g = upb_gmalloc(sizeof(*g)); + static const struct upb_refcounted_vtbl vtbl = {visitgroup, freegroup}; + upb_refcounted_init(mgroup_upcast_mutable(g), &vtbl, owner); + upb_inttable_init(&g->methods, UPB_CTYPE_PTR); + g->bytecode = NULL; + g->bytecode_end = NULL; + return g; +} + + +/* upb_pbdecodermethod ********************************************************/ + +static void freemethod(upb_refcounted *r) { + upb_pbdecodermethod *method = (upb_pbdecodermethod*)r; + + if (method->dest_handlers_) { + upb_handlers_unref(method->dest_handlers_, method); + } + + upb_inttable_uninit(&method->dispatch); + upb_gfree(method); +} + +static void visitmethod(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_pbdecodermethod *m = (const upb_pbdecodermethod*)r; + visit(r, m->group, closure); +} + +static upb_pbdecodermethod *newmethod(const upb_handlers *dest_handlers, + mgroup *group) { + static const struct upb_refcounted_vtbl vtbl = {visitmethod, freemethod}; + upb_pbdecodermethod *ret = upb_gmalloc(sizeof(*ret)); + upb_refcounted_init(upb_pbdecodermethod_upcast_mutable(ret), &vtbl, &ret); + upb_byteshandler_init(&ret->input_handler_); + + /* The method references the group and vice-versa, in a circular reference. */ + upb_ref2(ret, group); + upb_ref2(group, ret); + upb_inttable_insertptr(&group->methods, dest_handlers, upb_value_ptr(ret)); + upb_pbdecodermethod_unref(ret, &ret); + + ret->group = mgroup_upcast_mutable(group); + ret->dest_handlers_ = dest_handlers; + ret->is_native_ = false; /* If we JIT, it will update this later. */ + upb_inttable_init(&ret->dispatch, UPB_CTYPE_UINT64); + + if (ret->dest_handlers_) { + upb_handlers_ref(ret->dest_handlers_, ret); + } + return ret; +} + +const upb_handlers *upb_pbdecodermethod_desthandlers( + const upb_pbdecodermethod *m) { + return m->dest_handlers_; +} + +const upb_byteshandler *upb_pbdecodermethod_inputhandler( + const upb_pbdecodermethod *m) { + return &m->input_handler_; +} + +bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m) { + return m->is_native_; +} + +const upb_pbdecodermethod *upb_pbdecodermethod_new( + const upb_pbdecodermethodopts *opts, const void *owner) { + const upb_pbdecodermethod *ret; + upb_pbcodecache cache; + + upb_pbcodecache_init(&cache); + ret = upb_pbcodecache_getdecodermethod(&cache, opts); + upb_pbdecodermethod_ref(ret, owner); + upb_pbcodecache_uninit(&cache); + return ret; +} + + +/* bytecode compiler **********************************************************/ + +/* Data used only at compilation time. */ +typedef struct { + mgroup *group; + + uint32_t *pc; + int fwd_labels[MAXLABEL]; + int back_labels[MAXLABEL]; + + /* For fields marked "lazy", parse them lazily or eagerly? */ + bool lazy; +} compiler; + +static compiler *newcompiler(mgroup *group, bool lazy) { + compiler *ret = upb_gmalloc(sizeof(*ret)); + int i; + + ret->group = group; + ret->lazy = lazy; + for (i = 0; i < MAXLABEL; i++) { + ret->fwd_labels[i] = EMPTYLABEL; + ret->back_labels[i] = EMPTYLABEL; + } + return ret; +} + +static void freecompiler(compiler *c) { + upb_gfree(c); +} + +const size_t ptr_words = sizeof(void*) / sizeof(uint32_t); + +/* How many words an instruction is. */ +static int instruction_len(uint32_t instr) { + switch (getop(instr)) { + case OP_SETDISPATCH: return 1 + ptr_words; + case OP_TAGN: return 3; + case OP_SETBIGGROUPNUM: return 2; + default: return 1; + } +} + +bool op_has_longofs(int32_t instruction) { + switch (getop(instruction)) { + case OP_CALL: + case OP_BRANCH: + case OP_CHECKDELIM: + return true; + /* The "tag" instructions only have 8 bytes available for the jump target, + * but that is ok because these opcodes only require short jumps. */ + case OP_TAG1: + case OP_TAG2: + case OP_TAGN: + return false; + default: + assert(false); + return false; + } +} + +static int32_t getofs(uint32_t instruction) { + if (op_has_longofs(instruction)) { + return (int32_t)instruction >> 8; + } else { + return (int8_t)(instruction >> 8); + } +} + +static void setofs(uint32_t *instruction, int32_t ofs) { + if (op_has_longofs(*instruction)) { + *instruction = getop(*instruction) | ofs << 8; + } else { + *instruction = (*instruction & ~0xff00) | ((ofs & 0xff) << 8); + } + assert(getofs(*instruction) == ofs); /* Would fail in cases of overflow. */ +} + +static uint32_t pcofs(compiler *c) { return c->pc - c->group->bytecode; } + +/* Defines a local label at the current PC location. All previous forward + * references are updated to point to this location. The location is noted + * for any future backward references. */ +static void label(compiler *c, unsigned int label) { + int val; + uint32_t *codep; + + assert(label < MAXLABEL); + val = c->fwd_labels[label]; + codep = (val == EMPTYLABEL) ? NULL : c->group->bytecode + val; + while (codep) { + int ofs = getofs(*codep); + setofs(codep, c->pc - codep - instruction_len(*codep)); + codep = ofs ? codep + ofs : NULL; + } + c->fwd_labels[label] = EMPTYLABEL; + c->back_labels[label] = pcofs(c); +} + +/* Creates a reference to a numbered label; either a forward reference + * (positive arg) or backward reference (negative arg). For forward references + * the value returned now is actually a "next" pointer into a linked list of all + * instructions that use this label and will be patched later when the label is + * defined with label(). + * + * The returned value is the offset that should be written into the instruction. + */ +static int32_t labelref(compiler *c, int label) { + assert(label < MAXLABEL); + if (label == LABEL_DISPATCH) { + /* No resolving required. */ + return 0; + } else if (label < 0) { + /* Backward local label. Relative to the next instruction. */ + uint32_t from = (c->pc + 1) - c->group->bytecode; + return c->back_labels[-label] - from; + } else { + /* Forward local label: prepend to (possibly-empty) linked list. */ + int *lptr = &c->fwd_labels[label]; + int32_t ret = (*lptr == EMPTYLABEL) ? 0 : *lptr - pcofs(c); + *lptr = pcofs(c); + return ret; + } +} + +static void put32(compiler *c, uint32_t v) { + mgroup *g = c->group; + if (c->pc == g->bytecode_end) { + int ofs = pcofs(c); + size_t oldsize = g->bytecode_end - g->bytecode; + size_t newsize = UPB_MAX(oldsize * 2, 64); + /* TODO(haberman): handle OOM. */ + g->bytecode = upb_grealloc(g->bytecode, oldsize * sizeof(uint32_t), + newsize * sizeof(uint32_t)); + g->bytecode_end = g->bytecode + newsize; + c->pc = g->bytecode + ofs; + } + *c->pc++ = v; +} + +static void putop(compiler *c, opcode op, ...) { + va_list ap; + va_start(ap, op); + + switch (op) { + case OP_SETDISPATCH: { + uintptr_t ptr = (uintptr_t)va_arg(ap, void*); + put32(c, OP_SETDISPATCH); + put32(c, ptr); + if (sizeof(uintptr_t) > sizeof(uint32_t)) + put32(c, (uint64_t)ptr >> 32); + break; + } + case OP_STARTMSG: + case OP_ENDMSG: + case OP_PUSHLENDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_HALT: + case OP_RET: + case OP_DISPATCH: + put32(c, op); + break; + case OP_PARSE_DOUBLE: + case OP_PARSE_FLOAT: + case OP_PARSE_INT64: + case OP_PARSE_UINT64: + case OP_PARSE_INT32: + case OP_PARSE_FIXED64: + case OP_PARSE_FIXED32: + case OP_PARSE_BOOL: + case OP_PARSE_UINT32: + case OP_PARSE_SFIXED32: + case OP_PARSE_SFIXED64: + case OP_PARSE_SINT32: + case OP_PARSE_SINT64: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_STRING: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + put32(c, op | va_arg(ap, upb_selector_t) << 8); + break; + case OP_SETBIGGROUPNUM: + put32(c, op); + put32(c, va_arg(ap, int)); + break; + case OP_CALL: { + const upb_pbdecodermethod *method = va_arg(ap, upb_pbdecodermethod *); + put32(c, op | (method->code_base.ofs - (pcofs(c) + 1)) << 8); + break; + } + case OP_CHECKDELIM: + case OP_BRANCH: { + uint32_t instruction = op; + int label = va_arg(ap, int); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + break; + } + case OP_TAG1: + case OP_TAG2: { + int label = va_arg(ap, int); + uint64_t tag = va_arg(ap, uint64_t); + uint32_t instruction = op | (tag << 16); + assert(tag <= 0xffff); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + break; + } + case OP_TAGN: { + int label = va_arg(ap, int); + uint64_t tag = va_arg(ap, uint64_t); + uint32_t instruction = op | (upb_value_size(tag) << 16); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + put32(c, tag); + put32(c, tag >> 32); + break; + } + } + + va_end(ap); +} + +#if defined(UPB_USE_JIT_X64) || defined(UPB_DUMP_BYTECODE) + +const char *upb_pbdecoder_getopname(unsigned int op) { +#define QUOTE(x) #x +#define EXPAND_AND_QUOTE(x) QUOTE(x) +#define OPNAME(x) OP_##x +#define OP(x) case OPNAME(x): return EXPAND_AND_QUOTE(OPNAME(x)); +#define T(x) OP(PARSE_##x) + /* Keep in sync with list in decoder.int.h. */ + switch ((opcode)op) { + T(DOUBLE) T(FLOAT) T(INT64) T(UINT64) T(INT32) T(FIXED64) T(FIXED32) + T(BOOL) T(UINT32) T(SFIXED32) T(SFIXED64) T(SINT32) T(SINT64) + OP(STARTMSG) OP(ENDMSG) OP(STARTSEQ) OP(ENDSEQ) OP(STARTSUBMSG) + OP(ENDSUBMSG) OP(STARTSTR) OP(STRING) OP(ENDSTR) OP(CALL) OP(RET) + OP(PUSHLENDELIM) OP(PUSHTAGDELIM) OP(SETDELIM) OP(CHECKDELIM) + OP(BRANCH) OP(TAG1) OP(TAG2) OP(TAGN) OP(SETDISPATCH) OP(POP) + OP(SETBIGGROUPNUM) OP(DISPATCH) OP(HALT) + } + return "<unknown op>"; +#undef OP +#undef T +} + +#endif + +#ifdef UPB_DUMP_BYTECODE + +static void dumpbc(uint32_t *p, uint32_t *end, FILE *f) { + + uint32_t *begin = p; + + while (p < end) { + fprintf(f, "%p %8tx", p, p - begin); + uint32_t instr = *p++; + uint8_t op = getop(instr); + fprintf(f, " %s", upb_pbdecoder_getopname(op)); + switch ((opcode)op) { + case OP_SETDISPATCH: { + const upb_inttable *dispatch; + memcpy(&dispatch, p, sizeof(void*)); + p += ptr_words; + const upb_pbdecodermethod *method = + (void *)((char *)dispatch - + offsetof(upb_pbdecodermethod, dispatch)); + fprintf(f, " %s", upb_msgdef_fullname( + upb_handlers_msgdef(method->dest_handlers_))); + break; + } + case OP_DISPATCH: + case OP_STARTMSG: + case OP_ENDMSG: + case OP_PUSHLENDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_HALT: + case OP_RET: + break; + case OP_PARSE_DOUBLE: + case OP_PARSE_FLOAT: + case OP_PARSE_INT64: + case OP_PARSE_UINT64: + case OP_PARSE_INT32: + case OP_PARSE_FIXED64: + case OP_PARSE_FIXED32: + case OP_PARSE_BOOL: + case OP_PARSE_UINT32: + case OP_PARSE_SFIXED32: + case OP_PARSE_SFIXED64: + case OP_PARSE_SINT32: + case OP_PARSE_SINT64: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_STRING: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + fprintf(f, " %d", instr >> 8); + break; + case OP_SETBIGGROUPNUM: + fprintf(f, " %d", *p++); + break; + case OP_CHECKDELIM: + case OP_CALL: + case OP_BRANCH: + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + break; + case OP_TAG1: + case OP_TAG2: { + fprintf(f, " tag:0x%x", instr >> 16); + if (getofs(instr)) { + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + } + break; + } + case OP_TAGN: { + uint64_t tag = *p++; + tag |= (uint64_t)*p++ << 32; + fprintf(f, " tag:0x%llx", (long long)tag); + fprintf(f, " n:%d", instr >> 16); + if (getofs(instr)) { + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + } + break; + } + } + fputs("\n", f); + } +} + +#endif + +static uint64_t get_encoded_tag(const upb_fielddef *f, int wire_type) { + uint32_t tag = (upb_fielddef_number(f) << 3) | wire_type; + uint64_t encoded_tag = upb_vencode32(tag); + /* No tag should be greater than 5 bytes. */ + assert(encoded_tag <= 0xffffffffff); + return encoded_tag; +} + +static void putchecktag(compiler *c, const upb_fielddef *f, + int wire_type, int dest) { + uint64_t tag = get_encoded_tag(f, wire_type); + switch (upb_value_size(tag)) { + case 1: + putop(c, OP_TAG1, dest, tag); + break; + case 2: + putop(c, OP_TAG2, dest, tag); + break; + default: + putop(c, OP_TAGN, dest, tag); + break; + } +} + +static upb_selector_t getsel(const upb_fielddef *f, upb_handlertype_t type) { + upb_selector_t selector; + bool ok = upb_handlers_getselector(f, type, &selector); + UPB_ASSERT_VAR(ok, ok); + return selector; +} + +/* Takes an existing, primary dispatch table entry and repacks it with a + * different alternate wire type. Called when we are inserting a secondary + * dispatch table entry for an alternate wire type. */ +static uint64_t repack(uint64_t dispatch, int new_wt2) { + uint64_t ofs; + uint8_t wt1; + uint8_t old_wt2; + upb_pbdecoder_unpackdispatch(dispatch, &ofs, &wt1, &old_wt2); + assert(old_wt2 == NO_WIRE_TYPE); /* wt2 should not be set yet. */ + return upb_pbdecoder_packdispatch(ofs, wt1, new_wt2); +} + +/* Marks the current bytecode position as the dispatch target for this message, + * field, and wire type. */ +static void dispatchtarget(compiler *c, upb_pbdecodermethod *method, + const upb_fielddef *f, int wire_type) { + /* Offset is relative to msg base. */ + uint64_t ofs = pcofs(c) - method->code_base.ofs; + uint32_t fn = upb_fielddef_number(f); + upb_inttable *d = &method->dispatch; + upb_value v; + if (upb_inttable_remove(d, fn, &v)) { + /* TODO: prioritize based on packed setting in .proto file. */ + uint64_t repacked = repack(upb_value_getuint64(v), wire_type); + upb_inttable_insert(d, fn, upb_value_uint64(repacked)); + upb_inttable_insert(d, fn + UPB_MAX_FIELDNUMBER, upb_value_uint64(ofs)); + } else { + uint64_t val = upb_pbdecoder_packdispatch(ofs, wire_type, NO_WIRE_TYPE); + upb_inttable_insert(d, fn, upb_value_uint64(val)); + } +} + +static void putpush(compiler *c, const upb_fielddef *f) { + if (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE) { + putop(c, OP_PUSHLENDELIM); + } else { + uint32_t fn = upb_fielddef_number(f); + if (fn >= 1 << 24) { + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_SETBIGGROUPNUM, fn); + } else { + putop(c, OP_PUSHTAGDELIM, fn); + } + } +} + +static upb_pbdecodermethod *find_submethod(const compiler *c, + const upb_pbdecodermethod *method, + const upb_fielddef *f) { + const upb_handlers *sub = + upb_handlers_getsubhandlers(method->dest_handlers_, f); + upb_value v; + return upb_inttable_lookupptr(&c->group->methods, sub, &v) + ? upb_value_getptr(v) + : NULL; +} + +static void putsel(compiler *c, opcode op, upb_selector_t sel, + const upb_handlers *h) { + if (upb_handlers_gethandler(h, sel)) { + putop(c, op, sel); + } +} + +/* Puts an opcode to call a callback, but only if a callback actually exists for + * this field and handler type. */ +static void maybeput(compiler *c, opcode op, const upb_handlers *h, + const upb_fielddef *f, upb_handlertype_t type) { + putsel(c, op, getsel(f, type), h); +} + +static bool haslazyhandlers(const upb_handlers *h, const upb_fielddef *f) { + if (!upb_fielddef_lazy(f)) + return false; + + return upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STARTSTR)) || + upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STRING)) || + upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_ENDSTR)); +} + + +/* bytecode compiler code generation ******************************************/ + +/* Symbolic names for our local labels. */ +#define LABEL_LOOPSTART 1 /* Top of a repeated field loop. */ +#define LABEL_LOOPBREAK 2 /* To jump out of a repeated loop */ +#define LABEL_FIELD 3 /* Jump backward to find the most recent field. */ +#define LABEL_ENDMSG 4 /* To reach the OP_ENDMSG instr for this msg. */ + +/* Generates bytecode to parse a single non-lazy message field. */ +static void generate_msgfield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + const upb_pbdecodermethod *sub_m = find_submethod(c, method, f); + int wire_type; + + if (!sub_m) { + /* Don't emit any code for this field at all; it will be parsed as an + * unknown field. + * + * TODO(haberman): we should change this to parse it as a string field + * instead. It will probably be faster, but more importantly, once we + * start vending unknown fields, a field shouldn't be treated as unknown + * just because it doesn't have subhandlers registered. */ + return; + } + + label(c, LABEL_FIELD); + + wire_type = + (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE) + ? UPB_WIRE_TYPE_DELIMITED + : UPB_WIRE_TYPE_START_GROUP; + + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); + label(c, LABEL_LOOPSTART); + putpush(c, f); + putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG)); + putop(c, OP_CALL, sub_m); + putop(c, OP_POP); + maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG); + if (wire_type == UPB_WIRE_TYPE_DELIMITED) { + putop(c, OP_SETDELIM); + } + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, wire_type, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putpush(c, f); + putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG)); + putop(c, OP_CALL, sub_m); + putop(c, OP_POP); + maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG); + if (wire_type == UPB_WIRE_TYPE_DELIMITED) { + putop(c, OP_SETDELIM); + } + } +} + +/* Generates bytecode to parse a single string or lazy submessage field. */ +static void generate_delimfield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + + label(c, LABEL_FIELD); + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); + label(c, LABEL_LOOPSTART); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR)); + /* Need to emit even if no handler to skip past the string. */ + putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING)); + putop(c, OP_POP); + maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR); + putop(c, OP_SETDELIM); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR)); + putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING)); + putop(c, OP_POP); + maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR); + putop(c, OP_SETDELIM); + } +} + +/* Generates bytecode to parse a single primitive field. */ +static void generate_primitivefield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + upb_descriptortype_t descriptor_type = upb_fielddef_descriptortype(f); + opcode parse_type; + upb_selector_t sel; + int wire_type; + + label(c, LABEL_FIELD); + + /* From a decoding perspective, ENUM is the same as INT32. */ + if (descriptor_type == UPB_DESCRIPTOR_TYPE_ENUM) + descriptor_type = UPB_DESCRIPTOR_TYPE_INT32; + + parse_type = (opcode)descriptor_type; + + /* TODO(haberman): generate packed or non-packed first depending on "packed" + * setting in the fielddef. This will favor (in speed) whichever was + * specified. */ + + assert((int)parse_type >= 0 && parse_type <= OP_MAX); + sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + wire_type = upb_pb_native_wire_types[upb_fielddef_descriptortype(f)]; + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); /* Packed */ + label(c, LABEL_LOOPSTART); + putop(c, parse_type, sel); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + dispatchtarget(c, method, f, wire_type); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); /* Non-packed */ + label(c, LABEL_LOOPSTART); + putop(c, parse_type, sel); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, wire_type, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); /* Packed and non-packed join. */ + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + putop(c, OP_SETDELIM); /* Could remove for non-packed by dup ENDSEQ. */ + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putop(c, parse_type, sel); + } +} + +/* Adds bytecode for parsing the given message to the given decoderplan, + * while adding all dispatch targets to this message's dispatch table. */ +static void compile_method(compiler *c, upb_pbdecodermethod *method) { + const upb_handlers *h; + const upb_msgdef *md; + uint32_t* start_pc; + upb_msg_field_iter i; + upb_value val; + + assert(method); + + /* Clear all entries in the dispatch table. */ + upb_inttable_uninit(&method->dispatch); + upb_inttable_init(&method->dispatch, UPB_CTYPE_UINT64); + + h = upb_pbdecodermethod_desthandlers(method); + md = upb_handlers_msgdef(h); + + method->code_base.ofs = pcofs(c); + putop(c, OP_SETDISPATCH, &method->dispatch); + putsel(c, OP_STARTMSG, UPB_STARTMSG_SELECTOR, h); + label(c, LABEL_FIELD); + start_pc = c->pc; + for(upb_msg_field_begin(&i, md); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + upb_fieldtype_t type = upb_fielddef_type(f); + + if (type == UPB_TYPE_MESSAGE && !(haslazyhandlers(h, f) && c->lazy)) { + generate_msgfield(c, f, method); + } else if (type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES || + type == UPB_TYPE_MESSAGE) { + generate_delimfield(c, f, method); + } else { + generate_primitivefield(c, f, method); + } + } + + /* If there were no fields, or if no handlers were defined, we need to + * generate a non-empty loop body so that we can at least dispatch for unknown + * fields and check for the end of the message. */ + if (c->pc == start_pc) { + /* Check for end-of-message. */ + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + /* Unconditionally dispatch. */ + putop(c, OP_DISPATCH, 0); + } + + /* For now we just loop back to the last field of the message (or if none, + * the DISPATCH opcode for the message). */ + putop(c, OP_BRANCH, -LABEL_FIELD); + + /* Insert both a label and a dispatch table entry for this end-of-msg. */ + label(c, LABEL_ENDMSG); + val = upb_value_uint64(pcofs(c) - method->code_base.ofs); + upb_inttable_insert(&method->dispatch, DISPATCH_ENDMSG, val); + + putsel(c, OP_ENDMSG, UPB_ENDMSG_SELECTOR, h); + putop(c, OP_RET); + + upb_inttable_compact(&method->dispatch); +} + +/* Populate "methods" with new upb_pbdecodermethod objects reachable from "h". + * Returns the method for these handlers. + * + * Generates a new method for every destination handlers reachable from "h". */ +static void find_methods(compiler *c, const upb_handlers *h) { + upb_value v; + upb_msg_field_iter i; + const upb_msgdef *md; + + if (upb_inttable_lookupptr(&c->group->methods, h, &v)) + return; + newmethod(h, c->group); + + /* Find submethods. */ + md = upb_handlers_msgdef(h); + for(upb_msg_field_begin(&i, md); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + const upb_handlers *sub_h; + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE && + (sub_h = upb_handlers_getsubhandlers(h, f)) != NULL) { + /* We only generate a decoder method for submessages with handlers. + * Others will be parsed as unknown fields. */ + find_methods(c, sub_h); + } + } +} + +/* (Re-)compile bytecode for all messages in "msgs." + * Overwrites any existing bytecode in "c". */ +static void compile_methods(compiler *c) { + upb_inttable_iter i; + + /* Start over at the beginning of the bytecode. */ + c->pc = c->group->bytecode; + + upb_inttable_begin(&i, &c->group->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i)); + compile_method(c, method); + } +} + +static void set_bytecode_handlers(mgroup *g) { + upb_inttable_iter i; + upb_inttable_begin(&i, &g->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *m = upb_value_getptr(upb_inttable_iter_value(&i)); + upb_byteshandler *h = &m->input_handler_; + + m->code_base.ptr = g->bytecode + m->code_base.ofs; + + upb_byteshandler_setstartstr(h, upb_pbdecoder_startbc, m->code_base.ptr); + upb_byteshandler_setstring(h, upb_pbdecoder_decode, g); + upb_byteshandler_setendstr(h, upb_pbdecoder_end, m); + } +} + + +/* JIT setup. *****************************************************************/ + +#ifdef UPB_USE_JIT_X64 + +static void sethandlers(mgroup *g, bool allowjit) { + g->jit_code = NULL; + if (allowjit) { + /* Compile byte-code into machine code, create handlers. */ + upb_pbdecoder_jit(g); + } else { + set_bytecode_handlers(g); + } +} + +#else /* UPB_USE_JIT_X64 */ + +static void sethandlers(mgroup *g, bool allowjit) { + /* No JIT compiled in; use bytecode handlers unconditionally. */ + UPB_UNUSED(allowjit); + set_bytecode_handlers(g); +} + +#endif /* UPB_USE_JIT_X64 */ + + +/* TODO(haberman): allow this to be constructed for an arbitrary set of dest + * handlers and other mgroups (but verify we have a transitive closure). */ +const mgroup *mgroup_new(const upb_handlers *dest, bool allowjit, bool lazy, + const void *owner) { + mgroup *g; + compiler *c; + + UPB_UNUSED(allowjit); + assert(upb_handlers_isfrozen(dest)); + + g = newgroup(owner); + c = newcompiler(g, lazy); + find_methods(c, dest); + + /* We compile in two passes: + * 1. all messages are assigned relative offsets from the beginning of the + * bytecode (saved in method->code_base). + * 2. forwards OP_CALL instructions can be correctly linked since message + * offsets have been previously assigned. + * + * Could avoid the second pass by linking OP_CALL instructions somehow. */ + compile_methods(c); + compile_methods(c); + g->bytecode_end = c->pc; + freecompiler(c); + +#ifdef UPB_DUMP_BYTECODE + { + FILE *f = fopen("/tmp/upb-bytecode", "w"); + assert(f); + dumpbc(g->bytecode, g->bytecode_end, stderr); + dumpbc(g->bytecode, g->bytecode_end, f); + fclose(f); + + f = fopen("/tmp/upb-bytecode.bin", "wb"); + assert(f); + fwrite(g->bytecode, 1, g->bytecode_end - g->bytecode, f); + fclose(f); + } +#endif + + sethandlers(g, allowjit); + return g; +} + + +/* upb_pbcodecache ************************************************************/ + +void upb_pbcodecache_init(upb_pbcodecache *c) { + upb_inttable_init(&c->groups, UPB_CTYPE_CONSTPTR); + c->allow_jit_ = true; +} + +void upb_pbcodecache_uninit(upb_pbcodecache *c) { + upb_inttable_iter i; + upb_inttable_begin(&i, &c->groups); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + const mgroup *group = upb_value_getconstptr(upb_inttable_iter_value(&i)); + mgroup_unref(group, c); + } + upb_inttable_uninit(&c->groups); +} + +bool upb_pbcodecache_allowjit(const upb_pbcodecache *c) { + return c->allow_jit_; +} + +bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow) { + if (upb_inttable_count(&c->groups) > 0) + return false; + c->allow_jit_ = allow; + return true; +} + +const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod( + upb_pbcodecache *c, const upb_pbdecodermethodopts *opts) { + upb_value v; + bool ok; + + /* Right now we build a new DecoderMethod every time. + * TODO(haberman): properly cache methods by their true key. */ + const mgroup *g = mgroup_new(opts->handlers, c->allow_jit_, opts->lazy, c); + upb_inttable_push(&c->groups, upb_value_constptr(g)); + + ok = upb_inttable_lookupptr(&g->methods, opts->handlers, &v); + UPB_ASSERT_VAR(ok, ok); + return upb_value_getptr(v); +} + + +/* upb_pbdecodermethodopts ****************************************************/ + +void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts, + const upb_handlers *h) { + opts->handlers = h; + opts->lazy = false; +} + +void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy) { + opts->lazy = lazy; +} +/* +** upb::Decoder (Bytecode Decoder VM) +** +** Bytecode must previously have been generated using the bytecode compiler in +** compile_decoder.c. This decoder then walks through the bytecode op-by-op to +** parse the input. +** +** Decoding is fully resumable; we just keep a pointer to the current bytecode +** instruction and resume from there. A fair amount of the logic here is to +** handle the fact that values can span buffer seams and we have to be able to +** be capable of suspending/resuming from any byte in the stream. This +** sometimes requires keeping a few trailing bytes from the last buffer around +** in the "residual" buffer. +*/ + +#include <inttypes.h> +#include <stddef.h> + +#ifdef UPB_DUMP_BYTECODE +#include <stdio.h> +#endif + +#define CHECK_SUSPEND(x) if (!(x)) return upb_pbdecoder_suspend(d); + +/* Error messages that are shared between the bytecode and JIT decoders. */ +const char *kPbDecoderStackOverflow = "Nesting too deep."; +const char *kPbDecoderSubmessageTooLong = + "Submessage end extends past enclosing submessage."; + +/* Error messages shared within this file. */ +static const char *kUnterminatedVarint = "Unterminated varint."; + +/* upb_pbdecoder **************************************************************/ + +static opcode halt = OP_HALT; + +/* A dummy character we can point to when the user passes us a NULL buffer. + * We need this because in C (NULL + 0) and (NULL - NULL) are undefined + * behavior, which would invalidate functions like curbufleft(). */ +static const char dummy_char; + +/* Whether an op consumes any of the input buffer. */ +static bool consumes_input(opcode op) { + switch (op) { + case OP_SETDISPATCH: + case OP_STARTMSG: + case OP_ENDMSG: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_SETBIGGROUPNUM: + case OP_CHECKDELIM: + case OP_CALL: + case OP_RET: + case OP_BRANCH: + return false; + default: + return true; + } +} + +static size_t stacksize(upb_pbdecoder *d, size_t entries) { + UPB_UNUSED(d); + return entries * sizeof(upb_pbdecoder_frame); +} + +static size_t callstacksize(upb_pbdecoder *d, size_t entries) { + UPB_UNUSED(d); + +#ifdef UPB_USE_JIT_X64 + if (d->method_->is_native_) { + /* Each native stack frame needs two pointers, plus we need a few frames for + * the enter/exit trampolines. */ + size_t ret = entries * sizeof(void*) * 2; + ret += sizeof(void*) * 10; + return ret; + } +#endif + + return entries * sizeof(uint32_t*); +} + + +static bool in_residual_buf(const upb_pbdecoder *d, const char *p); + +/* It's unfortunate that we have to micro-manage the compiler with + * UPB_FORCEINLINE and UPB_NOINLINE, especially since this tuning is necessarily + * specific to one hardware configuration. But empirically on a Core i7, + * performance increases 30-50% with these annotations. Every instance where + * these appear, gcc 4.2.1 made the wrong decision and degraded performance in + * benchmarks. */ + +static void seterr(upb_pbdecoder *d, const char *msg) { + upb_status status = UPB_STATUS_INIT; + upb_status_seterrmsg(&status, msg); + upb_env_reporterror(d->env, &status); +} + +void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg) { + seterr(d, msg); +} + + +/* Buffering ******************************************************************/ + +/* We operate on one buffer at a time, which is either the user's buffer passed + * to our "decode" callback or some residual bytes from the previous buffer. */ + +/* How many bytes can be safely read from d->ptr without reading past end-of-buf + * or past the current delimited end. */ +static size_t curbufleft(const upb_pbdecoder *d) { + assert(d->data_end >= d->ptr); + return d->data_end - d->ptr; +} + +/* How many bytes are available before end-of-buffer. */ +static size_t bufleft(const upb_pbdecoder *d) { + return d->end - d->ptr; +} + +/* Overall stream offset of d->ptr. */ +uint64_t offset(const upb_pbdecoder *d) { + return d->bufstart_ofs + (d->ptr - d->buf); +} + +/* How many bytes are available before the end of this delimited region. */ +size_t delim_remaining(const upb_pbdecoder *d) { + return d->top->end_ofs - offset(d); +} + +/* Advances d->ptr. */ +static void advance(upb_pbdecoder *d, size_t len) { + assert(curbufleft(d) >= len); + d->ptr += len; +} + +static bool in_buf(const char *p, const char *buf, const char *end) { + return p >= buf && p <= end; +} + +static bool in_residual_buf(const upb_pbdecoder *d, const char *p) { + return in_buf(p, d->residual, d->residual_end); +} + +/* Calculates the delim_end value, which is affected by both the current buffer + * and the parsing stack, so must be called whenever either is updated. */ +static void set_delim_end(upb_pbdecoder *d) { + size_t delim_ofs = d->top->end_ofs - d->bufstart_ofs; + if (delim_ofs <= (size_t)(d->end - d->buf)) { + d->delim_end = d->buf + delim_ofs; + d->data_end = d->delim_end; + } else { + d->data_end = d->end; + d->delim_end = NULL; + } +} + +static void switchtobuf(upb_pbdecoder *d, const char *buf, const char *end) { + d->ptr = buf; + d->buf = buf; + d->end = end; + set_delim_end(d); +} + +static void advancetobuf(upb_pbdecoder *d, const char *buf, size_t len) { + assert(curbufleft(d) == 0); + d->bufstart_ofs += (d->end - d->buf); + switchtobuf(d, buf, buf + len); +} + +static void checkpoint(upb_pbdecoder *d) { + /* The assertion here is in the interests of efficiency, not correctness. + * We are trying to ensure that we don't checkpoint() more often than + * necessary. */ + assert(d->checkpoint != d->ptr); + d->checkpoint = d->ptr; +} + +/* Skips "bytes" bytes in the stream, which may be more than available. If we + * skip more bytes than are available, we return a long read count to the caller + * indicating how many bytes can be skipped over before passing actual data + * again. Skipped bytes can pass a NULL buffer and the decoder guarantees they + * won't actually be read. + */ +static int32_t skip(upb_pbdecoder *d, size_t bytes) { + assert(!in_residual_buf(d, d->ptr) || d->size_param == 0); + assert(d->skip == 0); + if (bytes > delim_remaining(d)) { + seterr(d, "Skipped value extended beyond enclosing submessage."); + return upb_pbdecoder_suspend(d); + } else if (bufleft(d) >= bytes) { + /* Skipped data is all in current buffer, and more is still available. */ + advance(d, bytes); + d->skip = 0; + return DECODE_OK; + } else { + /* Skipped data extends beyond currently available buffers. */ + d->pc = d->last; + d->skip = bytes - curbufleft(d); + d->bufstart_ofs += (d->end - d->buf); + d->residual_end = d->residual; + switchtobuf(d, d->residual, d->residual_end); + return d->size_param + d->skip; + } +} + + +/* Resumes the decoder from an initial state or from a previous suspend. */ +int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf, + size_t size, const upb_bufhandle *handle) { + UPB_UNUSED(p); /* Useless; just for the benefit of the JIT. */ + + /* d->skip and d->residual_end could probably elegantly be represented + * as a single variable, to more easily represent this invariant. */ + assert(!(d->skip && d->residual_end > d->residual)); + + /* We need to remember the original size_param, so that the value we return + * is relative to it, even if we do some skipping first. */ + d->size_param = size; + d->handle = handle; + + /* Have to handle this case specially (ie. not with skip()) because the user + * is allowed to pass a NULL buffer here, which won't allow us to safely + * calculate a d->end or use our normal functions like curbufleft(). */ + if (d->skip && d->skip >= size) { + d->skip -= size; + d->bufstart_ofs += size; + buf = &dummy_char; + size = 0; + + /* We can't just return now, because we might need to execute some ops + * like CHECKDELIM, which could call some callbacks and pop the stack. */ + } + + /* We need to pretend that this was the actual buffer param, since some of the + * calculations assume that d->ptr/d->buf is relative to this. */ + d->buf_param = buf; + + if (!buf) { + /* NULL buf is ok if its entire span is covered by the "skip" above, but + * by this point we know that "skip" doesn't cover the buffer. */ + seterr(d, "Passed NULL buffer over non-skippable region."); + return upb_pbdecoder_suspend(d); + } + + if (d->residual_end > d->residual) { + /* We have residual bytes from the last buffer. */ + assert(d->ptr == d->residual); + } else { + switchtobuf(d, buf, buf + size); + } + + d->checkpoint = d->ptr; + + /* Handle skips that don't cover the whole buffer (as above). */ + if (d->skip) { + size_t skip_bytes = d->skip; + d->skip = 0; + CHECK_RETURN(skip(d, skip_bytes)); + checkpoint(d); + } + + /* If we're inside an unknown group, continue to parse unknown values. */ + if (d->top->groupnum < 0) { + CHECK_RETURN(upb_pbdecoder_skipunknown(d, -1, 0)); + checkpoint(d); + } + + return DECODE_OK; +} + +/* Suspends the decoder at the last checkpoint, without saving any residual + * bytes. If there are any unconsumed bytes, returns a short byte count. */ +size_t upb_pbdecoder_suspend(upb_pbdecoder *d) { + d->pc = d->last; + if (d->checkpoint == d->residual) { + /* Checkpoint was in residual buf; no user bytes were consumed. */ + d->ptr = d->residual; + return 0; + } else { + size_t ret = d->size_param - (d->end - d->checkpoint); + assert(!in_residual_buf(d, d->checkpoint)); + assert(d->buf == d->buf_param || d->buf == &dummy_char); + + d->bufstart_ofs += (d->checkpoint - d->buf); + d->residual_end = d->residual; + switchtobuf(d, d->residual, d->residual_end); + return ret; + } +} + +/* Suspends the decoder at the last checkpoint, and saves any unconsumed + * bytes in our residual buffer. This is necessary if we need more user + * bytes to form a complete value, which might not be contiguous in the + * user's buffers. Always consumes all user bytes. */ +static size_t suspend_save(upb_pbdecoder *d) { + /* We hit end-of-buffer before we could parse a full value. + * Save any unconsumed bytes (if any) to the residual buffer. */ + d->pc = d->last; + + if (d->checkpoint == d->residual) { + /* Checkpoint was in residual buf; append user byte(s) to residual buf. */ + assert((d->residual_end - d->residual) + d->size_param <= + sizeof(d->residual)); + if (!in_residual_buf(d, d->ptr)) { + d->bufstart_ofs -= (d->residual_end - d->residual); + } + memcpy(d->residual_end, d->buf_param, d->size_param); + d->residual_end += d->size_param; + } else { + /* Checkpoint was in user buf; old residual bytes not needed. */ + size_t save; + assert(!in_residual_buf(d, d->checkpoint)); + + d->ptr = d->checkpoint; + save = curbufleft(d); + assert(save <= sizeof(d->residual)); + memcpy(d->residual, d->ptr, save); + d->residual_end = d->residual + save; + d->bufstart_ofs = offset(d); + } + + switchtobuf(d, d->residual, d->residual_end); + return d->size_param; +} + +/* Copies the next "bytes" bytes into "buf" and advances the stream. + * Requires that this many bytes are available in the current buffer. */ +UPB_FORCEINLINE static void consumebytes(upb_pbdecoder *d, void *buf, + size_t bytes) { + assert(bytes <= curbufleft(d)); + memcpy(buf, d->ptr, bytes); + advance(d, bytes); +} + +/* Slow path for getting the next "bytes" bytes, regardless of whether they are + * available in the current buffer or not. Returns a status code as described + * in decoder.int.h. */ +UPB_NOINLINE static int32_t getbytes_slow(upb_pbdecoder *d, void *buf, + size_t bytes) { + const size_t avail = curbufleft(d); + consumebytes(d, buf, avail); + bytes -= avail; + assert(bytes > 0); + if (in_residual_buf(d, d->ptr)) { + advancetobuf(d, d->buf_param, d->size_param); + } + if (curbufleft(d) >= bytes) { + consumebytes(d, (char *)buf + avail, bytes); + return DECODE_OK; + } else if (d->data_end == d->delim_end) { + seterr(d, "Submessage ended in the middle of a value or group"); + return upb_pbdecoder_suspend(d); + } else { + return suspend_save(d); + } +} + +/* Gets the next "bytes" bytes, regardless of whether they are available in the + * current buffer or not. Returns a status code as described in decoder.int.h. + */ +UPB_FORCEINLINE static int32_t getbytes(upb_pbdecoder *d, void *buf, + size_t bytes) { + if (curbufleft(d) >= bytes) { + /* Buffer has enough data to satisfy. */ + consumebytes(d, buf, bytes); + return DECODE_OK; + } else { + return getbytes_slow(d, buf, bytes); + } +} + +UPB_NOINLINE static size_t peekbytes_slow(upb_pbdecoder *d, void *buf, + size_t bytes) { + size_t ret = curbufleft(d); + memcpy(buf, d->ptr, ret); + if (in_residual_buf(d, d->ptr)) { + size_t copy = UPB_MIN(bytes - ret, d->size_param); + memcpy((char *)buf + ret, d->buf_param, copy); + ret += copy; + } + return ret; +} + +UPB_FORCEINLINE static size_t peekbytes(upb_pbdecoder *d, void *buf, + size_t bytes) { + if (curbufleft(d) >= bytes) { + memcpy(buf, d->ptr, bytes); + return bytes; + } else { + return peekbytes_slow(d, buf, bytes); + } +} + + +/* Decoding of wire types *****************************************************/ + +/* Slow path for decoding a varint from the current buffer position. + * Returns a status code as described in decoder.int.h. */ +UPB_NOINLINE int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d, + uint64_t *u64) { + uint8_t byte = 0x80; + int bitpos; + *u64 = 0; + for(bitpos = 0; bitpos < 70 && (byte & 0x80); bitpos += 7) { + CHECK_RETURN(getbytes(d, &byte, 1)); + *u64 |= (uint64_t)(byte & 0x7F) << bitpos; + } + if(bitpos == 70 && (byte & 0x80)) { + seterr(d, kUnterminatedVarint); + return upb_pbdecoder_suspend(d); + } + return DECODE_OK; +} + +/* Decodes a varint from the current buffer position. + * Returns a status code as described in decoder.int.h. */ +UPB_FORCEINLINE static int32_t decode_varint(upb_pbdecoder *d, uint64_t *u64) { + if (curbufleft(d) > 0 && !(*d->ptr & 0x80)) { + *u64 = *d->ptr; + advance(d, 1); + return DECODE_OK; + } else if (curbufleft(d) >= 10) { + /* Fast case. */ + upb_decoderet r = upb_vdecode_fast(d->ptr); + if (r.p == NULL) { + seterr(d, kUnterminatedVarint); + return upb_pbdecoder_suspend(d); + } + advance(d, r.p - d->ptr); + *u64 = r.val; + return DECODE_OK; + } else { + /* Slow case -- varint spans buffer seam. */ + return upb_pbdecoder_decode_varint_slow(d, u64); + } +} + +/* Decodes a 32-bit varint from the current buffer position. + * Returns a status code as described in decoder.int.h. */ +UPB_FORCEINLINE static int32_t decode_v32(upb_pbdecoder *d, uint32_t *u32) { + uint64_t u64; + int32_t ret = decode_varint(d, &u64); + if (ret >= 0) return ret; + if (u64 > UINT32_MAX) { + seterr(d, "Unterminated 32-bit varint"); + /* TODO(haberman) guarantee that this function return is >= 0 somehow, + * so we know this path will always be treated as error by our caller. + * Right now the size_t -> int32_t can overflow and produce negative values. + */ + *u32 = 0; + return upb_pbdecoder_suspend(d); + } + *u32 = u64; + return DECODE_OK; +} + +/* Decodes a fixed32 from the current buffer position. + * Returns a status code as described in decoder.int.h. + * TODO: proper byte swapping for big-endian machines. */ +UPB_FORCEINLINE static int32_t decode_fixed32(upb_pbdecoder *d, uint32_t *u32) { + return getbytes(d, u32, 4); +} + +/* Decodes a fixed64 from the current buffer position. + * Returns a status code as described in decoder.int.h. + * TODO: proper byte swapping for big-endian machines. */ +UPB_FORCEINLINE static int32_t decode_fixed64(upb_pbdecoder *d, uint64_t *u64) { + return getbytes(d, u64, 8); +} + +/* Non-static versions of the above functions. + * These are called by the JIT for fallback paths. */ +int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32) { + return decode_fixed32(d, u32); +} + +int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64) { + return decode_fixed64(d, u64); +} + +static double as_double(uint64_t n) { double d; memcpy(&d, &n, 8); return d; } +static float as_float(uint32_t n) { float f; memcpy(&f, &n, 4); return f; } + +/* Pushes a frame onto the decoder stack. */ +static bool decoder_push(upb_pbdecoder *d, uint64_t end) { + upb_pbdecoder_frame *fr = d->top; + + if (end > fr->end_ofs) { + seterr(d, kPbDecoderSubmessageTooLong); + return false; + } else if (fr == d->limit) { + seterr(d, kPbDecoderStackOverflow); + return false; + } + + fr++; + fr->end_ofs = end; + fr->dispatch = NULL; + fr->groupnum = 0; + d->top = fr; + return true; +} + +static bool pushtagdelim(upb_pbdecoder *d, uint32_t arg) { + /* While we expect to see an "end" tag (either ENDGROUP or a non-sequence + * field number) prior to hitting any enclosing submessage end, pushing our + * existing delim end prevents us from continuing to parse values from a + * corrupt proto that doesn't give us an END tag in time. */ + if (!decoder_push(d, d->top->end_ofs)) + return false; + d->top->groupnum = arg; + return true; +} + +/* Pops a frame from the decoder stack. */ +static void decoder_pop(upb_pbdecoder *d) { d->top--; } + +UPB_NOINLINE int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d, + uint64_t expected) { + uint64_t data = 0; + size_t bytes = upb_value_size(expected); + size_t read = peekbytes(d, &data, bytes); + if (read == bytes && data == expected) { + /* Advance past matched bytes. */ + int32_t ok = getbytes(d, &data, read); + UPB_ASSERT_VAR(ok, ok < 0); + return DECODE_OK; + } else if (read < bytes && memcmp(&data, &expected, read) == 0) { + return suspend_save(d); + } else { + return DECODE_MISMATCH; + } +} + +int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum, + uint8_t wire_type) { + if (fieldnum >= 0) + goto have_tag; + + while (true) { + uint32_t tag; + CHECK_RETURN(decode_v32(d, &tag)); + wire_type = tag & 0x7; + fieldnum = tag >> 3; + +have_tag: + if (fieldnum == 0) { + seterr(d, "Saw invalid field number (0)"); + return upb_pbdecoder_suspend(d); + } + + /* TODO: deliver to unknown field callback. */ + switch (wire_type) { + case UPB_WIRE_TYPE_32BIT: + CHECK_RETURN(skip(d, 4)); + break; + case UPB_WIRE_TYPE_64BIT: + CHECK_RETURN(skip(d, 8)); + break; + case UPB_WIRE_TYPE_VARINT: { + uint64_t u64; + CHECK_RETURN(decode_varint(d, &u64)); + break; + } + case UPB_WIRE_TYPE_DELIMITED: { + uint32_t len; + CHECK_RETURN(decode_v32(d, &len)); + CHECK_RETURN(skip(d, len)); + break; + } + case UPB_WIRE_TYPE_START_GROUP: + CHECK_SUSPEND(pushtagdelim(d, -fieldnum)); + break; + case UPB_WIRE_TYPE_END_GROUP: + if (fieldnum == -d->top->groupnum) { + decoder_pop(d); + } else if (fieldnum == d->top->groupnum) { + return DECODE_ENDGROUP; + } else { + seterr(d, "Unmatched ENDGROUP tag."); + return upb_pbdecoder_suspend(d); + } + break; + default: + seterr(d, "Invalid wire type"); + return upb_pbdecoder_suspend(d); + } + + if (d->top->groupnum >= 0) { + return DECODE_OK; + } + + /* Unknown group -- continue looping over unknown fields. */ + checkpoint(d); + } +} + +static void goto_endmsg(upb_pbdecoder *d) { + upb_value v; + bool found = upb_inttable_lookup32(d->top->dispatch, DISPATCH_ENDMSG, &v); + UPB_ASSERT_VAR(found, found); + d->pc = d->top->base + upb_value_getuint64(v); +} + +/* Parses a tag and jumps to the corresponding bytecode instruction for this + * field. + * + * If the tag is unknown (or the wire type doesn't match), parses the field as + * unknown. If the tag is a valid ENDGROUP tag, jumps to the bytecode + * instruction for the end of message. */ +static int32_t dispatch(upb_pbdecoder *d) { + upb_inttable *dispatch = d->top->dispatch; + uint32_t tag; + uint8_t wire_type; + uint32_t fieldnum; + upb_value val; + int32_t retval; + + /* Decode tag. */ + CHECK_RETURN(decode_v32(d, &tag)); + wire_type = tag & 0x7; + fieldnum = tag >> 3; + + /* Lookup tag. Because of packed/non-packed compatibility, we have to + * check the wire type against two possibilities. */ + if (fieldnum != DISPATCH_ENDMSG && + upb_inttable_lookup32(dispatch, fieldnum, &val)) { + uint64_t v = upb_value_getuint64(val); + if (wire_type == (v & 0xff)) { + d->pc = d->top->base + (v >> 16); + return DECODE_OK; + } else if (wire_type == ((v >> 8) & 0xff)) { + bool found = + upb_inttable_lookup(dispatch, fieldnum + UPB_MAX_FIELDNUMBER, &val); + UPB_ASSERT_VAR(found, found); + d->pc = d->top->base + upb_value_getuint64(val); + return DECODE_OK; + } + } + + /* We have some unknown fields (or ENDGROUP) to parse. The DISPATCH or TAG + * bytecode that triggered this is preceded by a CHECKDELIM bytecode which + * we need to back up to, so that when we're done skipping unknown data we + * can re-check the delimited end. */ + d->last--; /* Necessary if we get suspended */ + d->pc = d->last; + assert(getop(*d->last) == OP_CHECKDELIM); + + /* Unknown field or ENDGROUP. */ + retval = upb_pbdecoder_skipunknown(d, fieldnum, wire_type); + + CHECK_RETURN(retval); + + if (retval == DECODE_ENDGROUP) { + goto_endmsg(d); + return DECODE_OK; + } + + return DECODE_OK; +} + +/* Callers know that the stack is more than one deep because the opcodes that + * call this only occur after PUSH operations. */ +upb_pbdecoder_frame *outer_frame(upb_pbdecoder *d) { + assert(d->top != d->stack); + return d->top - 1; +} + + +/* The main decoding loop *****************************************************/ + +/* The main decoder VM function. Uses traditional bytecode dispatch loop with a + * switch() statement. */ +size_t run_decoder_vm(upb_pbdecoder *d, const mgroup *group, + const upb_bufhandle* handle) { + +#define VMCASE(op, code) \ + case op: { code; if (consumes_input(op)) checkpoint(d); break; } +#define PRIMITIVE_OP(type, wt, name, convfunc, ctype) \ + VMCASE(OP_PARSE_ ## type, { \ + ctype val; \ + CHECK_RETURN(decode_ ## wt(d, &val)); \ + upb_sink_put ## name(&d->top->sink, arg, (convfunc)(val)); \ + }) + + while(1) { + int32_t instruction; + opcode op; + uint32_t arg; + int32_t longofs; + + d->last = d->pc; + instruction = *d->pc++; + op = getop(instruction); + arg = instruction >> 8; + longofs = arg; + assert(d->ptr != d->residual_end); + UPB_UNUSED(group); +#ifdef UPB_DUMP_BYTECODE + fprintf(stderr, "s_ofs=%d buf_ofs=%d data_rem=%d buf_rem=%d delim_rem=%d " + "%x %s (%d)\n", + (int)offset(d), + (int)(d->ptr - d->buf), + (int)(d->data_end - d->ptr), + (int)(d->end - d->ptr), + (int)((d->top->end_ofs - d->bufstart_ofs) - (d->ptr - d->buf)), + (int)(d->pc - 1 - group->bytecode), + upb_pbdecoder_getopname(op), + arg); +#endif + switch (op) { + /* Technically, we are losing data if we see a 32-bit varint that is not + * properly sign-extended. We could detect this and error about the data + * loss, but proto2 does not do this, so we pass. */ + PRIMITIVE_OP(INT32, varint, int32, int32_t, uint64_t) + PRIMITIVE_OP(INT64, varint, int64, int64_t, uint64_t) + PRIMITIVE_OP(UINT32, varint, uint32, uint32_t, uint64_t) + PRIMITIVE_OP(UINT64, varint, uint64, uint64_t, uint64_t) + PRIMITIVE_OP(FIXED32, fixed32, uint32, uint32_t, uint32_t) + PRIMITIVE_OP(FIXED64, fixed64, uint64, uint64_t, uint64_t) + PRIMITIVE_OP(SFIXED32, fixed32, int32, int32_t, uint32_t) + PRIMITIVE_OP(SFIXED64, fixed64, int64, int64_t, uint64_t) + PRIMITIVE_OP(BOOL, varint, bool, bool, uint64_t) + PRIMITIVE_OP(DOUBLE, fixed64, double, as_double, uint64_t) + PRIMITIVE_OP(FLOAT, fixed32, float, as_float, uint32_t) + PRIMITIVE_OP(SINT32, varint, int32, upb_zzdec_32, uint64_t) + PRIMITIVE_OP(SINT64, varint, int64, upb_zzdec_64, uint64_t) + + VMCASE(OP_SETDISPATCH, + d->top->base = d->pc - 1; + memcpy(&d->top->dispatch, d->pc, sizeof(void*)); + d->pc += sizeof(void*) / sizeof(uint32_t); + ) + VMCASE(OP_STARTMSG, + CHECK_SUSPEND(upb_sink_startmsg(&d->top->sink)); + ) + VMCASE(OP_ENDMSG, + CHECK_SUSPEND(upb_sink_endmsg(&d->top->sink, d->status)); + ) + VMCASE(OP_STARTSEQ, + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startseq(&outer->sink, arg, &d->top->sink)); + ) + VMCASE(OP_ENDSEQ, + CHECK_SUSPEND(upb_sink_endseq(&d->top->sink, arg)); + ) + VMCASE(OP_STARTSUBMSG, + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startsubmsg(&outer->sink, arg, &d->top->sink)); + ) + VMCASE(OP_ENDSUBMSG, + CHECK_SUSPEND(upb_sink_endsubmsg(&d->top->sink, arg)); + ) + VMCASE(OP_STARTSTR, + uint32_t len = delim_remaining(d); + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startstr(&outer->sink, arg, len, &d->top->sink)); + if (len == 0) { + d->pc++; /* Skip OP_STRING. */ + } + ) + VMCASE(OP_STRING, + uint32_t len = curbufleft(d); + size_t n = upb_sink_putstring(&d->top->sink, arg, d->ptr, len, handle); + if (n > len) { + if (n > delim_remaining(d)) { + seterr(d, "Tried to skip past end of string."); + return upb_pbdecoder_suspend(d); + } else { + int32_t ret = skip(d, n); + /* This shouldn't return DECODE_OK, because n > len. */ + assert(ret >= 0); + return ret; + } + } + advance(d, n); + if (n < len || d->delim_end == NULL) { + /* We aren't finished with this string yet. */ + d->pc--; /* Repeat OP_STRING. */ + if (n > 0) checkpoint(d); + return upb_pbdecoder_suspend(d); + } + ) + VMCASE(OP_ENDSTR, + CHECK_SUSPEND(upb_sink_endstr(&d->top->sink, arg)); + ) + VMCASE(OP_PUSHTAGDELIM, + CHECK_SUSPEND(pushtagdelim(d, arg)); + ) + VMCASE(OP_SETBIGGROUPNUM, + d->top->groupnum = *d->pc++; + ) + VMCASE(OP_POP, + assert(d->top > d->stack); + decoder_pop(d); + ) + VMCASE(OP_PUSHLENDELIM, + uint32_t len; + CHECK_RETURN(decode_v32(d, &len)); + CHECK_SUSPEND(decoder_push(d, offset(d) + len)); + set_delim_end(d); + ) + VMCASE(OP_SETDELIM, + set_delim_end(d); + ) + VMCASE(OP_CHECKDELIM, + /* We are guaranteed of this assert because we never allow ourselves to + * consume bytes beyond data_end, which covers delim_end when non-NULL. + */ + assert(!(d->delim_end && d->ptr > d->delim_end)); + if (d->ptr == d->delim_end) + d->pc += longofs; + ) + VMCASE(OP_CALL, + d->callstack[d->call_len++] = d->pc; + d->pc += longofs; + ) + VMCASE(OP_RET, + assert(d->call_len > 0); + d->pc = d->callstack[--d->call_len]; + ) + VMCASE(OP_BRANCH, + d->pc += longofs; + ) + VMCASE(OP_TAG1, + uint8_t expected; + CHECK_SUSPEND(curbufleft(d) > 0); + expected = (arg >> 8) & 0xff; + if (*d->ptr == expected) { + advance(d, 1); + } else { + int8_t shortofs; + badtag: + shortofs = arg; + if (shortofs == LABEL_DISPATCH) { + CHECK_RETURN(dispatch(d)); + } else { + d->pc += shortofs; + break; /* Avoid checkpoint(). */ + } + } + ) + VMCASE(OP_TAG2, + uint16_t expected; + CHECK_SUSPEND(curbufleft(d) > 0); + expected = (arg >> 8) & 0xffff; + if (curbufleft(d) >= 2) { + uint16_t actual; + memcpy(&actual, d->ptr, 2); + if (expected == actual) { + advance(d, 2); + } else { + goto badtag; + } + } else { + int32_t result = upb_pbdecoder_checktag_slow(d, expected); + if (result == DECODE_MISMATCH) goto badtag; + if (result >= 0) return result; + } + ) + VMCASE(OP_TAGN, { + uint64_t expected; + int32_t result; + memcpy(&expected, d->pc, 8); + d->pc += 2; + result = upb_pbdecoder_checktag_slow(d, expected); + if (result == DECODE_MISMATCH) goto badtag; + if (result >= 0) return result; + }) + VMCASE(OP_DISPATCH, { + CHECK_RETURN(dispatch(d)); + }) + VMCASE(OP_HALT, { + return d->size_param; + }) + } + } +} + + +/* BytesHandler handlers ******************************************************/ + +void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint) { + upb_pbdecoder *d = closure; + UPB_UNUSED(size_hint); + d->top->end_ofs = UINT64_MAX; + d->bufstart_ofs = 0; + d->call_len = 1; + d->callstack[0] = &halt; + d->pc = pc; + d->skip = 0; + return d; +} + +void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint) { + upb_pbdecoder *d = closure; + UPB_UNUSED(hd); + UPB_UNUSED(size_hint); + d->top->end_ofs = UINT64_MAX; + d->bufstart_ofs = 0; + d->call_len = 0; + d->skip = 0; + return d; +} + +bool upb_pbdecoder_end(void *closure, const void *handler_data) { + upb_pbdecoder *d = closure; + const upb_pbdecodermethod *method = handler_data; + uint64_t end; + char dummy; + + if (d->residual_end > d->residual) { + seterr(d, "Unexpected EOF: decoder still has buffered unparsed data"); + return false; + } + + if (d->skip) { + seterr(d, "Unexpected EOF inside skipped data"); + return false; + } + + if (d->top->end_ofs != UINT64_MAX) { + seterr(d, "Unexpected EOF inside delimited string"); + return false; + } + + /* The user's end() call indicates that the message ends here. */ + end = offset(d); + d->top->end_ofs = end; + +#ifdef UPB_USE_JIT_X64 + if (method->is_native_) { + const mgroup *group = (const mgroup*)method->group; + if (d->top != d->stack) + d->stack->end_ofs = 0; + group->jit_code(closure, method->code_base.ptr, &dummy, 0, NULL); + } else +#endif + { + const uint32_t *p = d->pc; + d->stack->end_ofs = end; + /* Check the previous bytecode, but guard against beginning. */ + if (p != method->code_base.ptr) p--; + if (getop(*p) == OP_CHECKDELIM) { + /* Rewind from OP_TAG* to OP_CHECKDELIM. */ + assert(getop(*d->pc) == OP_TAG1 || + getop(*d->pc) == OP_TAG2 || + getop(*d->pc) == OP_TAGN || + getop(*d->pc) == OP_DISPATCH); + d->pc = p; + } + upb_pbdecoder_decode(closure, handler_data, &dummy, 0, NULL); + } + + if (d->call_len != 0) { + seterr(d, "Unexpected EOF inside submessage or group"); + return false; + } + + return true; +} + +size_t upb_pbdecoder_decode(void *decoder, const void *group, const char *buf, + size_t size, const upb_bufhandle *handle) { + int32_t result = upb_pbdecoder_resume(decoder, NULL, buf, size, handle); + + if (result == DECODE_ENDGROUP) goto_endmsg(decoder); + CHECK_RETURN(result); + + return run_decoder_vm(decoder, group, handle); +} + + +/* Public API *****************************************************************/ + +void upb_pbdecoder_reset(upb_pbdecoder *d) { + d->top = d->stack; + d->top->groupnum = 0; + d->ptr = d->residual; + d->buf = d->residual; + d->end = d->residual; + d->residual_end = d->residual; +} + +upb_pbdecoder *upb_pbdecoder_create(upb_env *e, const upb_pbdecodermethod *m, + upb_sink *sink) { + const size_t default_max_nesting = 64; +#ifndef NDEBUG + size_t size_before = upb_env_bytesallocated(e); +#endif + + upb_pbdecoder *d = upb_env_malloc(e, sizeof(upb_pbdecoder)); + if (!d) return NULL; + + d->method_ = m; + d->callstack = upb_env_malloc(e, callstacksize(d, default_max_nesting)); + d->stack = upb_env_malloc(e, stacksize(d, default_max_nesting)); + if (!d->stack || !d->callstack) { + return NULL; + } + + d->env = e; + d->limit = d->stack + default_max_nesting - 1; + d->stack_size = default_max_nesting; + d->status = NULL; + + upb_pbdecoder_reset(d); + upb_bytessink_reset(&d->input_, &m->input_handler_, d); + + assert(sink); + if (d->method_->dest_handlers_) { + if (sink->handlers != d->method_->dest_handlers_) + return NULL; + } + upb_sink_reset(&d->top->sink, sink->handlers, sink->closure); + + /* If this fails, increase the value in decoder.h. */ + assert(upb_env_bytesallocated(e) - size_before <= UPB_PB_DECODER_SIZE); + return d; +} + +uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d) { + return offset(d); +} + +const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d) { + return d->method_; +} + +upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d) { + return &d->input_; +} + +size_t upb_pbdecoder_maxnesting(const upb_pbdecoder *d) { + return d->stack_size; +} + +bool upb_pbdecoder_setmaxnesting(upb_pbdecoder *d, size_t max) { + assert(d->top >= d->stack); + + if (max < (size_t)(d->top - d->stack)) { + /* Can't set a limit smaller than what we are currently at. */ + return false; + } + + if (max > d->stack_size) { + /* Need to reallocate stack and callstack to accommodate. */ + size_t old_size = stacksize(d, d->stack_size); + size_t new_size = stacksize(d, max); + void *p = upb_env_realloc(d->env, d->stack, old_size, new_size); + if (!p) { + return false; + } + d->stack = p; + + old_size = callstacksize(d, d->stack_size); + new_size = callstacksize(d, max); + p = upb_env_realloc(d->env, d->callstack, old_size, new_size); + if (!p) { + return false; + } + d->callstack = p; + + d->stack_size = max; + } + + d->limit = d->stack + max - 1; + return true; +} +/* +** upb::Encoder +** +** Since we are implementing pure handlers (ie. without any out-of-band access +** to pre-computed lengths), we have to buffer all submessages before we can +** emit even their first byte. +** +** Not knowing the size of submessages also means we can't write a perfect +** zero-copy implementation, even with buffering. Lengths are stored as +** varints, which means that we don't know how many bytes to reserve for the +** length until we know what the length is. +** +** This leaves us with three main choices: +** +** 1. buffer all submessage data in a temporary buffer, then copy it exactly +** once into the output buffer. +** +** 2. attempt to buffer data directly into the output buffer, estimating how +** many bytes each length will take. When our guesses are wrong, use +** memmove() to grow or shrink the allotted space. +** +** 3. buffer directly into the output buffer, allocating a max length +** ahead-of-time for each submessage length. If we overallocated, we waste +** space, but no memcpy() or memmove() is required. This approach requires +** defining a maximum size for submessages and rejecting submessages that +** exceed that size. +** +** (2) and (3) have the potential to have better performance, but they are more +** complicated and subtle to implement: +** +** (3) requires making an arbitrary choice of the maximum message size; it +** wastes space when submessages are shorter than this and fails +** completely when they are longer. This makes it more finicky and +** requires configuration based on the input. It also makes it impossible +** to perfectly match the output of reference encoders that always use the +** optimal amount of space for each length. +** +** (2) requires guessing the the size upfront, and if multiple lengths are +** guessed wrong the minimum required number of memmove() operations may +** be complicated to compute correctly. Implemented properly, it may have +** a useful amortized or average cost, but more investigation is required +** to determine this and what the optimal algorithm is to achieve it. +** +** (1) makes you always pay for exactly one copy, but its implementation is +** the simplest and its performance is predictable. +** +** So for now, we implement (1) only. If we wish to optimize later, we should +** be able to do it without affecting users. +** +** The strategy is to buffer the segments of data that do *not* depend on +** unknown lengths in one buffer, and keep a separate buffer of segment pointers +** and lengths. When the top-level submessage ends, we can go beginning to end, +** alternating the writing of lengths with memcpy() of the rest of the data. +** At the top level though, no buffering is required. +*/ + + + +/* The output buffer is divided into segments; a segment is a string of data + * that is "ready to go" -- it does not need any varint lengths inserted into + * the middle. The seams between segments are where varints will be inserted + * once they are known. + * + * We also use the concept of a "run", which is a range of encoded bytes that + * occur at a single submessage level. Every segment contains one or more runs. + * + * A segment can span messages. Consider: + * + * .--Submessage lengths---------. + * | | | + * | V V + * V | |--------------- | |----------------- + * Submessages: | |----------------------------------------------- + * Top-level msg: ------------------------------------------------------------ + * + * Segments: ----- ------------------- ----------------- + * Runs: *---- *--------------*--- *---------------- + * (* marks the start) + * + * Note that the top-level menssage is not in any segment because it does not + * have any length preceding it. + * + * A segment is only interrupted when another length needs to be inserted. So + * observe how the second segment spans both the inner submessage and part of + * the next enclosing message. */ +typedef struct { + uint32_t msglen; /* The length to varint-encode before this segment. */ + uint32_t seglen; /* Length of the segment. */ +} upb_pb_encoder_segment; + +struct upb_pb_encoder { + upb_env *env; + + /* Our input and output. */ + upb_sink input_; + upb_bytessink *output_; + + /* The "subclosure" -- used as the inner closure as part of the bytessink + * protocol. */ + void *subc; + + /* The output buffer and limit, and our current write position. "buf" + * initially points to "initbuf", but is dynamically allocated if we need to + * grow beyond the initial size. */ + char *buf, *ptr, *limit; + + /* The beginning of the current run, or undefined if we are at the top + * level. */ + char *runbegin; + + /* The list of segments we are accumulating. */ + upb_pb_encoder_segment *segbuf, *segptr, *seglimit; + + /* The stack of enclosing submessages. Each entry in the stack points to the + * segment where this submessage's length is being accumulated. */ + int *stack, *top, *stacklimit; + + /* Depth of startmsg/endmsg calls. */ + int depth; +}; + +/* low-level buffering ********************************************************/ + +/* Low-level functions for interacting with the output buffer. */ + +/* TODO(haberman): handle pushback */ +static void putbuf(upb_pb_encoder *e, const char *buf, size_t len) { + size_t n = upb_bytessink_putbuf(e->output_, e->subc, buf, len, NULL); + UPB_ASSERT_VAR(n, n == len); +} + +static upb_pb_encoder_segment *top(upb_pb_encoder *e) { + return &e->segbuf[*e->top]; +} + +/* Call to ensure that at least "bytes" bytes are available for writing at + * e->ptr. Returns false if the bytes could not be allocated. */ +static bool reserve(upb_pb_encoder *e, size_t bytes) { + if ((size_t)(e->limit - e->ptr) < bytes) { + /* Grow buffer. */ + char *new_buf; + size_t needed = bytes + (e->ptr - e->buf); + size_t old_size = e->limit - e->buf; + + size_t new_size = old_size; + + while (new_size < needed) { + new_size *= 2; + } + + new_buf = upb_env_realloc(e->env, e->buf, old_size, new_size); + + if (new_buf == NULL) { + return false; + } + + e->ptr = new_buf + (e->ptr - e->buf); + e->runbegin = new_buf + (e->runbegin - e->buf); + e->limit = new_buf + new_size; + e->buf = new_buf; + } + + return true; +} + +/* Call when "bytes" bytes have been writte at e->ptr. The caller *must* have + * previously called reserve() with at least this many bytes. */ +static void encoder_advance(upb_pb_encoder *e, size_t bytes) { + assert((size_t)(e->limit - e->ptr) >= bytes); + e->ptr += bytes; +} + +/* Call when all of the bytes for a handler have been written. Flushes the + * bytes if possible and necessary, returning false if this failed. */ +static bool commit(upb_pb_encoder *e) { + if (!e->top) { + /* We aren't inside a delimited region. Flush our accumulated bytes to + * the output. + * + * TODO(haberman): in the future we may want to delay flushing for + * efficiency reasons. */ + putbuf(e, e->buf, e->ptr - e->buf); + e->ptr = e->buf; + } + + return true; +} + +/* Writes the given bytes to the buffer, handling reserve/advance. */ +static bool encode_bytes(upb_pb_encoder *e, const void *data, size_t len) { + if (!reserve(e, len)) { + return false; + } + + memcpy(e->ptr, data, len); + encoder_advance(e, len); + return true; +} + +/* Finish the current run by adding the run totals to the segment and message + * length. */ +static void accumulate(upb_pb_encoder *e) { + size_t run_len; + assert(e->ptr >= e->runbegin); + run_len = e->ptr - e->runbegin; + e->segptr->seglen += run_len; + top(e)->msglen += run_len; + e->runbegin = e->ptr; +} + +/* Call to indicate the start of delimited region for which the full length is + * not yet known. All data will be buffered until the length is known. + * Delimited regions may be nested; their lengths will all be tracked properly. */ +static bool start_delim(upb_pb_encoder *e) { + if (e->top) { + /* We are already buffering, advance to the next segment and push it on the + * stack. */ + accumulate(e); + + if (++e->top == e->stacklimit) { + /* TODO(haberman): grow stack? */ + return false; + } + + if (++e->segptr == e->seglimit) { + /* Grow segment buffer. */ + size_t old_size = + (e->seglimit - e->segbuf) * sizeof(upb_pb_encoder_segment); + size_t new_size = old_size * 2; + upb_pb_encoder_segment *new_buf = + upb_env_realloc(e->env, e->segbuf, old_size, new_size); + + if (new_buf == NULL) { + return false; + } + + e->segptr = new_buf + (e->segptr - e->segbuf); + e->seglimit = new_buf + (new_size / sizeof(upb_pb_encoder_segment)); + e->segbuf = new_buf; + } + } else { + /* We were previously at the top level, start buffering. */ + e->segptr = e->segbuf; + e->top = e->stack; + e->runbegin = e->ptr; + } + + *e->top = e->segptr - e->segbuf; + e->segptr->seglen = 0; + e->segptr->msglen = 0; + + return true; +} + +/* Call to indicate the end of a delimited region. We now know the length of + * the delimited region. If we are not nested inside any other delimited + * regions, we can now emit all of the buffered data we accumulated. */ +static bool end_delim(upb_pb_encoder *e) { + size_t msglen; + accumulate(e); + msglen = top(e)->msglen; + + if (e->top == e->stack) { + /* All lengths are now available, emit all buffered data. */ + char buf[UPB_PB_VARINT_MAX_LEN]; + upb_pb_encoder_segment *s; + const char *ptr = e->buf; + for (s = e->segbuf; s <= e->segptr; s++) { + size_t lenbytes = upb_vencode64(s->msglen, buf); + putbuf(e, buf, lenbytes); + putbuf(e, ptr, s->seglen); + ptr += s->seglen; + } + + e->ptr = e->buf; + e->top = NULL; + } else { + /* Need to keep buffering; propagate length info into enclosing + * submessages. */ + --e->top; + top(e)->msglen += msglen + upb_varint_size(msglen); + } + + return true; +} + + +/* tag_t **********************************************************************/ + +/* A precomputed (pre-encoded) tag and length. */ + +typedef struct { + uint8_t bytes; + char tag[7]; +} tag_t; + +/* Allocates a new tag for this field, and sets it in these handlerattr. */ +static void new_tag(upb_handlers *h, const upb_fielddef *f, upb_wiretype_t wt, + upb_handlerattr *attr) { + uint32_t n = upb_fielddef_number(f); + + tag_t *tag = upb_gmalloc(sizeof(tag_t)); + tag->bytes = upb_vencode64((n << 3) | wt, tag->tag); + + upb_handlerattr_init(attr); + upb_handlerattr_sethandlerdata(attr, tag); + upb_handlers_addcleanup(h, tag, upb_gfree); +} + +static bool encode_tag(upb_pb_encoder *e, const tag_t *tag) { + return encode_bytes(e, tag->tag, tag->bytes); +} + + +/* encoding of wire types *****************************************************/ + +static bool encode_fixed64(upb_pb_encoder *e, uint64_t val) { + /* TODO(haberman): byte-swap for big endian. */ + return encode_bytes(e, &val, sizeof(uint64_t)); +} + +static bool encode_fixed32(upb_pb_encoder *e, uint32_t val) { + /* TODO(haberman): byte-swap for big endian. */ + return encode_bytes(e, &val, sizeof(uint32_t)); +} + +static bool encode_varint(upb_pb_encoder *e, uint64_t val) { + if (!reserve(e, UPB_PB_VARINT_MAX_LEN)) { + return false; + } + + encoder_advance(e, upb_vencode64(val, e->ptr)); + return true; +} + +static uint64_t dbl2uint64(double d) { + uint64_t ret; + memcpy(&ret, &d, sizeof(uint64_t)); + return ret; +} + +static uint32_t flt2uint32(float d) { + uint32_t ret; + memcpy(&ret, &d, sizeof(uint32_t)); + return ret; +} + + +/* encoding of proto types ****************************************************/ + +static bool startmsg(void *c, const void *hd) { + upb_pb_encoder *e = c; + UPB_UNUSED(hd); + if (e->depth++ == 0) { + upb_bytessink_start(e->output_, 0, &e->subc); + } + return true; +} + +static bool endmsg(void *c, const void *hd, upb_status *status) { + upb_pb_encoder *e = c; + UPB_UNUSED(hd); + UPB_UNUSED(status); + if (--e->depth == 0) { + upb_bytessink_end(e->output_); + } + return true; +} + +static void *encode_startdelimfield(void *c, const void *hd) { + bool ok = encode_tag(c, hd) && commit(c) && start_delim(c); + return ok ? c : UPB_BREAK; +} + +static bool encode_enddelimfield(void *c, const void *hd) { + UPB_UNUSED(hd); + return end_delim(c); +} + +static void *encode_startgroup(void *c, const void *hd) { + return (encode_tag(c, hd) && commit(c)) ? c : UPB_BREAK; +} + +static bool encode_endgroup(void *c, const void *hd) { + return encode_tag(c, hd) && commit(c); +} + +static void *encode_startstr(void *c, const void *hd, size_t size_hint) { + UPB_UNUSED(size_hint); + return encode_startdelimfield(c, hd); +} + +static size_t encode_strbuf(void *c, const void *hd, const char *buf, + size_t len, const upb_bufhandle *h) { + UPB_UNUSED(hd); + UPB_UNUSED(h); + return encode_bytes(c, buf, len) ? len : 0; +} + +#define T(type, ctype, convert, encode) \ + static bool encode_scalar_##type(void *e, const void *hd, ctype val) { \ + return encode_tag(e, hd) && encode(e, (convert)(val)) && commit(e); \ + } \ + static bool encode_packed_##type(void *e, const void *hd, ctype val) { \ + UPB_UNUSED(hd); \ + return encode(e, (convert)(val)); \ + } + +T(double, double, dbl2uint64, encode_fixed64) +T(float, float, flt2uint32, encode_fixed32) +T(int64, int64_t, uint64_t, encode_varint) +T(int32, int32_t, uint32_t, encode_varint) +T(fixed64, uint64_t, uint64_t, encode_fixed64) +T(fixed32, uint32_t, uint32_t, encode_fixed32) +T(bool, bool, bool, encode_varint) +T(uint32, uint32_t, uint32_t, encode_varint) +T(uint64, uint64_t, uint64_t, encode_varint) +T(enum, int32_t, uint32_t, encode_varint) +T(sfixed32, int32_t, uint32_t, encode_fixed32) +T(sfixed64, int64_t, uint64_t, encode_fixed64) +T(sint32, int32_t, upb_zzenc_32, encode_varint) +T(sint64, int64_t, upb_zzenc_64, encode_varint) + +#undef T + + +/* code to build the handlers *************************************************/ + +static void newhandlers_callback(const void *closure, upb_handlers *h) { + const upb_msgdef *m; + upb_msg_field_iter i; + + UPB_UNUSED(closure); + + upb_handlers_setstartmsg(h, startmsg, NULL); + upb_handlers_setendmsg(h, endmsg, NULL); + + m = upb_handlers_msgdef(h); + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + bool packed = upb_fielddef_isseq(f) && upb_fielddef_isprimitive(f) && + upb_fielddef_packed(f); + upb_handlerattr attr; + upb_wiretype_t wt = + packed ? UPB_WIRE_TYPE_DELIMITED + : upb_pb_native_wire_types[upb_fielddef_descriptortype(f)]; + + /* Pre-encode the tag for this field. */ + new_tag(h, f, wt, &attr); + + if (packed) { + upb_handlers_setstartseq(h, f, encode_startdelimfield, &attr); + upb_handlers_setendseq(h, f, encode_enddelimfield, &attr); + } + +#define T(upper, lower, upbtype) \ + case UPB_DESCRIPTOR_TYPE_##upper: \ + if (packed) { \ + upb_handlers_set##upbtype(h, f, encode_packed_##lower, &attr); \ + } else { \ + upb_handlers_set##upbtype(h, f, encode_scalar_##lower, &attr); \ + } \ + break; + + switch (upb_fielddef_descriptortype(f)) { + T(DOUBLE, double, double); + T(FLOAT, float, float); + T(INT64, int64, int64); + T(INT32, int32, int32); + T(FIXED64, fixed64, uint64); + T(FIXED32, fixed32, uint32); + T(BOOL, bool, bool); + T(UINT32, uint32, uint32); + T(UINT64, uint64, uint64); + T(ENUM, enum, int32); + T(SFIXED32, sfixed32, int32); + T(SFIXED64, sfixed64, int64); + T(SINT32, sint32, int32); + T(SINT64, sint64, int64); + case UPB_DESCRIPTOR_TYPE_STRING: + case UPB_DESCRIPTOR_TYPE_BYTES: + upb_handlers_setstartstr(h, f, encode_startstr, &attr); + upb_handlers_setendstr(h, f, encode_enddelimfield, &attr); + upb_handlers_setstring(h, f, encode_strbuf, &attr); + break; + case UPB_DESCRIPTOR_TYPE_MESSAGE: + upb_handlers_setstartsubmsg(h, f, encode_startdelimfield, &attr); + upb_handlers_setendsubmsg(h, f, encode_enddelimfield, &attr); + break; + case UPB_DESCRIPTOR_TYPE_GROUP: { + /* Endgroup takes a different tag (wire_type = END_GROUP). */ + upb_handlerattr attr2; + new_tag(h, f, UPB_WIRE_TYPE_END_GROUP, &attr2); + + upb_handlers_setstartsubmsg(h, f, encode_startgroup, &attr); + upb_handlers_setendsubmsg(h, f, encode_endgroup, &attr2); + + upb_handlerattr_uninit(&attr2); + break; + } + } + +#undef T + + upb_handlerattr_uninit(&attr); + } +} + +void upb_pb_encoder_reset(upb_pb_encoder *e) { + e->segptr = NULL; + e->top = NULL; + e->depth = 0; +} + + +/* public API *****************************************************************/ + +const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m, + const void *owner) { + return upb_handlers_newfrozen(m, owner, newhandlers_callback, NULL); +} + +upb_pb_encoder *upb_pb_encoder_create(upb_env *env, const upb_handlers *h, + upb_bytessink *output) { + const size_t initial_bufsize = 256; + const size_t initial_segbufsize = 16; + /* TODO(haberman): make this configurable. */ + const size_t stack_size = 64; +#ifndef NDEBUG + const size_t size_before = upb_env_bytesallocated(env); +#endif + + upb_pb_encoder *e = upb_env_malloc(env, sizeof(upb_pb_encoder)); + if (!e) return NULL; + + e->buf = upb_env_malloc(env, initial_bufsize); + e->segbuf = upb_env_malloc(env, initial_segbufsize * sizeof(*e->segbuf)); + e->stack = upb_env_malloc(env, stack_size * sizeof(*e->stack)); + + if (!e->buf || !e->segbuf || !e->stack) { + return NULL; + } + + e->limit = e->buf + initial_bufsize; + e->seglimit = e->segbuf + initial_segbufsize; + e->stacklimit = e->stack + stack_size; + + upb_pb_encoder_reset(e); + upb_sink_reset(&e->input_, h, e); + + e->env = env; + e->output_ = output; + e->subc = output->closure; + e->ptr = e->buf; + + /* If this fails, increase the value in encoder.h. */ + assert(upb_env_bytesallocated(env) - size_before <= UPB_PB_ENCODER_SIZE); + return e; +} + +upb_sink *upb_pb_encoder_input(upb_pb_encoder *e) { return &e->input_; } + + + +upb_filedef **upb_loaddescriptor(const char *buf, size_t n, const void *owner, + upb_status *status) { + /* Create handlers. */ + const upb_pbdecodermethod *decoder_m; + const upb_handlers *reader_h = upb_descreader_newhandlers(&reader_h); + upb_env env; + upb_pbdecodermethodopts opts; + upb_pbdecoder *decoder; + upb_descreader *reader; + bool ok; + size_t i; + upb_filedef **ret = NULL; + + upb_pbdecodermethodopts_init(&opts, reader_h); + decoder_m = upb_pbdecodermethod_new(&opts, &decoder_m); + + upb_env_init(&env); + upb_env_reporterrorsto(&env, status); + + reader = upb_descreader_create(&env, reader_h); + decoder = upb_pbdecoder_create(&env, decoder_m, upb_descreader_input(reader)); + + /* Push input data. */ + ok = upb_bufsrc_putbuf(buf, n, upb_pbdecoder_input(decoder)); + + if (!ok) { + goto cleanup; + } + + ret = upb_gmalloc(sizeof (*ret) * (upb_descreader_filecount(reader) + 1)); + + if (!ret) { + goto cleanup; + } + + for (i = 0; i < upb_descreader_filecount(reader); i++) { + ret[i] = upb_descreader_file(reader, i); + upb_filedef_ref(ret[i], owner); + } + + ret[i] = NULL; + +cleanup: + upb_env_uninit(&env); + upb_handlers_unref(reader_h, &reader_h); + upb_pbdecodermethod_unref(decoder_m, &decoder_m); + return ret; +} +/* + * upb::pb::TextPrinter + * + * OPT: This is not optimized at all. It uses printf() which parses the format + * string every time, and it allocates memory for every put. + */ + + +#include <ctype.h> +#include <float.h> +#include <inttypes.h> +#include <stdarg.h> +#include <stdio.h> +#include <string.h> + + +struct upb_textprinter { + upb_sink input_; + upb_bytessink *output_; + int indent_depth_; + bool single_line_; + void *subc; +}; + +#define CHECK(x) if ((x) < 0) goto err; + +static const char *shortname(const char *longname) { + const char *last = strrchr(longname, '.'); + return last ? last + 1 : longname; +} + +static int indent(upb_textprinter *p) { + int i; + if (!p->single_line_) + for (i = 0; i < p->indent_depth_; i++) + upb_bytessink_putbuf(p->output_, p->subc, " ", 2, NULL); + return 0; +} + +static int endfield(upb_textprinter *p) { + const char ch = (p->single_line_ ? ' ' : '\n'); + upb_bytessink_putbuf(p->output_, p->subc, &ch, 1, NULL); + return 0; +} + +static int putescaped(upb_textprinter *p, const char *buf, size_t len, + bool preserve_utf8) { + /* Based on CEscapeInternal() from Google's protobuf release. */ + char dstbuf[4096], *dst = dstbuf, *dstend = dstbuf + sizeof(dstbuf); + const char *end = buf + len; + + /* I think hex is prettier and more useful, but proto2 uses octal; should + * investigate whether it can parse hex also. */ + const bool use_hex = false; + bool last_hex_escape = false; /* true if last output char was \xNN */ + + for (; buf < end; buf++) { + bool is_hex_escape; + + if (dstend - dst < 4) { + upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL); + dst = dstbuf; + } + + is_hex_escape = false; + switch (*buf) { + case '\n': *(dst++) = '\\'; *(dst++) = 'n'; break; + case '\r': *(dst++) = '\\'; *(dst++) = 'r'; break; + case '\t': *(dst++) = '\\'; *(dst++) = 't'; break; + case '\"': *(dst++) = '\\'; *(dst++) = '\"'; break; + case '\'': *(dst++) = '\\'; *(dst++) = '\''; break; + case '\\': *(dst++) = '\\'; *(dst++) = '\\'; break; + default: + /* Note that if we emit \xNN and the buf character after that is a hex + * digit then that digit must be escaped too to prevent it being + * interpreted as part of the character code by C. */ + if ((!preserve_utf8 || (uint8_t)*buf < 0x80) && + (!isprint(*buf) || (last_hex_escape && isxdigit(*buf)))) { + sprintf(dst, (use_hex ? "\\x%02x" : "\\%03o"), (uint8_t)*buf); + is_hex_escape = use_hex; + dst += 4; + } else { + *(dst++) = *buf; break; + } + } + last_hex_escape = is_hex_escape; + } + /* Flush remaining data. */ + upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL); + return 0; +} + +bool putf(upb_textprinter *p, const char *fmt, ...) { + va_list args; + va_list args_copy; + char *str; + int written; + int len; + bool ok; + + va_start(args, fmt); + + /* Run once to get the length of the string. */ + _upb_va_copy(args_copy, args); + len = _upb_vsnprintf(NULL, 0, fmt, args_copy); + va_end(args_copy); + + /* + 1 for NULL terminator (vsprintf() requires it even if we don't). */ + str = upb_gmalloc(len + 1); + if (!str) return false; + written = vsprintf(str, fmt, args); + va_end(args); + UPB_ASSERT_VAR(written, written == len); + + ok = upb_bytessink_putbuf(p->output_, p->subc, str, len, NULL); + upb_gfree(str); + return ok; +} + + +/* handlers *******************************************************************/ + +static bool textprinter_startmsg(void *c, const void *hd) { + upb_textprinter *p = c; + UPB_UNUSED(hd); + if (p->indent_depth_ == 0) { + upb_bytessink_start(p->output_, 0, &p->subc); + } + return true; +} + +static bool textprinter_endmsg(void *c, const void *hd, upb_status *s) { + upb_textprinter *p = c; + UPB_UNUSED(hd); + UPB_UNUSED(s); + if (p->indent_depth_ == 0) { + upb_bytessink_end(p->output_); + } + return true; +} + +#define TYPE(name, ctype, fmt) \ + static bool textprinter_put ## name(void *closure, const void *handler_data, \ + ctype val) { \ + upb_textprinter *p = closure; \ + const upb_fielddef *f = handler_data; \ + CHECK(indent(p)); \ + putf(p, "%s: " fmt, upb_fielddef_name(f), val); \ + CHECK(endfield(p)); \ + return true; \ + err: \ + return false; \ +} + +static bool textprinter_putbool(void *closure, const void *handler_data, + bool val) { + upb_textprinter *p = closure; + const upb_fielddef *f = handler_data; + CHECK(indent(p)); + putf(p, "%s: %s", upb_fielddef_name(f), val ? "true" : "false"); + CHECK(endfield(p)); + return true; +err: + return false; +} + +#define STRINGIFY_HELPER(x) #x +#define STRINGIFY_MACROVAL(x) STRINGIFY_HELPER(x) + +TYPE(int32, int32_t, "%" PRId32) +TYPE(int64, int64_t, "%" PRId64) +TYPE(uint32, uint32_t, "%" PRIu32) +TYPE(uint64, uint64_t, "%" PRIu64) +TYPE(float, float, "%." STRINGIFY_MACROVAL(FLT_DIG) "g") +TYPE(double, double, "%." STRINGIFY_MACROVAL(DBL_DIG) "g") + +#undef TYPE + +/* Output a symbolic value from the enum if found, else just print as int32. */ +static bool textprinter_putenum(void *closure, const void *handler_data, + int32_t val) { + upb_textprinter *p = closure; + const upb_fielddef *f = handler_data; + const upb_enumdef *enum_def = upb_downcast_enumdef(upb_fielddef_subdef(f)); + const char *label = upb_enumdef_iton(enum_def, val); + if (label) { + indent(p); + putf(p, "%s: %s", upb_fielddef_name(f), label); + endfield(p); + } else { + if (!textprinter_putint32(closure, handler_data, val)) + return false; + } + return true; +} + +static void *textprinter_startstr(void *closure, const void *handler_data, + size_t size_hint) { + upb_textprinter *p = closure; + const upb_fielddef *f = handler_data; + UPB_UNUSED(size_hint); + indent(p); + putf(p, "%s: \"", upb_fielddef_name(f)); + return p; +} + +static bool textprinter_endstr(void *closure, const void *handler_data) { + upb_textprinter *p = closure; + UPB_UNUSED(handler_data); + putf(p, "\""); + endfield(p); + return true; +} + +static size_t textprinter_putstr(void *closure, const void *hd, const char *buf, + size_t len, const upb_bufhandle *handle) { + upb_textprinter *p = closure; + const upb_fielddef *f = hd; + UPB_UNUSED(handle); + CHECK(putescaped(p, buf, len, upb_fielddef_type(f) == UPB_TYPE_STRING)); + return len; +err: + return 0; +} + +static void *textprinter_startsubmsg(void *closure, const void *handler_data) { + upb_textprinter *p = closure; + const char *name = handler_data; + CHECK(indent(p)); + putf(p, "%s {%c", name, p->single_line_ ? ' ' : '\n'); + p->indent_depth_++; + return p; +err: + return UPB_BREAK; +} + +static bool textprinter_endsubmsg(void *closure, const void *handler_data) { + upb_textprinter *p = closure; + UPB_UNUSED(handler_data); + p->indent_depth_--; + CHECK(indent(p)); + upb_bytessink_putbuf(p->output_, p->subc, "}", 1, NULL); + CHECK(endfield(p)); + return true; +err: + return false; +} + +static void onmreg(const void *c, upb_handlers *h) { + const upb_msgdef *m = upb_handlers_msgdef(h); + upb_msg_field_iter i; + UPB_UNUSED(c); + + upb_handlers_setstartmsg(h, textprinter_startmsg, NULL); + upb_handlers_setendmsg(h, textprinter_endmsg, NULL); + + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, f); + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, f, textprinter_putint32, &attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, f, textprinter_putint64, &attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, f, textprinter_putuint32, &attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, f, textprinter_putuint64, &attr); + break; + case UPB_TYPE_FLOAT: + upb_handlers_setfloat(h, f, textprinter_putfloat, &attr); + break; + case UPB_TYPE_DOUBLE: + upb_handlers_setdouble(h, f, textprinter_putdouble, &attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, f, textprinter_putbool, &attr); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + upb_handlers_setstartstr(h, f, textprinter_startstr, &attr); + upb_handlers_setstring(h, f, textprinter_putstr, &attr); + upb_handlers_setendstr(h, f, textprinter_endstr, &attr); + break; + case UPB_TYPE_MESSAGE: { + const char *name = + upb_fielddef_istagdelim(f) + ? shortname(upb_msgdef_fullname(upb_fielddef_msgsubdef(f))) + : upb_fielddef_name(f); + upb_handlerattr_sethandlerdata(&attr, name); + upb_handlers_setstartsubmsg(h, f, textprinter_startsubmsg, &attr); + upb_handlers_setendsubmsg(h, f, textprinter_endsubmsg, &attr); + break; + } + case UPB_TYPE_ENUM: + upb_handlers_setint32(h, f, textprinter_putenum, &attr); + break; + } + } +} + +static void textprinter_reset(upb_textprinter *p, bool single_line) { + p->single_line_ = single_line; + p->indent_depth_ = 0; +} + + +/* Public API *****************************************************************/ + +upb_textprinter *upb_textprinter_create(upb_env *env, const upb_handlers *h, + upb_bytessink *output) { + upb_textprinter *p = upb_env_malloc(env, sizeof(upb_textprinter)); + if (!p) return NULL; + + p->output_ = output; + upb_sink_reset(&p->input_, h, p); + textprinter_reset(p, false); + + return p; +} + +const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m, + const void *owner) { + return upb_handlers_newfrozen(m, owner, &onmreg, NULL); +} + +upb_sink *upb_textprinter_input(upb_textprinter *p) { return &p->input_; } + +void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line) { + p->single_line_ = single_line; +} + + +/* Index is descriptor type. */ +const uint8_t upb_pb_native_wire_types[] = { + UPB_WIRE_TYPE_END_GROUP, /* ENDGROUP */ + UPB_WIRE_TYPE_64BIT, /* DOUBLE */ + UPB_WIRE_TYPE_32BIT, /* FLOAT */ + UPB_WIRE_TYPE_VARINT, /* INT64 */ + UPB_WIRE_TYPE_VARINT, /* UINT64 */ + UPB_WIRE_TYPE_VARINT, /* INT32 */ + UPB_WIRE_TYPE_64BIT, /* FIXED64 */ + UPB_WIRE_TYPE_32BIT, /* FIXED32 */ + UPB_WIRE_TYPE_VARINT, /* BOOL */ + UPB_WIRE_TYPE_DELIMITED, /* STRING */ + UPB_WIRE_TYPE_START_GROUP, /* GROUP */ + UPB_WIRE_TYPE_DELIMITED, /* MESSAGE */ + UPB_WIRE_TYPE_DELIMITED, /* BYTES */ + UPB_WIRE_TYPE_VARINT, /* UINT32 */ + UPB_WIRE_TYPE_VARINT, /* ENUM */ + UPB_WIRE_TYPE_32BIT, /* SFIXED32 */ + UPB_WIRE_TYPE_64BIT, /* SFIXED64 */ + UPB_WIRE_TYPE_VARINT, /* SINT32 */ + UPB_WIRE_TYPE_VARINT, /* SINT64 */ +}; + +/* A basic branch-based decoder, uses 32-bit values to get good performance + * on 32-bit architectures (but performs well on 64-bits also). + * This scheme comes from the original Google Protobuf implementation + * (proto2). */ +upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r) { + upb_decoderet err = {NULL, 0}; + const char *p = r.p; + uint32_t low = (uint32_t)r.val; + uint32_t high = 0; + uint32_t b; + b = *(p++); low |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done; + b = *(p++); low |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done; + b = *(p++); low |= (b & 0x7fU) << 28; + high = (b & 0x7fU) >> 4; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 3; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 10; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 17; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 24; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 31; if (!(b & 0x80)) goto done; + return err; + +done: + r.val = ((uint64_t)high << 32) | low; + r.p = p; + return r; +} + +/* Like the previous, but uses 64-bit values. */ +upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r) { + const char *p = r.p; + uint64_t val = r.val; + uint64_t b; + upb_decoderet err = {NULL, 0}; + b = *(p++); val |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 28; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 35; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 42; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 49; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 56; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 63; if (!(b & 0x80)) goto done; + return err; + +done: + r.val = val; + r.p = p; + return r; +} + +/* Given an encoded varint v, returns an integer with a single bit set that + * indicates the end of the varint. Subtracting one from this value will + * yield a mask that leaves only bits that are part of the varint. Returns + * 0 if the varint is unterminated. */ +static uint64_t upb_get_vstopbit(uint64_t v) { + uint64_t cbits = v | 0x7f7f7f7f7f7f7f7fULL; + return ~cbits & (cbits+1); +} + +/* A branchless decoder. Credit to Pascal Massimino for the bit-twiddling. */ +upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r) { + uint64_t b; + uint64_t stop_bit; + upb_decoderet my_r; + memcpy(&b, r.p, sizeof(b)); + stop_bit = upb_get_vstopbit(b); + b = (b & 0x7f7f7f7f7f7f7f7fULL) & (stop_bit - 1); + b += b & 0x007f007f007f007fULL; + b += 3 * (b & 0x0000ffff0000ffffULL); + b += 15 * (b & 0x00000000ffffffffULL); + if (stop_bit == 0) { + /* Error: unterminated varint. */ + upb_decoderet err_r = {(void*)0, 0}; + return err_r; + } + my_r = upb_decoderet_make(r.p + ((__builtin_ctzll(stop_bit) + 1) / 8), + r.val | (b << 7)); + return my_r; +} + +/* A branchless decoder. Credit to Daniel Wright for the bit-twiddling. */ +upb_decoderet upb_vdecode_max8_wright(upb_decoderet r) { + uint64_t b; + uint64_t stop_bit; + upb_decoderet my_r; + memcpy(&b, r.p, sizeof(b)); + stop_bit = upb_get_vstopbit(b); + b &= (stop_bit - 1); + b = ((b & 0x7f007f007f007f00ULL) >> 1) | (b & 0x007f007f007f007fULL); + b = ((b & 0xffff0000ffff0000ULL) >> 2) | (b & 0x0000ffff0000ffffULL); + b = ((b & 0xffffffff00000000ULL) >> 4) | (b & 0x00000000ffffffffULL); + if (stop_bit == 0) { + /* Error: unterminated varint. */ + upb_decoderet err_r = {(void*)0, 0}; + return err_r; + } + my_r = upb_decoderet_make(r.p + ((__builtin_ctzll(stop_bit) + 1) / 8), + r.val | (b << 14)); + return my_r; +} + +#line 1 "upb/json/parser.rl" +/* +** upb::json::Parser (upb_json_parser) +** +** A parser that uses the Ragel State Machine Compiler to generate +** the finite automata. +** +** Ragel only natively handles regular languages, but we can manually +** program it a bit to handle context-free languages like JSON, by using +** the "fcall" and "fret" constructs. +** +** This parser can handle the basics, but needs several things to be fleshed +** out: +** +** - handling of unicode escape sequences (including high surrogate pairs). +** - properly check and report errors for unknown fields, stack overflow, +** improper array nesting (or lack of nesting). +** - handling of base64 sequences with padding characters. +** - handling of push-back (non-success returns from sink functions). +** - handling of keys/escape-sequences/etc that span input buffers. +*/ + +#include <assert.h> +#include <errno.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> + + +#define UPB_JSON_MAX_DEPTH 64 + +typedef struct { + upb_sink sink; + + /* The current message in which we're parsing, and the field whose value we're + * expecting next. */ + const upb_msgdef *m; + const upb_fielddef *f; + + /* The table mapping json name to fielddef for this message. */ + upb_strtable *name_table; + + /* We are in a repeated-field context, ready to emit mapentries as + * submessages. This flag alters the start-of-object (open-brace) behavior to + * begin a sequence of mapentry messages rather than a single submessage. */ + bool is_map; + + /* We are in a map-entry message context. This flag is set when parsing the + * value field of a single map entry and indicates to all value-field parsers + * (subobjects, strings, numbers, and bools) that the map-entry submessage + * should end as soon as the value is parsed. */ + bool is_mapentry; + + /* If |is_map| or |is_mapentry| is true, |mapfield| refers to the parent + * message's map field that we're currently parsing. This differs from |f| + * because |f| is the field in the *current* message (i.e., the map-entry + * message itself), not the parent's field that leads to this map. */ + const upb_fielddef *mapfield; +} upb_jsonparser_frame; + +struct upb_json_parser { + upb_env *env; + const upb_json_parsermethod *method; + upb_bytessink input_; + + /* Stack to track the JSON scopes we are in. */ + upb_jsonparser_frame stack[UPB_JSON_MAX_DEPTH]; + upb_jsonparser_frame *top; + upb_jsonparser_frame *limit; + + upb_status status; + + /* Ragel's internal parsing stack for the parsing state machine. */ + int current_state; + int parser_stack[UPB_JSON_MAX_DEPTH]; + int parser_top; + + /* The handle for the current buffer. */ + const upb_bufhandle *handle; + + /* Accumulate buffer. See details in parser.rl. */ + const char *accumulated; + size_t accumulated_len; + char *accumulate_buf; + size_t accumulate_buf_size; + + /* Multi-part text data. See details in parser.rl. */ + int multipart_state; + upb_selector_t string_selector; + + /* Input capture. See details in parser.rl. */ + const char *capture; + + /* Intermediate result of parsing a unicode escape sequence. */ + uint32_t digit; +}; + +struct upb_json_parsermethod { + upb_refcounted base; + + upb_byteshandler input_handler_; + + /* Mainly for the purposes of refcounting, so all the fielddefs we point + * to stay alive. */ + const upb_msgdef *msg; + + /* Keys are upb_msgdef*, values are upb_strtable (json_name -> fielddef) */ + upb_inttable name_tables; +}; + +#define PARSER_CHECK_RETURN(x) if (!(x)) return false + +/* Used to signal that a capture has been suspended. */ +static char suspend_capture; + +static upb_selector_t getsel_for_handlertype(upb_json_parser *p, + upb_handlertype_t type) { + upb_selector_t sel; + bool ok = upb_handlers_getselector(p->top->f, type, &sel); + UPB_ASSERT_VAR(ok, ok); + return sel; +} + +static upb_selector_t parser_getsel(upb_json_parser *p) { + return getsel_for_handlertype( + p, upb_handlers_getprimitivehandlertype(p->top->f)); +} + +static bool check_stack(upb_json_parser *p) { + if ((p->top + 1) == p->limit) { + upb_status_seterrmsg(&p->status, "Nesting too deep"); + upb_env_reporterror(p->env, &p->status); + return false; + } + + return true; +} + +static void set_name_table(upb_json_parser *p, upb_jsonparser_frame *frame) { + upb_value v; + bool ok = upb_inttable_lookupptr(&p->method->name_tables, frame->m, &v); + UPB_ASSERT_VAR(ok, ok); + frame->name_table = upb_value_getptr(v); +} + +/* There are GCC/Clang built-ins for overflow checking which we could start + * using if there was any performance benefit to it. */ + +static bool checked_add(size_t a, size_t b, size_t *c) { + if (SIZE_MAX - a < b) return false; + *c = a + b; + return true; +} + +static size_t saturating_multiply(size_t a, size_t b) { + /* size_t is unsigned, so this is defined behavior even on overflow. */ + size_t ret = a * b; + if (b != 0 && ret / b != a) { + ret = SIZE_MAX; + } + return ret; +} + + +/* Base64 decoding ************************************************************/ + +/* TODO(haberman): make this streaming. */ + +static const signed char b64table[] = { + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, 62/*+*/, -1, -1, -1, 63/*/ */, + 52/*0*/, 53/*1*/, 54/*2*/, 55/*3*/, 56/*4*/, 57/*5*/, 58/*6*/, 59/*7*/, + 60/*8*/, 61/*9*/, -1, -1, -1, -1, -1, -1, + -1, 0/*A*/, 1/*B*/, 2/*C*/, 3/*D*/, 4/*E*/, 5/*F*/, 6/*G*/, + 07/*H*/, 8/*I*/, 9/*J*/, 10/*K*/, 11/*L*/, 12/*M*/, 13/*N*/, 14/*O*/, + 15/*P*/, 16/*Q*/, 17/*R*/, 18/*S*/, 19/*T*/, 20/*U*/, 21/*V*/, 22/*W*/, + 23/*X*/, 24/*Y*/, 25/*Z*/, -1, -1, -1, -1, -1, + -1, 26/*a*/, 27/*b*/, 28/*c*/, 29/*d*/, 30/*e*/, 31/*f*/, 32/*g*/, + 33/*h*/, 34/*i*/, 35/*j*/, 36/*k*/, 37/*l*/, 38/*m*/, 39/*n*/, 40/*o*/, + 41/*p*/, 42/*q*/, 43/*r*/, 44/*s*/, 45/*t*/, 46/*u*/, 47/*v*/, 48/*w*/, + 49/*x*/, 50/*y*/, 51/*z*/, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1 +}; + +/* Returns the table value sign-extended to 32 bits. Knowing that the upper + * bits will be 1 for unrecognized characters makes it easier to check for + * this error condition later (see below). */ +int32_t b64lookup(unsigned char ch) { return b64table[ch]; } + +/* Returns true if the given character is not a valid base64 character or + * padding. */ +bool nonbase64(unsigned char ch) { return b64lookup(ch) == -1 && ch != '='; } + +static bool base64_push(upb_json_parser *p, upb_selector_t sel, const char *ptr, + size_t len) { + const char *limit = ptr + len; + for (; ptr < limit; ptr += 4) { + uint32_t val; + char output[3]; + + if (limit - ptr < 4) { + upb_status_seterrf(&p->status, + "Base64 input for bytes field not a multiple of 4: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } + + val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12 | + b64lookup(ptr[2]) << 6 | + b64lookup(ptr[3]); + + /* Test the upper bit; returns true if any of the characters returned -1. */ + if (val & 0x80000000) { + goto otherchar; + } + + output[0] = val >> 16; + output[1] = (val >> 8) & 0xff; + output[2] = val & 0xff; + upb_sink_putstring(&p->top->sink, sel, output, 3, NULL); + } + return true; + +otherchar: + if (nonbase64(ptr[0]) || nonbase64(ptr[1]) || nonbase64(ptr[2]) || + nonbase64(ptr[3]) ) { + upb_status_seterrf(&p->status, + "Non-base64 characters in bytes field: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } if (ptr[2] == '=') { + uint32_t val; + char output; + + /* Last group contains only two input bytes, one output byte. */ + if (ptr[0] == '=' || ptr[1] == '=' || ptr[3] != '=') { + goto badpadding; + } + + val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12; + + assert(!(val & 0x80000000)); + output = val >> 16; + upb_sink_putstring(&p->top->sink, sel, &output, 1, NULL); + return true; + } else { + uint32_t val; + char output[2]; + + /* Last group contains only three input bytes, two output bytes. */ + if (ptr[0] == '=' || ptr[1] == '=' || ptr[2] == '=') { + goto badpadding; + } + + val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12 | + b64lookup(ptr[2]) << 6; + + output[0] = val >> 16; + output[1] = (val >> 8) & 0xff; + upb_sink_putstring(&p->top->sink, sel, output, 2, NULL); + return true; + } + +badpadding: + upb_status_seterrf(&p->status, + "Incorrect base64 padding for field: %s (%.*s)", + upb_fielddef_name(p->top->f), + 4, ptr); + upb_env_reporterror(p->env, &p->status); + return false; +} + + +/* Accumulate buffer **********************************************************/ + +/* Functionality for accumulating a buffer. + * + * Some parts of the parser need an entire value as a contiguous string. For + * example, to look up a member name in a hash table, or to turn a string into + * a number, the relevant library routines need the input string to be in + * contiguous memory, even if the value spanned two or more buffers in the + * input. These routines handle that. + * + * In the common case we can just point to the input buffer to get this + * contiguous string and avoid any actual copy. So we optimistically begin + * this way. But there are a few cases where we must instead copy into a + * separate buffer: + * + * 1. The string was not contiguous in the input (it spanned buffers). + * + * 2. The string included escape sequences that need to be interpreted to get + * the true value in a contiguous buffer. */ + +static void assert_accumulate_empty(upb_json_parser *p) { + UPB_UNUSED(p); + assert(p->accumulated == NULL); + assert(p->accumulated_len == 0); +} + +static void accumulate_clear(upb_json_parser *p) { + p->accumulated = NULL; + p->accumulated_len = 0; +} + +/* Used internally by accumulate_append(). */ +static bool accumulate_realloc(upb_json_parser *p, size_t need) { + void *mem; + size_t old_size = p->accumulate_buf_size; + size_t new_size = UPB_MAX(old_size, 128); + while (new_size < need) { + new_size = saturating_multiply(new_size, 2); + } + + mem = upb_env_realloc(p->env, p->accumulate_buf, old_size, new_size); + if (!mem) { + upb_status_seterrmsg(&p->status, "Out of memory allocating buffer."); + upb_env_reporterror(p->env, &p->status); + return false; + } + + p->accumulate_buf = mem; + p->accumulate_buf_size = new_size; + return true; +} + +/* Logically appends the given data to the append buffer. + * If "can_alias" is true, we will try to avoid actually copying, but the buffer + * must be valid until the next accumulate_append() call (if any). */ +static bool accumulate_append(upb_json_parser *p, const char *buf, size_t len, + bool can_alias) { + size_t need; + + if (!p->accumulated && can_alias) { + p->accumulated = buf; + p->accumulated_len = len; + return true; + } + + if (!checked_add(p->accumulated_len, len, &need)) { + upb_status_seterrmsg(&p->status, "Integer overflow."); + upb_env_reporterror(p->env, &p->status); + return false; + } + + if (need > p->accumulate_buf_size && !accumulate_realloc(p, need)) { + return false; + } + + if (p->accumulated != p->accumulate_buf) { + memcpy(p->accumulate_buf, p->accumulated, p->accumulated_len); + p->accumulated = p->accumulate_buf; + } + + memcpy(p->accumulate_buf + p->accumulated_len, buf, len); + p->accumulated_len += len; + return true; +} + +/* Returns a pointer to the data accumulated since the last accumulate_clear() + * call, and writes the length to *len. This with point either to the input + * buffer or a temporary accumulate buffer. */ +static const char *accumulate_getptr(upb_json_parser *p, size_t *len) { + assert(p->accumulated); + *len = p->accumulated_len; + return p->accumulated; +} + + +/* Mult-part text data ********************************************************/ + +/* When we have text data in the input, it can often come in multiple segments. + * For example, there may be some raw string data followed by an escape + * sequence. The two segments are processed with different logic. Also buffer + * seams in the input can cause multiple segments. + * + * As we see segments, there are two main cases for how we want to process them: + * + * 1. we want to push the captured input directly to string handlers. + * + * 2. we need to accumulate all the parts into a contiguous buffer for further + * processing (field name lookup, string->number conversion, etc). */ + +/* This is the set of states for p->multipart_state. */ +enum { + /* We are not currently processing multipart data. */ + MULTIPART_INACTIVE = 0, + + /* We are processing multipart data by accumulating it into a contiguous + * buffer. */ + MULTIPART_ACCUMULATE = 1, + + /* We are processing multipart data by pushing each part directly to the + * current string handlers. */ + MULTIPART_PUSHEAGERLY = 2 +}; + +/* Start a multi-part text value where we accumulate the data for processing at + * the end. */ +static void multipart_startaccum(upb_json_parser *p) { + assert_accumulate_empty(p); + assert(p->multipart_state == MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_ACCUMULATE; +} + +/* Start a multi-part text value where we immediately push text data to a string + * value with the given selector. */ +static void multipart_start(upb_json_parser *p, upb_selector_t sel) { + assert_accumulate_empty(p); + assert(p->multipart_state == MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_PUSHEAGERLY; + p->string_selector = sel; +} + +static bool multipart_text(upb_json_parser *p, const char *buf, size_t len, + bool can_alias) { + switch (p->multipart_state) { + case MULTIPART_INACTIVE: + upb_status_seterrmsg( + &p->status, "Internal error: unexpected state MULTIPART_INACTIVE"); + upb_env_reporterror(p->env, &p->status); + return false; + + case MULTIPART_ACCUMULATE: + if (!accumulate_append(p, buf, len, can_alias)) { + return false; + } + break; + + case MULTIPART_PUSHEAGERLY: { + const upb_bufhandle *handle = can_alias ? p->handle : NULL; + upb_sink_putstring(&p->top->sink, p->string_selector, buf, len, handle); + break; + } + } + + return true; +} + +/* Note: this invalidates the accumulate buffer! Call only after reading its + * contents. */ +static void multipart_end(upb_json_parser *p) { + assert(p->multipart_state != MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_INACTIVE; + accumulate_clear(p); +} + + +/* Input capture **************************************************************/ + +/* Functionality for capturing a region of the input as text. Gracefully + * handles the case where a buffer seam occurs in the middle of the captured + * region. */ + +static void capture_begin(upb_json_parser *p, const char *ptr) { + assert(p->multipart_state != MULTIPART_INACTIVE); + assert(p->capture == NULL); + p->capture = ptr; +} + +static bool capture_end(upb_json_parser *p, const char *ptr) { + assert(p->capture); + if (multipart_text(p, p->capture, ptr - p->capture, true)) { + p->capture = NULL; + return true; + } else { + return false; + } +} + +/* This is called at the end of each input buffer (ie. when we have hit a + * buffer seam). If we are in the middle of capturing the input, this + * processes the unprocessed capture region. */ +static void capture_suspend(upb_json_parser *p, const char **ptr) { + if (!p->capture) return; + + if (multipart_text(p, p->capture, *ptr - p->capture, false)) { + /* We use this as a signal that we were in the middle of capturing, and + * that capturing should resume at the beginning of the next buffer. + * + * We can't use *ptr here, because we have no guarantee that this pointer + * will be valid when we resume (if the underlying memory is freed, then + * using the pointer at all, even to compare to NULL, is likely undefined + * behavior). */ + p->capture = &suspend_capture; + } else { + /* Need to back up the pointer to the beginning of the capture, since + * we were not able to actually preserve it. */ + *ptr = p->capture; + } +} + +static void capture_resume(upb_json_parser *p, const char *ptr) { + if (p->capture) { + assert(p->capture == &suspend_capture); + p->capture = ptr; + } +} + + +/* Callbacks from the parser **************************************************/ + +/* These are the functions called directly from the parser itself. + * We define these in the same order as their declarations in the parser. */ + +static char escape_char(char in) { + switch (in) { + case 'r': return '\r'; + case 't': return '\t'; + case 'n': return '\n'; + case 'f': return '\f'; + case 'b': return '\b'; + case '/': return '/'; + case '"': return '"'; + case '\\': return '\\'; + default: + assert(0); + return 'x'; + } +} + +static bool escape(upb_json_parser *p, const char *ptr) { + char ch = escape_char(*ptr); + return multipart_text(p, &ch, 1, false); +} + +static void start_hex(upb_json_parser *p) { + p->digit = 0; +} + +static void hexdigit(upb_json_parser *p, const char *ptr) { + char ch = *ptr; + + p->digit <<= 4; + + if (ch >= '0' && ch <= '9') { + p->digit += (ch - '0'); + } else if (ch >= 'a' && ch <= 'f') { + p->digit += ((ch - 'a') + 10); + } else { + assert(ch >= 'A' && ch <= 'F'); + p->digit += ((ch - 'A') + 10); + } +} + +static bool end_hex(upb_json_parser *p) { + uint32_t codepoint = p->digit; + + /* emit the codepoint as UTF-8. */ + char utf8[3]; /* support \u0000 -- \uFFFF -- need only three bytes. */ + int length = 0; + if (codepoint <= 0x7F) { + utf8[0] = codepoint; + length = 1; + } else if (codepoint <= 0x07FF) { + utf8[1] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[0] = (codepoint & 0x1F) | 0xC0; + length = 2; + } else /* codepoint <= 0xFFFF */ { + utf8[2] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[1] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[0] = (codepoint & 0x0F) | 0xE0; + length = 3; + } + /* TODO(haberman): Handle high surrogates: if codepoint is a high surrogate + * we have to wait for the next escape to get the full code point). */ + + return multipart_text(p, utf8, length, false); +} + +static void start_text(upb_json_parser *p, const char *ptr) { + capture_begin(p, ptr); +} + +static bool end_text(upb_json_parser *p, const char *ptr) { + return capture_end(p, ptr); +} + +static void start_number(upb_json_parser *p, const char *ptr) { + multipart_startaccum(p); + capture_begin(p, ptr); +} + +static bool parse_number(upb_json_parser *p); + +static bool end_number(upb_json_parser *p, const char *ptr) { + if (!capture_end(p, ptr)) { + return false; + } + + return parse_number(p); +} + +static bool parse_number(upb_json_parser *p) { + size_t len; + const char *buf; + const char *myend; + char *end; + + /* strtol() and friends unfortunately do not support specifying the length of + * the input string, so we need to force a copy into a NULL-terminated buffer. */ + if (!multipart_text(p, "\0", 1, false)) { + return false; + } + + buf = accumulate_getptr(p, &len); + myend = buf + len - 1; /* One for NULL. */ + + /* XXX: We are using strtol to parse integers, but this is wrong as even + * integers can be represented as 1e6 (for example), which strtol can't + * handle correctly. + * + * XXX: Also, we can't handle large integers properly because strto[u]ll + * isn't in C89. + * + * XXX: Also, we don't properly check floats for overflow, since strtof + * isn't in C89. */ + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_ENUM: + case UPB_TYPE_INT32: { + long val = strtol(p->accumulated, &end, 0); + if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putint32(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_INT64: { + long long val = strtol(p->accumulated, &end, 0); + if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putint64(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_UINT32: { + unsigned long val = strtoul(p->accumulated, &end, 0); + if (val > UINT32_MAX || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putuint32(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_UINT64: { + unsigned long long val = strtoul(p->accumulated, &end, 0); + if (val > UINT64_MAX || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putuint64(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_DOUBLE: { + double val = strtod(p->accumulated, &end); + if (errno == ERANGE || end != myend) + goto err; + else + upb_sink_putdouble(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_FLOAT: { + float val = strtod(p->accumulated, &end); + if (errno == ERANGE || end != myend) + goto err; + else + upb_sink_putfloat(&p->top->sink, parser_getsel(p), val); + break; + } + default: + assert(false); + } + + multipart_end(p); + + return true; + +err: + upb_status_seterrf(&p->status, "error parsing number: %s", buf); + upb_env_reporterror(p->env, &p->status); + multipart_end(p); + return false; +} + +static bool parser_putbool(upb_json_parser *p, bool val) { + bool ok; + + if (upb_fielddef_type(p->top->f) != UPB_TYPE_BOOL) { + upb_status_seterrf(&p->status, + "Boolean value specified for non-bool field: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } + + ok = upb_sink_putbool(&p->top->sink, parser_getsel(p), val); + UPB_ASSERT_VAR(ok, ok); + + return true; +} + +static bool start_stringval(upb_json_parser *p) { + assert(p->top->f); + + if (upb_fielddef_isstring(p->top->f)) { + upb_jsonparser_frame *inner; + upb_selector_t sel; + + if (!check_stack(p)) return false; + + /* Start a new parser frame: parser frames correspond one-to-one with + * handler frames, and string events occur in a sub-frame. */ + inner = p->top + 1; + sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR); + upb_sink_startstr(&p->top->sink, sel, 0, &inner->sink); + inner->m = p->top->m; + inner->f = p->top->f; + inner->name_table = NULL; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + if (upb_fielddef_type(p->top->f) == UPB_TYPE_STRING) { + /* For STRING fields we push data directly to the handlers as it is + * parsed. We don't do this yet for BYTES fields, because our base64 + * decoder is not streaming. + * + * TODO(haberman): make base64 decoding streaming also. */ + multipart_start(p, getsel_for_handlertype(p, UPB_HANDLER_STRING)); + return true; + } else { + multipart_startaccum(p); + return true; + } + } else if (upb_fielddef_type(p->top->f) == UPB_TYPE_ENUM) { + /* No need to push a frame -- symbolic enum names in quotes remain in the + * current parser frame. + * + * Enum string values must accumulate so we can look up the value in a table + * once it is complete. */ + multipart_startaccum(p); + return true; + } else { + upb_status_seterrf(&p->status, + "String specified for non-string/non-enum field: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } +} + +static bool end_stringval(upb_json_parser *p) { + bool ok = true; + + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_BYTES: + if (!base64_push(p, getsel_for_handlertype(p, UPB_HANDLER_STRING), + p->accumulated, p->accumulated_len)) { + return false; + } + /* Fall through. */ + + case UPB_TYPE_STRING: { + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR); + upb_sink_endstr(&p->top->sink, sel); + p->top--; + break; + } + + case UPB_TYPE_ENUM: { + /* Resolve enum symbolic name to integer value. */ + const upb_enumdef *enumdef = + (const upb_enumdef*)upb_fielddef_subdef(p->top->f); + + size_t len; + const char *buf = accumulate_getptr(p, &len); + + int32_t int_val = 0; + ok = upb_enumdef_ntoi(enumdef, buf, len, &int_val); + + if (ok) { + upb_selector_t sel = parser_getsel(p); + upb_sink_putint32(&p->top->sink, sel, int_val); + } else { + upb_status_seterrf(&p->status, "Enum value unknown: '%.*s'", len, buf); + upb_env_reporterror(p->env, &p->status); + } + + break; + } + + default: + assert(false); + upb_status_seterrmsg(&p->status, "Internal error in JSON decoder"); + upb_env_reporterror(p->env, &p->status); + ok = false; + break; + } + + multipart_end(p); + + return ok; +} + +static void start_member(upb_json_parser *p) { + assert(!p->top->f); + multipart_startaccum(p); +} + +/* Helper: invoked during parse_mapentry() to emit the mapentry message's key + * field based on the current contents of the accumulate buffer. */ +static bool parse_mapentry_key(upb_json_parser *p) { + + size_t len; + const char *buf = accumulate_getptr(p, &len); + + /* Emit the key field. We do a bit of ad-hoc parsing here because the + * parser state machine has already decided that this is a string field + * name, and we are reinterpreting it as some arbitrary key type. In + * particular, integer and bool keys are quoted, so we need to parse the + * quoted string contents here. */ + + p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_KEY); + if (p->top->f == NULL) { + upb_status_seterrmsg(&p->status, "mapentry message has no key"); + upb_env_reporterror(p->env, &p->status); + return false; + } + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + /* Invoke end_number. The accum buffer has the number's text already. */ + if (!parse_number(p)) { + return false; + } + break; + case UPB_TYPE_BOOL: + if (len == 4 && !strncmp(buf, "true", 4)) { + if (!parser_putbool(p, true)) { + return false; + } + } else if (len == 5 && !strncmp(buf, "false", 5)) { + if (!parser_putbool(p, false)) { + return false; + } + } else { + upb_status_seterrmsg(&p->status, + "Map bool key not 'true' or 'false'"); + upb_env_reporterror(p->env, &p->status); + return false; + } + multipart_end(p); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + upb_sink subsink; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR); + upb_sink_startstr(&p->top->sink, sel, len, &subsink); + sel = getsel_for_handlertype(p, UPB_HANDLER_STRING); + upb_sink_putstring(&subsink, sel, buf, len, NULL); + sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR); + upb_sink_endstr(&subsink, sel); + multipart_end(p); + break; + } + default: + upb_status_seterrmsg(&p->status, "Invalid field type for map key"); + upb_env_reporterror(p->env, &p->status); + return false; + } + + return true; +} + +/* Helper: emit one map entry (as a submessage in the map field sequence). This + * is invoked from end_membername(), at the end of the map entry's key string, + * with the map key in the accumulate buffer. It parses the key from that + * buffer, emits the handler calls to start the mapentry submessage (setting up + * its subframe in the process), and sets up state in the subframe so that the + * value parser (invoked next) will emit the mapentry's value field and then + * end the mapentry message. */ + +static bool handle_mapentry(upb_json_parser *p) { + const upb_fielddef *mapfield; + const upb_msgdef *mapentrymsg; + upb_jsonparser_frame *inner; + upb_selector_t sel; + + /* Map entry: p->top->sink is the seq frame, so we need to start a frame + * for the mapentry itself, and then set |f| in that frame so that the map + * value field is parsed, and also set a flag to end the frame after the + * map-entry value is parsed. */ + if (!check_stack(p)) return false; + + mapfield = p->top->mapfield; + mapentrymsg = upb_fielddef_msgsubdef(mapfield); + + inner = p->top + 1; + p->top->f = mapfield; + sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG); + upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink); + inner->m = mapentrymsg; + inner->name_table = NULL; + inner->mapfield = mapfield; + inner->is_map = false; + + /* Don't set this to true *yet* -- we reuse parsing handlers below to push + * the key field value to the sink, and these handlers will pop the frame + * if they see is_mapentry (when invoked by the parser state machine, they + * would have just seen the map-entry value, not key). */ + inner->is_mapentry = false; + p->top = inner; + + /* send STARTMSG in submsg frame. */ + upb_sink_startmsg(&p->top->sink); + + parse_mapentry_key(p); + + /* Set up the value field to receive the map-entry value. */ + p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_VALUE); + p->top->is_mapentry = true; /* set up to pop frame after value is parsed. */ + p->top->mapfield = mapfield; + if (p->top->f == NULL) { + upb_status_seterrmsg(&p->status, "mapentry message has no value"); + upb_env_reporterror(p->env, &p->status); + return false; + } + + return true; +} + +static bool end_membername(upb_json_parser *p) { + assert(!p->top->f); + + if (p->top->is_map) { + return handle_mapentry(p); + } else { + size_t len; + const char *buf = accumulate_getptr(p, &len); + upb_value v; + + if (upb_strtable_lookup2(p->top->name_table, buf, len, &v)) { + p->top->f = upb_value_getconstptr(v); + multipart_end(p); + + return true; + } else { + /* TODO(haberman): Ignore unknown fields if requested/configured to do + * so. */ + upb_status_seterrf(&p->status, "No such field: %.*s\n", (int)len, buf); + upb_env_reporterror(p->env, &p->status); + return false; + } + } +} + +static void end_member(upb_json_parser *p) { + /* If we just parsed a map-entry value, end that frame too. */ + if (p->top->is_mapentry) { + upb_status s = UPB_STATUS_INIT; + upb_selector_t sel; + bool ok; + const upb_fielddef *mapfield; + + assert(p->top > p->stack); + /* send ENDMSG on submsg. */ + upb_sink_endmsg(&p->top->sink, &s); + mapfield = p->top->mapfield; + + /* send ENDSUBMSG in repeated-field-of-mapentries frame. */ + p->top--; + ok = upb_handlers_getselector(mapfield, UPB_HANDLER_ENDSUBMSG, &sel); + UPB_ASSERT_VAR(ok, ok); + upb_sink_endsubmsg(&p->top->sink, sel); + } + + p->top->f = NULL; +} + +static bool start_subobject(upb_json_parser *p) { + assert(p->top->f); + + if (upb_fielddef_ismap(p->top->f)) { + upb_jsonparser_frame *inner; + upb_selector_t sel; + + /* Beginning of a map. Start a new parser frame in a repeated-field + * context. */ + if (!check_stack(p)) return false; + + inner = p->top + 1; + sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ); + upb_sink_startseq(&p->top->sink, sel, &inner->sink); + inner->m = upb_fielddef_msgsubdef(p->top->f); + inner->name_table = NULL; + inner->mapfield = p->top->f; + inner->f = NULL; + inner->is_map = true; + inner->is_mapentry = false; + p->top = inner; + + return true; + } else if (upb_fielddef_issubmsg(p->top->f)) { + upb_jsonparser_frame *inner; + upb_selector_t sel; + + /* Beginning of a subobject. Start a new parser frame in the submsg + * context. */ + if (!check_stack(p)) return false; + + inner = p->top + 1; + + sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG); + upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink); + inner->m = upb_fielddef_msgsubdef(p->top->f); + set_name_table(p, inner); + inner->f = NULL; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + return true; + } else { + upb_status_seterrf(&p->status, + "Object specified for non-message/group field: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } +} + +static void end_subobject(upb_json_parser *p) { + if (p->top->is_map) { + upb_selector_t sel; + p->top--; + sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ); + upb_sink_endseq(&p->top->sink, sel); + } else { + upb_selector_t sel; + p->top--; + sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSUBMSG); + upb_sink_endsubmsg(&p->top->sink, sel); + } +} + +static bool start_array(upb_json_parser *p) { + upb_jsonparser_frame *inner; + upb_selector_t sel; + + assert(p->top->f); + + if (!upb_fielddef_isseq(p->top->f)) { + upb_status_seterrf(&p->status, + "Array specified for non-repeated field: %s", + upb_fielddef_name(p->top->f)); + upb_env_reporterror(p->env, &p->status); + return false; + } + + if (!check_stack(p)) return false; + + inner = p->top + 1; + sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ); + upb_sink_startseq(&p->top->sink, sel, &inner->sink); + inner->m = p->top->m; + inner->name_table = NULL; + inner->f = p->top->f; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + return true; +} + +static void end_array(upb_json_parser *p) { + upb_selector_t sel; + + assert(p->top > p->stack); + + p->top--; + sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ); + upb_sink_endseq(&p->top->sink, sel); +} + +static void start_object(upb_json_parser *p) { + if (!p->top->is_map) { + upb_sink_startmsg(&p->top->sink); + } +} + +static void end_object(upb_json_parser *p) { + if (!p->top->is_map) { + upb_status status; + upb_status_clear(&status); + upb_sink_endmsg(&p->top->sink, &status); + if (!upb_ok(&status)) { + upb_env_reporterror(p->env, &status); + } + } +} + + +#define CHECK_RETURN_TOP(x) if (!(x)) goto error + + +/* The actual parser **********************************************************/ + +/* What follows is the Ragel parser itself. The language is specified in Ragel + * and the actions call our C functions above. + * + * Ragel has an extensive set of functionality, and we use only a small part of + * it. There are many action types but we only use a few: + * + * ">" -- transition into a machine + * "%" -- transition out of a machine + * "@" -- transition into a final state of a machine. + * + * "@" transitions are tricky because a machine can transition into a final + * state repeatedly. But in some cases we know this can't happen, for example + * a string which is delimited by a final '"' can only transition into its + * final state once, when the closing '"' is seen. */ + + +#line 1245 "upb/json/parser.rl" + + + +#line 1157 "upb/json/parser.c" +static const char _json_actions[] = { + 0, 1, 0, 1, 2, 1, 3, 1, + 5, 1, 6, 1, 7, 1, 8, 1, + 10, 1, 12, 1, 13, 1, 14, 1, + 15, 1, 16, 1, 17, 1, 21, 1, + 25, 1, 27, 2, 3, 8, 2, 4, + 5, 2, 6, 2, 2, 6, 8, 2, + 11, 9, 2, 13, 15, 2, 14, 15, + 2, 18, 1, 2, 19, 27, 2, 20, + 9, 2, 22, 27, 2, 23, 27, 2, + 24, 27, 2, 26, 27, 3, 14, 11, + 9 +}; + +static const unsigned char _json_key_offsets[] = { + 0, 0, 4, 9, 14, 15, 19, 24, + 29, 34, 38, 42, 45, 48, 50, 54, + 58, 60, 62, 67, 69, 71, 80, 86, + 92, 98, 104, 106, 115, 116, 116, 116, + 121, 126, 131, 132, 133, 134, 135, 135, + 136, 137, 138, 138, 139, 140, 141, 141, + 146, 151, 152, 156, 161, 166, 171, 175, + 175, 178, 178, 178 +}; + +static const char _json_trans_keys[] = { + 32, 123, 9, 13, 32, 34, 125, 9, + 13, 32, 34, 125, 9, 13, 34, 32, + 58, 9, 13, 32, 93, 125, 9, 13, + 32, 44, 125, 9, 13, 32, 44, 125, + 9, 13, 32, 34, 9, 13, 45, 48, + 49, 57, 48, 49, 57, 46, 69, 101, + 48, 57, 69, 101, 48, 57, 43, 45, + 48, 57, 48, 57, 48, 57, 46, 69, + 101, 48, 57, 34, 92, 34, 92, 34, + 47, 92, 98, 102, 110, 114, 116, 117, + 48, 57, 65, 70, 97, 102, 48, 57, + 65, 70, 97, 102, 48, 57, 65, 70, + 97, 102, 48, 57, 65, 70, 97, 102, + 34, 92, 34, 45, 91, 102, 110, 116, + 123, 48, 57, 34, 32, 93, 125, 9, + 13, 32, 44, 93, 9, 13, 32, 93, + 125, 9, 13, 97, 108, 115, 101, 117, + 108, 108, 114, 117, 101, 32, 34, 125, + 9, 13, 32, 34, 125, 9, 13, 34, + 32, 58, 9, 13, 32, 93, 125, 9, + 13, 32, 44, 125, 9, 13, 32, 44, + 125, 9, 13, 32, 34, 9, 13, 32, + 9, 13, 0 +}; + +static const char _json_single_lengths[] = { + 0, 2, 3, 3, 1, 2, 3, 3, + 3, 2, 2, 1, 3, 0, 2, 2, + 0, 0, 3, 2, 2, 9, 0, 0, + 0, 0, 2, 7, 1, 0, 0, 3, + 3, 3, 1, 1, 1, 1, 0, 1, + 1, 1, 0, 1, 1, 1, 0, 3, + 3, 1, 2, 3, 3, 3, 2, 0, + 1, 0, 0, 0 +}; + +static const char _json_range_lengths[] = { + 0, 1, 1, 1, 0, 1, 1, 1, + 1, 1, 1, 1, 0, 1, 1, 1, + 1, 1, 1, 0, 0, 0, 3, 3, + 3, 3, 0, 1, 0, 0, 0, 1, + 1, 1, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 1, + 1, 0, 1, 1, 1, 1, 1, 0, + 1, 0, 0, 0 +}; + +static const short _json_index_offsets[] = { + 0, 0, 4, 9, 14, 16, 20, 25, + 30, 35, 39, 43, 46, 50, 52, 56, + 60, 62, 64, 69, 72, 75, 85, 89, + 93, 97, 101, 104, 113, 115, 116, 117, + 122, 127, 132, 134, 136, 138, 140, 141, + 143, 145, 147, 148, 150, 152, 154, 155, + 160, 165, 167, 171, 176, 181, 186, 190, + 191, 194, 195, 196 +}; + +static const char _json_indicies[] = { + 0, 2, 0, 1, 3, 4, 5, 3, + 1, 6, 7, 8, 6, 1, 9, 1, + 10, 11, 10, 1, 11, 1, 1, 11, + 12, 13, 14, 15, 13, 1, 16, 17, + 8, 16, 1, 17, 7, 17, 1, 18, + 19, 20, 1, 19, 20, 1, 22, 23, + 23, 21, 24, 1, 23, 23, 24, 21, + 25, 25, 26, 1, 26, 1, 26, 21, + 22, 23, 23, 20, 21, 28, 29, 27, + 31, 32, 30, 33, 33, 33, 33, 33, + 33, 33, 33, 34, 1, 35, 35, 35, + 1, 36, 36, 36, 1, 37, 37, 37, + 1, 38, 38, 38, 1, 40, 41, 39, + 42, 43, 44, 45, 46, 47, 48, 43, + 1, 49, 1, 50, 51, 53, 54, 1, + 53, 52, 55, 56, 54, 55, 1, 56, + 1, 1, 56, 52, 57, 1, 58, 1, + 59, 1, 60, 1, 61, 62, 1, 63, + 1, 64, 1, 65, 66, 1, 67, 1, + 68, 1, 69, 70, 71, 72, 70, 1, + 73, 74, 75, 73, 1, 76, 1, 77, + 78, 77, 1, 78, 1, 1, 78, 79, + 80, 81, 82, 80, 1, 83, 84, 75, + 83, 1, 84, 74, 84, 1, 85, 86, + 86, 1, 1, 1, 1, 0 +}; + +static const char _json_trans_targs[] = { + 1, 0, 2, 3, 4, 56, 3, 4, + 56, 5, 5, 6, 7, 8, 9, 56, + 8, 9, 11, 12, 18, 57, 13, 15, + 14, 16, 17, 20, 58, 21, 20, 58, + 21, 19, 22, 23, 24, 25, 26, 20, + 58, 21, 28, 30, 31, 34, 39, 43, + 47, 29, 59, 59, 32, 31, 29, 32, + 33, 35, 36, 37, 38, 59, 40, 41, + 42, 59, 44, 45, 46, 59, 48, 49, + 55, 48, 49, 55, 50, 50, 51, 52, + 53, 54, 55, 53, 54, 59, 56 +}; + +static const char _json_trans_actions[] = { + 0, 0, 0, 21, 77, 53, 0, 47, + 23, 17, 0, 0, 15, 19, 19, 50, + 0, 0, 0, 0, 0, 1, 0, 0, + 0, 0, 0, 3, 13, 0, 0, 35, + 5, 11, 0, 38, 7, 7, 7, 41, + 44, 9, 62, 56, 25, 0, 0, 0, + 31, 29, 33, 59, 15, 0, 27, 0, + 0, 0, 0, 0, 0, 68, 0, 0, + 0, 71, 0, 0, 0, 65, 21, 77, + 53, 0, 47, 23, 17, 0, 0, 15, + 19, 19, 50, 0, 0, 74, 0 +}; + +static const int json_start = 1; + +static const int json_en_number_machine = 10; +static const int json_en_string_machine = 19; +static const int json_en_value_machine = 27; +static const int json_en_main = 1; + + +#line 1248 "upb/json/parser.rl" + +size_t parse(void *closure, const void *hd, const char *buf, size_t size, + const upb_bufhandle *handle) { + upb_json_parser *parser = closure; + + /* Variables used by Ragel's generated code. */ + int cs = parser->current_state; + int *stack = parser->parser_stack; + int top = parser->parser_top; + + const char *p = buf; + const char *pe = buf + size; + + parser->handle = handle; + + UPB_UNUSED(hd); + UPB_UNUSED(handle); + + capture_resume(parser, buf); + + +#line 1328 "upb/json/parser.c" + { + int _klen; + unsigned int _trans; + const char *_acts; + unsigned int _nacts; + const char *_keys; + + if ( p == pe ) + goto _test_eof; + if ( cs == 0 ) + goto _out; +_resume: + _keys = _json_trans_keys + _json_key_offsets[cs]; + _trans = _json_index_offsets[cs]; + + _klen = _json_single_lengths[cs]; + if ( _klen > 0 ) { + const char *_lower = _keys; + const char *_mid; + const char *_upper = _keys + _klen - 1; + while (1) { + if ( _upper < _lower ) + break; + + _mid = _lower + ((_upper-_lower) >> 1); + if ( (*p) < *_mid ) + _upper = _mid - 1; + else if ( (*p) > *_mid ) + _lower = _mid + 1; + else { + _trans += (unsigned int)(_mid - _keys); + goto _match; + } + } + _keys += _klen; + _trans += _klen; + } + + _klen = _json_range_lengths[cs]; + if ( _klen > 0 ) { + const char *_lower = _keys; + const char *_mid; + const char *_upper = _keys + (_klen<<1) - 2; + while (1) { + if ( _upper < _lower ) + break; + + _mid = _lower + (((_upper-_lower) >> 1) & ~1); + if ( (*p) < _mid[0] ) + _upper = _mid - 2; + else if ( (*p) > _mid[1] ) + _lower = _mid + 2; + else { + _trans += (unsigned int)((_mid - _keys)>>1); + goto _match; + } + } + _trans += _klen; + } + +_match: + _trans = _json_indicies[_trans]; + cs = _json_trans_targs[_trans]; + + if ( _json_trans_actions[_trans] == 0 ) + goto _again; + + _acts = _json_actions + _json_trans_actions[_trans]; + _nacts = (unsigned int) *_acts++; + while ( _nacts-- > 0 ) + { + switch ( *_acts++ ) + { + case 0: +#line 1160 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; + case 1: +#line 1161 "upb/json/parser.rl" + { p--; {stack[top++] = cs; cs = 10; goto _again;} } + break; + case 2: +#line 1165 "upb/json/parser.rl" + { start_text(parser, p); } + break; + case 3: +#line 1166 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_text(parser, p)); } + break; + case 4: +#line 1172 "upb/json/parser.rl" + { start_hex(parser); } + break; + case 5: +#line 1173 "upb/json/parser.rl" + { hexdigit(parser, p); } + break; + case 6: +#line 1174 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_hex(parser)); } + break; + case 7: +#line 1180 "upb/json/parser.rl" + { CHECK_RETURN_TOP(escape(parser, p)); } + break; + case 8: +#line 1186 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; + case 9: +#line 1189 "upb/json/parser.rl" + { {stack[top++] = cs; cs = 19; goto _again;} } + break; + case 10: +#line 1191 "upb/json/parser.rl" + { p--; {stack[top++] = cs; cs = 27; goto _again;} } + break; + case 11: +#line 1196 "upb/json/parser.rl" + { start_member(parser); } + break; + case 12: +#line 1197 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_membername(parser)); } + break; + case 13: +#line 1200 "upb/json/parser.rl" + { end_member(parser); } + break; + case 14: +#line 1206 "upb/json/parser.rl" + { start_object(parser); } + break; + case 15: +#line 1209 "upb/json/parser.rl" + { end_object(parser); } + break; + case 16: +#line 1215 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_array(parser)); } + break; + case 17: +#line 1219 "upb/json/parser.rl" + { end_array(parser); } + break; + case 18: +#line 1224 "upb/json/parser.rl" + { start_number(parser, p); } + break; + case 19: +#line 1225 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_number(parser, p)); } + break; + case 20: +#line 1227 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_stringval(parser)); } + break; + case 21: +#line 1228 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_stringval(parser)); } + break; + case 22: +#line 1230 "upb/json/parser.rl" + { CHECK_RETURN_TOP(parser_putbool(parser, true)); } + break; + case 23: +#line 1232 "upb/json/parser.rl" + { CHECK_RETURN_TOP(parser_putbool(parser, false)); } + break; + case 24: +#line 1234 "upb/json/parser.rl" + { /* null value */ } + break; + case 25: +#line 1236 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_subobject(parser)); } + break; + case 26: +#line 1237 "upb/json/parser.rl" + { end_subobject(parser); } + break; + case 27: +#line 1242 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; +#line 1514 "upb/json/parser.c" + } + } + +_again: + if ( cs == 0 ) + goto _out; + if ( ++p != pe ) + goto _resume; + _test_eof: {} + _out: {} + } + +#line 1269 "upb/json/parser.rl" + + if (p != pe) { + upb_status_seterrf(&parser->status, "Parse error at '%.*s'\n", pe - p, p); + upb_env_reporterror(parser->env, &parser->status); + } else { + capture_suspend(parser, &p); + } + +error: + /* Save parsing state back to parser. */ + parser->current_state = cs; + parser->parser_top = top; + + return p - buf; +} + +bool end(void *closure, const void *hd) { + UPB_UNUSED(closure); + UPB_UNUSED(hd); + + /* Prevent compile warning on unused static constants. */ + UPB_UNUSED(json_start); + UPB_UNUSED(json_en_number_machine); + UPB_UNUSED(json_en_string_machine); + UPB_UNUSED(json_en_value_machine); + UPB_UNUSED(json_en_main); + return true; +} + +static void json_parser_reset(upb_json_parser *p) { + int cs; + int top; + + p->top = p->stack; + p->top->f = NULL; + p->top->is_map = false; + p->top->is_mapentry = false; + + /* Emit Ragel initialization of the parser. */ + +#line 1568 "upb/json/parser.c" + { + cs = json_start; + top = 0; + } + +#line 1309 "upb/json/parser.rl" + p->current_state = cs; + p->parser_top = top; + accumulate_clear(p); + p->multipart_state = MULTIPART_INACTIVE; + p->capture = NULL; + p->accumulated = NULL; + upb_status_clear(&p->status); +} + +static void visit_json_parsermethod(const upb_refcounted *r, + upb_refcounted_visit *visit, + void *closure) { + const upb_json_parsermethod *method = (upb_json_parsermethod*)r; + visit(r, upb_msgdef_upcast2(method->msg), closure); +} + +static void free_json_parsermethod(upb_refcounted *r) { + upb_json_parsermethod *method = (upb_json_parsermethod*)r; + + upb_inttable_iter i; + upb_inttable_begin(&i, &method->name_tables); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_value val = upb_inttable_iter_value(&i); + upb_strtable *t = upb_value_getptr(val); + upb_strtable_uninit(t); + upb_gfree(t); + } + + upb_inttable_uninit(&method->name_tables); + + upb_gfree(r); +} + +static void add_jsonname_table(upb_json_parsermethod *m, const upb_msgdef* md) { + upb_msg_field_iter i; + upb_strtable *t; + + /* It would be nice to stack-allocate this, but protobufs do not limit the + * length of fields to any reasonable limit. */ + char *buf = NULL; + size_t len = 0; + + if (upb_inttable_lookupptr(&m->name_tables, md, NULL)) { + return; + } + + /* TODO(haberman): handle malloc failure. */ + t = upb_gmalloc(sizeof(*t)); + upb_strtable_init(t, UPB_CTYPE_CONSTPTR); + upb_inttable_insertptr(&m->name_tables, md, upb_value_ptr(t)); + + for(upb_msg_field_begin(&i, md); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + + /* Add an entry for the JSON name. */ + size_t field_len = upb_fielddef_getjsonname(f, buf, len); + if (field_len > len) { + size_t len2; + buf = upb_grealloc(buf, 0, field_len); + len = field_len; + len2 = upb_fielddef_getjsonname(f, buf, len); + UPB_ASSERT_VAR(len2, len == len2); + } + upb_strtable_insert(t, buf, upb_value_constptr(f)); + + if (strcmp(buf, upb_fielddef_name(f)) != 0) { + /* Since the JSON name is different from the regular field name, add an + * entry for the raw name (compliant proto3 JSON parsers must accept + * both). */ + upb_strtable_insert(t, upb_fielddef_name(f), upb_value_constptr(f)); + } + + if (upb_fielddef_issubmsg(f)) { + add_jsonname_table(m, upb_fielddef_msgsubdef(f)); + } + } + + upb_gfree(buf); +} + +/* Public API *****************************************************************/ + +upb_json_parser *upb_json_parser_create(upb_env *env, + const upb_json_parsermethod *method, + upb_sink *output) { +#ifndef NDEBUG + const size_t size_before = upb_env_bytesallocated(env); +#endif + upb_json_parser *p = upb_env_malloc(env, sizeof(upb_json_parser)); + if (!p) return false; + + p->env = env; + p->method = method; + p->limit = p->stack + UPB_JSON_MAX_DEPTH; + p->accumulate_buf = NULL; + p->accumulate_buf_size = 0; + upb_bytessink_reset(&p->input_, &method->input_handler_, p); + + json_parser_reset(p); + upb_sink_reset(&p->top->sink, output->handlers, output->closure); + p->top->m = upb_handlers_msgdef(output->handlers); + set_name_table(p, p->top); + + /* If this fails, uncomment and increase the value in parser.h. */ + /* fprintf(stderr, "%zd\n", upb_env_bytesallocated(env) - size_before); */ + assert(upb_env_bytesallocated(env) - size_before <= UPB_JSON_PARSER_SIZE); + return p; +} + +upb_bytessink *upb_json_parser_input(upb_json_parser *p) { + return &p->input_; +} + +upb_json_parsermethod *upb_json_parsermethod_new(const upb_msgdef* md, + const void* owner) { + static const struct upb_refcounted_vtbl vtbl = {visit_json_parsermethod, + free_json_parsermethod}; + upb_json_parsermethod *ret = upb_gmalloc(sizeof(*ret)); + upb_refcounted_init(upb_json_parsermethod_upcast_mutable(ret), &vtbl, owner); + + ret->msg = md; + upb_ref2(md, ret); + + upb_byteshandler_init(&ret->input_handler_); + upb_byteshandler_setstring(&ret->input_handler_, parse, ret); + upb_byteshandler_setendstr(&ret->input_handler_, end, ret); + + upb_inttable_init(&ret->name_tables, UPB_CTYPE_PTR); + + add_jsonname_table(ret, md); + + return ret; +} + +const upb_byteshandler *upb_json_parsermethod_inputhandler( + const upb_json_parsermethod *m) { + return &m->input_handler_; +} +/* +** This currently uses snprintf() to format primitives, and could be optimized +** further. +*/ + + +#include <string.h> +#include <stdint.h> + +struct upb_json_printer { + upb_sink input_; + /* BytesSink closure. */ + void *subc_; + upb_bytessink *output_; + + /* We track the depth so that we know when to emit startstr/endstr on the + * output. */ + int depth_; + + /* Have we emitted the first element? This state is necessary to emit commas + * without leaving a trailing comma in arrays/maps. We keep this state per + * frame depth. + * + * Why max_depth * 2? UPB_MAX_HANDLER_DEPTH counts depth as nested messages. + * We count frames (contexts in which we separate elements by commas) as both + * repeated fields and messages (maps), and the worst case is a + * message->repeated field->submessage->repeated field->... nesting. */ + bool first_elem_[UPB_MAX_HANDLER_DEPTH * 2]; +}; + +/* StringPiece; a pointer plus a length. */ +typedef struct { + char *ptr; + size_t len; +} strpc; + +void freestrpc(void *ptr) { + strpc *pc = ptr; + upb_gfree(pc->ptr); + upb_gfree(pc); +} + +/* Convert fielddef name to JSON name and return as a string piece. */ +strpc *newstrpc(upb_handlers *h, const upb_fielddef *f, + bool preserve_fieldnames) { + /* TODO(haberman): handle malloc failure. */ + strpc *ret = upb_gmalloc(sizeof(*ret)); + if (preserve_fieldnames) { + ret->ptr = upb_gstrdup(upb_fielddef_name(f)); + ret->len = strlen(ret->ptr); + } else { + size_t len; + ret->len = upb_fielddef_getjsonname(f, NULL, 0); + ret->ptr = upb_gmalloc(ret->len); + len = upb_fielddef_getjsonname(f, ret->ptr, ret->len); + UPB_ASSERT_VAR(len, len == ret->len); + ret->len--; /* NULL */ + } + + upb_handlers_addcleanup(h, ret, freestrpc); + return ret; +} + +/* ------------ JSON string printing: values, maps, arrays ------------------ */ + +static void print_data( + upb_json_printer *p, const char *buf, unsigned int len) { + /* TODO: Will need to change if we support pushback from the sink. */ + size_t n = upb_bytessink_putbuf(p->output_, p->subc_, buf, len, NULL); + UPB_ASSERT_VAR(n, n == len); +} + +static void print_comma(upb_json_printer *p) { + if (!p->first_elem_[p->depth_]) { + print_data(p, ",", 1); + } + p->first_elem_[p->depth_] = false; +} + +/* Helpers that print properly formatted elements to the JSON output stream. */ + +/* Used for escaping control chars in strings. */ +static const char kControlCharLimit = 0x20; + +UPB_INLINE bool is_json_escaped(char c) { + /* See RFC 4627. */ + unsigned char uc = (unsigned char)c; + return uc < kControlCharLimit || uc == '"' || uc == '\\'; +} + +UPB_INLINE const char* json_nice_escape(char c) { + switch (c) { + case '"': return "\\\""; + case '\\': return "\\\\"; + case '\b': return "\\b"; + case '\f': return "\\f"; + case '\n': return "\\n"; + case '\r': return "\\r"; + case '\t': return "\\t"; + default: return NULL; + } +} + +/* Write a properly escaped string chunk. The surrounding quotes are *not* + * printed; this is so that the caller has the option of emitting the string + * content in chunks. */ +static void putstring(upb_json_printer *p, const char *buf, unsigned int len) { + const char* unescaped_run = NULL; + unsigned int i; + for (i = 0; i < len; i++) { + char c = buf[i]; + /* Handle escaping. */ + if (is_json_escaped(c)) { + /* Use a "nice" escape, like \n, if one exists for this character. */ + const char* escape = json_nice_escape(c); + /* If we don't have a specific 'nice' escape code, use a \uXXXX-style + * escape. */ + char escape_buf[8]; + if (!escape) { + unsigned char byte = (unsigned char)c; + _upb_snprintf(escape_buf, sizeof(escape_buf), "\\u%04x", (int)byte); + escape = escape_buf; + } + + /* N.B. that we assume that the input encoding is equal to the output + * encoding (both UTF-8 for now), so for chars >= 0x20 and != \, ", we + * can simply pass the bytes through. */ + + /* If there's a current run of unescaped chars, print that run first. */ + if (unescaped_run) { + print_data(p, unescaped_run, &buf[i] - unescaped_run); + unescaped_run = NULL; + } + /* Then print the escape code. */ + print_data(p, escape, strlen(escape)); + } else { + /* Add to the current unescaped run of characters. */ + if (unescaped_run == NULL) { + unescaped_run = &buf[i]; + } + } + } + + /* If the string ended in a run of unescaped characters, print that last run. */ + if (unescaped_run) { + print_data(p, unescaped_run, &buf[len] - unescaped_run); + } +} + +#define CHKLENGTH(x) if (!(x)) return -1; + +/* Helpers that format floating point values according to our custom formats. + * Right now we use %.8g and %.17g for float/double, respectively, to match + * proto2::util::JsonFormat's defaults. May want to change this later. */ + +static size_t fmt_double(double val, char* buf, size_t length) { + size_t n = _upb_snprintf(buf, length, "%.17g", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_float(float val, char* buf, size_t length) { + size_t n = _upb_snprintf(buf, length, "%.8g", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_bool(bool val, char* buf, size_t length) { + size_t n = _upb_snprintf(buf, length, "%s", (val ? "true" : "false")); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_int64(long val, char* buf, size_t length) { + size_t n = _upb_snprintf(buf, length, "%ld", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_uint64(unsigned long long val, char* buf, size_t length) { + size_t n = _upb_snprintf(buf, length, "%llu", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +/* Print a map key given a field name. Called by scalar field handlers and by + * startseq for repeated fields. */ +static bool putkey(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + const strpc *key = handler_data; + print_comma(p); + print_data(p, "\"", 1); + putstring(p, key->ptr, key->len); + print_data(p, "\":", 2); + return true; +} + +#define CHKFMT(val) if ((val) == (size_t)-1) return false; +#define CHK(val) if (!(val)) return false; + +#define TYPE_HANDLERS(type, fmt_func) \ + static bool put##type(void *closure, const void *handler_data, type val) { \ + upb_json_printer *p = closure; \ + char data[64]; \ + size_t length = fmt_func(val, data, sizeof(data)); \ + UPB_UNUSED(handler_data); \ + CHKFMT(length); \ + print_data(p, data, length); \ + return true; \ + } \ + static bool scalar_##type(void *closure, const void *handler_data, \ + type val) { \ + CHK(putkey(closure, handler_data)); \ + CHK(put##type(closure, handler_data, val)); \ + return true; \ + } \ + static bool repeated_##type(void *closure, const void *handler_data, \ + type val) { \ + upb_json_printer *p = closure; \ + print_comma(p); \ + CHK(put##type(closure, handler_data, val)); \ + return true; \ + } + +#define TYPE_HANDLERS_MAPKEY(type, fmt_func) \ + static bool putmapkey_##type(void *closure, const void *handler_data, \ + type val) { \ + upb_json_printer *p = closure; \ + print_data(p, "\"", 1); \ + CHK(put##type(closure, handler_data, val)); \ + print_data(p, "\":", 2); \ + return true; \ + } + +TYPE_HANDLERS(double, fmt_double) +TYPE_HANDLERS(float, fmt_float) +TYPE_HANDLERS(bool, fmt_bool) +TYPE_HANDLERS(int32_t, fmt_int64) +TYPE_HANDLERS(uint32_t, fmt_int64) +TYPE_HANDLERS(int64_t, fmt_int64) +TYPE_HANDLERS(uint64_t, fmt_uint64) + +/* double and float are not allowed to be map keys. */ +TYPE_HANDLERS_MAPKEY(bool, fmt_bool) +TYPE_HANDLERS_MAPKEY(int32_t, fmt_int64) +TYPE_HANDLERS_MAPKEY(uint32_t, fmt_int64) +TYPE_HANDLERS_MAPKEY(int64_t, fmt_int64) +TYPE_HANDLERS_MAPKEY(uint64_t, fmt_uint64) + +#undef TYPE_HANDLERS +#undef TYPE_HANDLERS_MAPKEY + +typedef struct { + void *keyname; + const upb_enumdef *enumdef; +} EnumHandlerData; + +static bool scalar_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + const char *symbolic_name; + + CHK(putkey(closure, hd->keyname)); + + symbolic_name = upb_enumdef_iton(hd->enumdef, val); + if (symbolic_name) { + print_data(p, "\"", 1); + putstring(p, symbolic_name, strlen(symbolic_name)); + print_data(p, "\"", 1); + } else { + putint32_t(closure, NULL, val); + } + + return true; +} + +static void print_enum_symbolic_name(upb_json_printer *p, + const upb_enumdef *def, + int32_t val) { + const char *symbolic_name = upb_enumdef_iton(def, val); + if (symbolic_name) { + print_data(p, "\"", 1); + putstring(p, symbolic_name, strlen(symbolic_name)); + print_data(p, "\"", 1); + } else { + putint32_t(p, NULL, val); + } +} + +static bool repeated_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + print_comma(p); + + print_enum_symbolic_name(p, hd->enumdef, val); + + return true; +} + +static bool mapvalue_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + + print_enum_symbolic_name(p, hd->enumdef, val); + + return true; +} + +static void *scalar_startsubmsg(void *closure, const void *handler_data) { + return putkey(closure, handler_data) ? closure : UPB_BREAK; +} + +static void *repeated_startsubmsg(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_comma(p); + return closure; +} + +static void start_frame(upb_json_printer *p) { + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "{", 1); +} + +static void end_frame(upb_json_printer *p) { + print_data(p, "}", 1); + p->depth_--; +} + +static bool printer_startmsg(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + if (p->depth_ == 0) { + upb_bytessink_start(p->output_, 0, &p->subc_); + } + start_frame(p); + return true; +} + +static bool printer_endmsg(void *closure, const void *handler_data, upb_status *s) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + UPB_UNUSED(s); + end_frame(p); + if (p->depth_ == 0) { + upb_bytessink_end(p->output_); + } + return true; +} + +static void *startseq(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + CHK(putkey(closure, handler_data)); + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "[", 1); + return closure; +} + +static bool endseq(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "]", 1); + p->depth_--; + return true; +} + +static void *startmap(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + CHK(putkey(closure, handler_data)); + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "{", 1); + return closure; +} + +static bool endmap(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "}", 1); + p->depth_--; + return true; +} + +static size_t putstr(void *closure, const void *handler_data, const char *str, + size_t len, const upb_bufhandle *handle) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + UPB_UNUSED(handle); + putstring(p, str, len); + return len; +} + +/* This has to Base64 encode the bytes, because JSON has no "bytes" type. */ +static size_t putbytes(void *closure, const void *handler_data, const char *str, + size_t len, const upb_bufhandle *handle) { + upb_json_printer *p = closure; + + /* This is the regular base64, not the "web-safe" version. */ + static const char base64[] = + "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; + + /* Base64-encode. */ + char data[16000]; + const char *limit = data + sizeof(data); + const unsigned char *from = (const unsigned char*)str; + char *to = data; + size_t remaining = len; + size_t bytes; + + UPB_UNUSED(handler_data); + UPB_UNUSED(handle); + + while (remaining > 2) { + /* TODO(haberman): handle encoded lengths > sizeof(data) */ + UPB_ASSERT_VAR(limit, (limit - to) >= 4); + + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)]; + to[2] = base64[((from[1] & 0xf) << 2) | (from[2] >> 6)]; + to[3] = base64[from[2] & 0x3f]; + + remaining -= 3; + to += 4; + from += 3; + } + + switch (remaining) { + case 2: + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)]; + to[2] = base64[(from[1] & 0xf) << 2]; + to[3] = '='; + to += 4; + from += 2; + break; + case 1: + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4)]; + to[2] = '='; + to[3] = '='; + to += 4; + from += 1; + break; + } + + bytes = to - data; + print_data(p, "\"", 1); + putstring(p, data, bytes); + print_data(p, "\"", 1); + return len; +} + +static void *scalar_startstr(void *closure, const void *handler_data, + size_t size_hint) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + CHK(putkey(closure, handler_data)); + print_data(p, "\"", 1); + return p; +} + +static size_t scalar_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool scalar_endstr(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "\"", 1); + return true; +} + +static void *repeated_startstr(void *closure, const void *handler_data, + size_t size_hint) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + print_comma(p); + print_data(p, "\"", 1); + return p; +} + +static size_t repeated_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool repeated_endstr(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "\"", 1); + return true; +} + +static void *mapkeyval_startstr(void *closure, const void *handler_data, + size_t size_hint) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + print_data(p, "\"", 1); + return p; +} + +static size_t mapkey_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool mapkey_endstr(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "\":", 2); + return true; +} + +static bool mapvalue_endstr(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + UPB_UNUSED(handler_data); + print_data(p, "\"", 1); + return true; +} + +static size_t scalar_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putkey(closure, handler_data)); + CHK(putbytes(closure, handler_data, str, len, handle)); + return len; +} + +static size_t repeated_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + upb_json_printer *p = closure; + print_comma(p); + CHK(putbytes(closure, handler_data, str, len, handle)); + return len; +} + +static size_t mapkey_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + upb_json_printer *p = closure; + CHK(putbytes(closure, handler_data, str, len, handle)); + print_data(p, ":", 1); + return len; +} + +static void set_enum_hd(upb_handlers *h, + const upb_fielddef *f, + bool preserve_fieldnames, + upb_handlerattr *attr) { + EnumHandlerData *hd = upb_gmalloc(sizeof(EnumHandlerData)); + hd->enumdef = (const upb_enumdef *)upb_fielddef_subdef(f); + hd->keyname = newstrpc(h, f, preserve_fieldnames); + upb_handlers_addcleanup(h, hd, upb_gfree); + upb_handlerattr_sethandlerdata(attr, hd); +} + +/* Set up handlers for a mapentry submessage (i.e., an individual key/value pair + * in a map). + * + * TODO: Handle missing key, missing value, out-of-order key/value, or repeated + * key or value cases properly. The right way to do this is to allocate a + * temporary structure at the start of a mapentry submessage, store key and + * value data in it as key and value handlers are called, and then print the + * key/value pair once at the end of the submessage. If we don't do this, we + * should at least detect the case and throw an error. However, so far all of + * our sources that emit mapentry messages do so canonically (with one key + * field, and then one value field), so this is not a pressing concern at the + * moment. */ +void printer_sethandlers_mapentry(const void *closure, bool preserve_fieldnames, + upb_handlers *h) { + const upb_msgdef *md = upb_handlers_msgdef(h); + + /* A mapentry message is printed simply as '"key": value'. Rather than + * special-case key and value for every type below, we just handle both + * fields explicitly here. */ + const upb_fielddef* key_field = upb_msgdef_itof(md, UPB_MAPENTRY_KEY); + const upb_fielddef* value_field = upb_msgdef_itof(md, UPB_MAPENTRY_VALUE); + + upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER; + + UPB_UNUSED(closure); + + switch (upb_fielddef_type(key_field)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, key_field, putmapkey_int32_t, &empty_attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, key_field, putmapkey_int64_t, &empty_attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, key_field, putmapkey_uint32_t, &empty_attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, key_field, putmapkey_uint64_t, &empty_attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, key_field, putmapkey_bool, &empty_attr); + break; + case UPB_TYPE_STRING: + upb_handlers_setstartstr(h, key_field, mapkeyval_startstr, &empty_attr); + upb_handlers_setstring(h, key_field, mapkey_str, &empty_attr); + upb_handlers_setendstr(h, key_field, mapkey_endstr, &empty_attr); + break; + case UPB_TYPE_BYTES: + upb_handlers_setstring(h, key_field, mapkey_bytes, &empty_attr); + break; + default: + assert(false); + break; + } + + switch (upb_fielddef_type(value_field)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, value_field, putint32_t, &empty_attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, value_field, putint64_t, &empty_attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, value_field, putuint32_t, &empty_attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, value_field, putuint64_t, &empty_attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, value_field, putbool, &empty_attr); + break; + case UPB_TYPE_FLOAT: + upb_handlers_setfloat(h, value_field, putfloat, &empty_attr); + break; + case UPB_TYPE_DOUBLE: + upb_handlers_setdouble(h, value_field, putdouble, &empty_attr); + break; + case UPB_TYPE_STRING: + upb_handlers_setstartstr(h, value_field, mapkeyval_startstr, &empty_attr); + upb_handlers_setstring(h, value_field, putstr, &empty_attr); + upb_handlers_setendstr(h, value_field, mapvalue_endstr, &empty_attr); + break; + case UPB_TYPE_BYTES: + upb_handlers_setstring(h, value_field, putbytes, &empty_attr); + break; + case UPB_TYPE_ENUM: { + upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER; + set_enum_hd(h, value_field, preserve_fieldnames, &enum_attr); + upb_handlers_setint32(h, value_field, mapvalue_enum, &enum_attr); + upb_handlerattr_uninit(&enum_attr); + break; + } + case UPB_TYPE_MESSAGE: + /* No handler necessary -- the submsg handlers will print the message + * as appropriate. */ + break; + } + + upb_handlerattr_uninit(&empty_attr); +} + +void printer_sethandlers(const void *closure, upb_handlers *h) { + const upb_msgdef *md = upb_handlers_msgdef(h); + bool is_mapentry = upb_msgdef_mapentry(md); + upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER; + upb_msg_field_iter i; + const bool *preserve_fieldnames_ptr = closure; + const bool preserve_fieldnames = *preserve_fieldnames_ptr; + + if (is_mapentry) { + /* mapentry messages are sufficiently different that we handle them + * separately. */ + printer_sethandlers_mapentry(closure, preserve_fieldnames, h); + return; + } + + upb_handlers_setstartmsg(h, printer_startmsg, &empty_attr); + upb_handlers_setendmsg(h, printer_endmsg, &empty_attr); + +#define TYPE(type, name, ctype) \ + case type: \ + if (upb_fielddef_isseq(f)) { \ + upb_handlers_set##name(h, f, repeated_##ctype, &empty_attr); \ + } else { \ + upb_handlers_set##name(h, f, scalar_##ctype, &name_attr); \ + } \ + break; + + upb_msg_field_begin(&i, md); + for(; !upb_msg_field_done(&i); upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + + upb_handlerattr name_attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&name_attr, + newstrpc(h, f, preserve_fieldnames)); + + if (upb_fielddef_ismap(f)) { + upb_handlers_setstartseq(h, f, startmap, &name_attr); + upb_handlers_setendseq(h, f, endmap, &name_attr); + } else if (upb_fielddef_isseq(f)) { + upb_handlers_setstartseq(h, f, startseq, &name_attr); + upb_handlers_setendseq(h, f, endseq, &empty_attr); + } + + switch (upb_fielddef_type(f)) { + TYPE(UPB_TYPE_FLOAT, float, float); + TYPE(UPB_TYPE_DOUBLE, double, double); + TYPE(UPB_TYPE_BOOL, bool, bool); + TYPE(UPB_TYPE_INT32, int32, int32_t); + TYPE(UPB_TYPE_UINT32, uint32, uint32_t); + TYPE(UPB_TYPE_INT64, int64, int64_t); + TYPE(UPB_TYPE_UINT64, uint64, uint64_t); + case UPB_TYPE_ENUM: { + /* For now, we always emit symbolic names for enums. We may want an + * option later to control this behavior, but we will wait for a real + * need first. */ + upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER; + set_enum_hd(h, f, preserve_fieldnames, &enum_attr); + + if (upb_fielddef_isseq(f)) { + upb_handlers_setint32(h, f, repeated_enum, &enum_attr); + } else { + upb_handlers_setint32(h, f, scalar_enum, &enum_attr); + } + + upb_handlerattr_uninit(&enum_attr); + break; + } + case UPB_TYPE_STRING: + if (upb_fielddef_isseq(f)) { + upb_handlers_setstartstr(h, f, repeated_startstr, &empty_attr); + upb_handlers_setstring(h, f, repeated_str, &empty_attr); + upb_handlers_setendstr(h, f, repeated_endstr, &empty_attr); + } else { + upb_handlers_setstartstr(h, f, scalar_startstr, &name_attr); + upb_handlers_setstring(h, f, scalar_str, &empty_attr); + upb_handlers_setendstr(h, f, scalar_endstr, &empty_attr); + } + break; + case UPB_TYPE_BYTES: + /* XXX: this doesn't support strings that span buffers yet. The base64 + * encoder will need to be made resumable for this to work properly. */ + if (upb_fielddef_isseq(f)) { + upb_handlers_setstring(h, f, repeated_bytes, &empty_attr); + } else { + upb_handlers_setstring(h, f, scalar_bytes, &name_attr); + } + break; + case UPB_TYPE_MESSAGE: + if (upb_fielddef_isseq(f)) { + upb_handlers_setstartsubmsg(h, f, repeated_startsubmsg, &name_attr); + } else { + upb_handlers_setstartsubmsg(h, f, scalar_startsubmsg, &name_attr); + } + break; + } + + upb_handlerattr_uninit(&name_attr); + } + + upb_handlerattr_uninit(&empty_attr); +#undef TYPE +} + +static void json_printer_reset(upb_json_printer *p) { + p->depth_ = 0; +} + + +/* Public API *****************************************************************/ + +upb_json_printer *upb_json_printer_create(upb_env *e, const upb_handlers *h, + upb_bytessink *output) { +#ifndef NDEBUG + size_t size_before = upb_env_bytesallocated(e); +#endif + + upb_json_printer *p = upb_env_malloc(e, sizeof(upb_json_printer)); + if (!p) return NULL; + + p->output_ = output; + json_printer_reset(p); + upb_sink_reset(&p->input_, h, p); + + /* If this fails, increase the value in printer.h. */ + assert(upb_env_bytesallocated(e) - size_before <= UPB_JSON_PRINTER_SIZE); + return p; +} + +upb_sink *upb_json_printer_input(upb_json_printer *p) { + return &p->input_; +} + +const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md, + bool preserve_fieldnames, + const void *owner) { + return upb_handlers_newfrozen( + md, owner, printer_sethandlers, &preserve_fieldnames); +} diff --git a/ruby/ext/google/protobuf_c/upb.h b/ruby/ext/google/protobuf_c/upb.h new file mode 100644 index 00000000..e8b683bd --- /dev/null +++ b/ruby/ext/google/protobuf_c/upb.h @@ -0,0 +1,8569 @@ +// Amalgamated source file +/* +** Defs are upb's internal representation of the constructs that can appear +** in a .proto file: +** +** - upb::MessageDef (upb_msgdef): describes a "message" construct. +** - upb::FieldDef (upb_fielddef): describes a message field. +** - upb::FileDef (upb_filedef): describes a .proto file and its defs. +** - upb::EnumDef (upb_enumdef): describes an enum. +** - upb::OneofDef (upb_oneofdef): describes a oneof. +** - upb::Def (upb_def): base class of all the others. +** +** TODO: definitions of services. +** +** Like upb_refcounted objects, defs are mutable only until frozen, and are +** only thread-safe once frozen. +** +** This is a mixed C/C++ interface that offers a full API to both languages. +** See the top-level README for more information. +*/ + +#ifndef UPB_DEF_H_ +#define UPB_DEF_H_ + +/* +** upb::RefCounted (upb_refcounted) +** +** A refcounting scheme that supports circular refs. It accomplishes this by +** partitioning the set of objects into groups such that no cycle spans groups; +** we can then reference-count the group as a whole and ignore refs within the +** group. When objects are mutable, these groups are computed very +** conservatively; we group any objects that have ever had a link between them. +** When objects are frozen, we compute strongly-connected components which +** allows us to be precise and only group objects that are actually cyclic. +** +** This is a mixed C/C++ interface that offers a full API to both languages. +** See the top-level README for more information. +*/ + +#ifndef UPB_REFCOUNTED_H_ +#define UPB_REFCOUNTED_H_ + +/* +** upb_table +** +** This header is INTERNAL-ONLY! Its interfaces are not public or stable! +** This file defines very fast int->upb_value (inttable) and string->upb_value +** (strtable) hash tables. +** +** The table uses chained scatter with Brent's variation (inspired by the Lua +** implementation of hash tables). The hash function for strings is Austin +** Appleby's "MurmurHash." +** +** The inttable uses uintptr_t as its key, which guarantees it can be used to +** store pointers or integers of at least 32 bits (upb isn't really useful on +** systems where sizeof(void*) < 4). +** +** The table must be homogeneous (all values of the same type). In debug +** mode, we check this on insert and lookup. +*/ + +#ifndef UPB_TABLE_H_ +#define UPB_TABLE_H_ + +#include <assert.h> +#include <stdint.h> +#include <string.h> +/* +** This file contains shared definitions that are widely used across upb. +** +** This is a mixed C/C++ interface that offers a full API to both languages. +** See the top-level README for more information. +*/ + +#ifndef UPB_H_ +#define UPB_H_ + +#include <assert.h> +#include <stdarg.h> +#include <stdbool.h> +#include <stddef.h> + +#ifdef __cplusplus +namespace upb { +class Allocator; +class Arena; +class Environment; +class ErrorSpace; +class Status; +template <int N> class InlinedArena; +template <int N> class InlinedEnvironment; +} +#endif + +/* UPB_INLINE: inline if possible, emit standalone code if required. */ +#ifdef __cplusplus +#define UPB_INLINE inline +#elif defined (__GNUC__) +#define UPB_INLINE static __inline__ +#else +#define UPB_INLINE static +#endif + +/* Define UPB_BIG_ENDIAN manually if you're on big endian and your compiler + * doesn't provide these preprocessor symbols. */ +#if defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) +#define UPB_BIG_ENDIAN +#endif + +/* Macros for function attributes on compilers that support them. */ +#ifdef __GNUC__ +#define UPB_FORCEINLINE __inline__ __attribute__((always_inline)) +#define UPB_NOINLINE __attribute__((noinline)) +#define UPB_NORETURN __attribute__((__noreturn__)) +#else /* !defined(__GNUC__) */ +#define UPB_FORCEINLINE +#define UPB_NOINLINE +#define UPB_NORETURN +#endif + +/* A few hacky workarounds for functions not in C89. + * For internal use only! + * TODO(haberman): fix these by including our own implementations, or finding + * another workaround. + */ +#ifdef __GNUC__ +#define _upb_snprintf __builtin_snprintf +#define _upb_vsnprintf __builtin_vsnprintf +#define _upb_va_copy(a, b) __va_copy(a, b) +#elif __STDC_VERSION__ >= 199901L +/* C99 versions. */ +#define _upb_snprintf snprintf +#define _upb_vsnprintf vsnprintf +#define _upb_va_copy(a, b) va_copy(a, b) +#else +#error Need implementations of [v]snprintf and va_copy +#endif + + +#if ((defined(__cplusplus) && __cplusplus >= 201103L) || \ + defined(__GXX_EXPERIMENTAL_CXX0X__)) && !defined(UPB_NO_CXX11) +#define UPB_CXX11 +#endif + +/* UPB_DISALLOW_COPY_AND_ASSIGN() + * UPB_DISALLOW_POD_OPS() + * + * Declare these in the "private" section of a C++ class to forbid copy/assign + * or all POD ops (construct, destruct, copy, assign) on that class. */ +#ifdef UPB_CXX11 +#include <type_traits> +#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \ + class_name(const class_name&) = delete; \ + void operator=(const class_name&) = delete; +#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \ + class_name() = delete; \ + ~class_name() = delete; \ + UPB_DISALLOW_COPY_AND_ASSIGN(class_name) +#define UPB_ASSERT_STDLAYOUT(type) \ + static_assert(std::is_standard_layout<type>::value, \ + #type " must be standard layout"); +#define UPB_FINAL final +#else /* !defined(UPB_CXX11) */ +#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \ + class_name(const class_name&); \ + void operator=(const class_name&); +#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \ + class_name(); \ + ~class_name(); \ + UPB_DISALLOW_COPY_AND_ASSIGN(class_name) +#define UPB_ASSERT_STDLAYOUT(type) +#define UPB_FINAL +#endif + +/* UPB_DECLARE_TYPE() + * UPB_DECLARE_DERIVED_TYPE() + * UPB_DECLARE_DERIVED_TYPE2() + * + * Macros for declaring C and C++ types both, including inheritance. + * The inheritance doesn't use real C++ inheritance, to stay compatible with C. + * + * These macros also provide upcasts: + * - in C: types-specific functions (ie. upb_foo_upcast(foo)) + * - in C++: upb::upcast(foo) along with implicit conversions + * + * Downcasts are not provided, but upb/def.h defines downcasts for upb::Def. */ + +#define UPB_C_UPCASTS(ty, base) \ + UPB_INLINE base *ty ## _upcast_mutable(ty *p) { return (base*)p; } \ + UPB_INLINE const base *ty ## _upcast(const ty *p) { return (const base*)p; } + +#define UPB_C_UPCASTS2(ty, base, base2) \ + UPB_C_UPCASTS(ty, base) \ + UPB_INLINE base2 *ty ## _upcast2_mutable(ty *p) { return (base2*)p; } \ + UPB_INLINE const base2 *ty ## _upcast2(const ty *p) { return (const base2*)p; } + +#ifdef __cplusplus + +#define UPB_BEGIN_EXTERN_C extern "C" { +#define UPB_END_EXTERN_C } +#define UPB_PRIVATE_FOR_CPP private: +#define UPB_DECLARE_TYPE(cppname, cname) typedef cppname cname; + +#define UPB_DECLARE_DERIVED_TYPE(cppname, cppbase, cname, cbase) \ + UPB_DECLARE_TYPE(cppname, cname) \ + UPB_C_UPCASTS(cname, cbase) \ + namespace upb { \ + template <> \ + class Pointer<cppname> : public PointerBase<cppname, cppbase> { \ + public: \ + explicit Pointer(cppname* ptr) \ + : PointerBase<cppname, cppbase>(ptr) {} \ + }; \ + template <> \ + class Pointer<const cppname> \ + : public PointerBase<const cppname, const cppbase> { \ + public: \ + explicit Pointer(const cppname* ptr) \ + : PointerBase<const cppname, const cppbase>(ptr) {} \ + }; \ + } + +#define UPB_DECLARE_DERIVED_TYPE2(cppname, cppbase, cppbase2, cname, cbase, \ + cbase2) \ + UPB_DECLARE_TYPE(cppname, cname) \ + UPB_C_UPCASTS2(cname, cbase, cbase2) \ + namespace upb { \ + template <> \ + class Pointer<cppname> : public PointerBase2<cppname, cppbase, cppbase2> { \ + public: \ + explicit Pointer(cppname* ptr) \ + : PointerBase2<cppname, cppbase, cppbase2>(ptr) {} \ + }; \ + template <> \ + class Pointer<const cppname> \ + : public PointerBase2<const cppname, const cppbase, const cppbase2> { \ + public: \ + explicit Pointer(const cppname* ptr) \ + : PointerBase2<const cppname, const cppbase, const cppbase2>(ptr) {} \ + }; \ + } + +#else /* !defined(__cplusplus) */ + +#define UPB_BEGIN_EXTERN_C +#define UPB_END_EXTERN_C +#define UPB_PRIVATE_FOR_CPP +#define UPB_DECLARE_TYPE(cppname, cname) \ + struct cname; \ + typedef struct cname cname; +#define UPB_DECLARE_DERIVED_TYPE(cppname, cppbase, cname, cbase) \ + UPB_DECLARE_TYPE(cppname, cname) \ + UPB_C_UPCASTS(cname, cbase) +#define UPB_DECLARE_DERIVED_TYPE2(cppname, cppbase, cppbase2, \ + cname, cbase, cbase2) \ + UPB_DECLARE_TYPE(cppname, cname) \ + UPB_C_UPCASTS2(cname, cbase, cbase2) + +#endif /* defined(__cplusplus) */ + +#define UPB_MAX(x, y) ((x) > (y) ? (x) : (y)) +#define UPB_MIN(x, y) ((x) < (y) ? (x) : (y)) + +#define UPB_UNUSED(var) (void)var + +/* For asserting something about a variable when the variable is not used for + * anything else. This prevents "unused variable" warnings when compiling in + * debug mode. */ +#define UPB_ASSERT_VAR(var, predicate) UPB_UNUSED(var); assert(predicate) + +/* Generic function type. */ +typedef void upb_func(); + + +/* C++ Casts ******************************************************************/ + +#ifdef __cplusplus + +namespace upb { + +template <class T> class Pointer; + +/* Casts to a subclass. The caller must know that cast is correct; an + * incorrect cast will throw an assertion failure in debug mode. + * + * Example: + * upb::Def* def = GetDef(); + * // Assert-fails if this was not actually a MessageDef. + * upb::MessgeDef* md = upb::down_cast<upb::MessageDef>(def); + * + * Note that downcasts are only defined for some types (at the moment you can + * only downcast from a upb::Def to a specific Def type). */ +template<class To, class From> To down_cast(From* f); + +/* Casts to a subclass. If the class does not actually match the given To type, + * returns NULL. + * + * Example: + * upb::Def* def = GetDef(); + * // md will be NULL if this was not actually a MessageDef. + * upb::MessgeDef* md = upb::down_cast<upb::MessageDef>(def); + * + * Note that dynamic casts are only defined for some types (at the moment you + * can only downcast from a upb::Def to a specific Def type).. */ +template<class To, class From> To dyn_cast(From* f); + +/* Casts to any base class, or the type itself (ie. can be a no-op). + * + * Example: + * upb::MessageDef* md = GetDef(); + * // This will fail to compile if this wasn't actually a base class. + * upb::Def* def = upb::upcast(md); + */ +template <class T> inline Pointer<T> upcast(T *f) { return Pointer<T>(f); } + +/* Attempt upcast to specific base class. + * + * Example: + * upb::MessageDef* md = GetDef(); + * upb::upcast_to<upb::Def>(md)->MethodOnDef(); + */ +template <class T, class F> inline T* upcast_to(F *f) { + return static_cast<T*>(upcast(f)); +} + +/* PointerBase<T>: implementation detail of upb::upcast(). + * It is implicitly convertable to pointers to the Base class(es). + */ +template <class T, class Base> +class PointerBase { + public: + explicit PointerBase(T* ptr) : ptr_(ptr) {} + operator T*() { return ptr_; } + operator Base*() { return (Base*)ptr_; } + + private: + T* ptr_; +}; + +template <class T, class Base, class Base2> +class PointerBase2 : public PointerBase<T, Base> { + public: + explicit PointerBase2(T* ptr) : PointerBase<T, Base>(ptr) {} + operator Base2*() { return Pointer<Base>(*this); } +}; + +} + +#endif + + +/* upb::ErrorSpace ************************************************************/ + +/* A upb::ErrorSpace represents some domain of possible error values. This lets + * upb::Status attach specific error codes to operations, like POSIX/C errno, + * Win32 error codes, etc. Clients who want to know the very specific error + * code can check the error space and then know the type of the integer code. + * + * NOTE: upb::ErrorSpace is currently not used and should be considered + * experimental. It is important primarily in cases where upb is performing + * I/O, but upb doesn't currently have any components that do this. */ + +UPB_DECLARE_TYPE(upb::ErrorSpace, upb_errorspace) + +#ifdef __cplusplus +class upb::ErrorSpace { +#else +struct upb_errorspace { +#endif + const char *name; +}; + + +/* upb::Status ****************************************************************/ + +/* upb::Status represents a success or failure status and error message. + * It owns no resources and allocates no memory, so it should work + * even in OOM situations. */ +UPB_DECLARE_TYPE(upb::Status, upb_status) + +/* The maximum length of an error message before it will get truncated. */ +#define UPB_STATUS_MAX_MESSAGE 128 + +UPB_BEGIN_EXTERN_C + +const char *upb_status_errmsg(const upb_status *status); +bool upb_ok(const upb_status *status); +upb_errorspace *upb_status_errspace(const upb_status *status); +int upb_status_errcode(const upb_status *status); + +/* Any of the functions that write to a status object allow status to be NULL, + * to support use cases where the function's caller does not care about the + * status message. */ +void upb_status_clear(upb_status *status); +void upb_status_seterrmsg(upb_status *status, const char *msg); +void upb_status_seterrf(upb_status *status, const char *fmt, ...); +void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args); +void upb_status_copy(upb_status *to, const upb_status *from); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +class upb::Status { + public: + Status() { upb_status_clear(this); } + + /* Returns true if there is no error. */ + bool ok() const { return upb_ok(this); } + + /* Optional error space and code, useful if the caller wants to + * programmatically check the specific kind of error. */ + ErrorSpace* error_space() { return upb_status_errspace(this); } + int error_code() const { return upb_status_errcode(this); } + + /* The returned string is invalidated by any other call into the status. */ + const char *error_message() const { return upb_status_errmsg(this); } + + /* The error message will be truncated if it is longer than + * UPB_STATUS_MAX_MESSAGE-4. */ + void SetErrorMessage(const char* msg) { upb_status_seterrmsg(this, msg); } + void SetFormattedErrorMessage(const char* fmt, ...) { + va_list args; + va_start(args, fmt); + upb_status_vseterrf(this, fmt, args); + va_end(args); + } + + /* Resets the status to a successful state with no message. */ + void Clear() { upb_status_clear(this); } + + void CopyFrom(const Status& other) { upb_status_copy(this, &other); } + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Status) +#else +struct upb_status { +#endif + bool ok_; + + /* Specific status code defined by some error space (optional). */ + int code_; + upb_errorspace *error_space_; + + /* TODO(haberman): add file/line of error? */ + + /* Error message; NULL-terminated. */ + char msg[UPB_STATUS_MAX_MESSAGE]; +}; + +#define UPB_STATUS_INIT {true, 0, NULL, {0}} + + +/** Built-in error spaces. ****************************************************/ + +/* Errors raised by upb that we want to be able to detect programmatically. */ +typedef enum { + UPB_NOMEM /* Can't reuse ENOMEM because it is POSIX, not ISO C. */ +} upb_errcode_t; + +extern upb_errorspace upb_upberr; + +void upb_upberr_setoom(upb_status *s); + +/* Since errno is defined by standard C, we define an error space for it in + * core upb. Other error spaces should be defined in other, platform-specific + * modules. */ + +extern upb_errorspace upb_errnoerr; + + +/** upb::Allocator ************************************************************/ + +/* A upb::Allocator is a possibly-stateful allocator object. + * + * It could either be an arena allocator (which doesn't require individual + * free() calls) or a regular malloc() (which does). The client must therefore + * free memory unless it knows that the allocator is an arena allocator. */ +UPB_DECLARE_TYPE(upb::Allocator, upb_alloc) + +/* A malloc()/free() function. + * If "size" is 0 then the function acts like free(), otherwise it acts like + * realloc(). Only "oldsize" bytes from a previous allocation are preserved. */ +typedef void *upb_alloc_func(upb_alloc *alloc, void *ptr, size_t oldsize, + size_t size); + +#ifdef __cplusplus + +class upb::Allocator UPB_FINAL { + public: + Allocator() {} + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Allocator) + + public: +#else +struct upb_alloc { +#endif /* __cplusplus */ + upb_alloc_func *func; +}; + +UPB_INLINE void *upb_malloc(upb_alloc *alloc, size_t size) { + assert(size > 0); + return alloc->func(alloc, NULL, 0, size); +} + +UPB_INLINE void *upb_realloc(upb_alloc *alloc, void *ptr, size_t oldsize, + size_t size) { + assert(size > 0); + return alloc->func(alloc, ptr, oldsize, size); +} + +UPB_INLINE void upb_free(upb_alloc *alloc, void *ptr) { + alloc->func(alloc, ptr, 0, 0); +} + +/* The global allocator used by upb. Uses the standard malloc()/free(). */ + +extern upb_alloc upb_alloc_global; + +/* Functions that hard-code the global malloc. + * + * We still get benefit because we can put custom logic into our global + * allocator, like injecting out-of-memory faults in debug/testing builds. */ + +UPB_INLINE void *upb_gmalloc(size_t size) { + return upb_malloc(&upb_alloc_global, size); +} + +UPB_INLINE void *upb_grealloc(void *ptr, size_t oldsize, size_t size) { + return upb_realloc(&upb_alloc_global, ptr, oldsize, size); +} + +UPB_INLINE void upb_gfree(void *ptr) { + upb_free(&upb_alloc_global, ptr); +} + +/* upb::Arena *****************************************************************/ + +/* upb::Arena is a specific allocator implementation that uses arena allocation. + * The user provides an allocator that will be used to allocate the underlying + * arena blocks. Arenas by nature do not require the individual allocations + * to be freed. However the Arena does allow users to register cleanup + * functions that will run when the arena is destroyed. + * + * A upb::Arena is *not* thread-safe. + * + * You could write a thread-safe arena allocator that satisfies the + * upb::Allocator interface, but it would not be as efficient for the + * single-threaded case. */ +UPB_DECLARE_TYPE(upb::Arena, upb_arena) + +typedef void upb_cleanup_func(void *ud); + +#define UPB_ARENA_BLOCK_OVERHEAD (sizeof(size_t)*4) + +UPB_BEGIN_EXTERN_C + +void upb_arena_init(upb_arena *a); +void upb_arena_init2(upb_arena *a, void *mem, size_t n, upb_alloc *alloc); +void upb_arena_uninit(upb_arena *a); +upb_alloc *upb_arena_alloc(upb_arena *a); +bool upb_arena_addcleanup(upb_arena *a, upb_cleanup_func *func, void *ud); +size_t upb_arena_bytesallocated(const upb_arena *a); +void upb_arena_setnextblocksize(upb_arena *a, size_t size); +void upb_arena_setmaxblocksize(upb_arena *a, size_t size); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +class upb::Arena { + public: + /* A simple arena with no initial memory block and the default allocator. */ + Arena() { upb_arena_init(this); } + + /* Constructs an arena with the given initial block which allocates blocks + * with the given allocator. The given allocator must outlive the Arena. + * + * If you pass NULL for the allocator it will default to the global allocator + * upb_alloc_global, and NULL/0 for the initial block will cause there to be + * no initial block. */ + Arena(void *mem, size_t len, Allocator* a) { + upb_arena_init2(this, mem, len, a); + } + + ~Arena() { upb_arena_uninit(this); } + + /* Sets the size of the next block the Arena will request (unless the + * requested allocation is larger). Each block will double in size until the + * max limit is reached. */ + void SetNextBlockSize(size_t size) { upb_arena_setnextblocksize(this, size); } + + /* Sets the maximum block size. No blocks larger than this will be requested + * from the underlying allocator unless individual arena allocations are + * larger. */ + void SetMaxBlockSize(size_t size) { upb_arena_setmaxblocksize(this, size); } + + /* Allows this arena to be used as a generic allocator. + * + * The arena does not need free() calls so when using Arena as an allocator + * it is safe to skip them. However they are no-ops so there is no harm in + * calling free() either. */ + Allocator* allocator() { return upb_arena_alloc(this); } + + /* Add a cleanup function to run when the arena is destroyed. + * Returns false on out-of-memory. */ + bool AddCleanup(upb_cleanup_func* func, void* ud) { + return upb_arena_addcleanup(this, func, ud); + } + + /* Total number of bytes that have been allocated. It is undefined what + * Realloc() does to this counter. */ + size_t BytesAllocated() const { + return upb_arena_bytesallocated(this); + } + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Arena) + +#else +struct upb_arena { +#endif /* __cplusplus */ + /* We implement the allocator interface. + * This must be the first member of upb_arena! */ + upb_alloc alloc; + + /* Allocator to allocate arena blocks. We are responsible for freeing these + * when we are destroyed. */ + upb_alloc *block_alloc; + + size_t bytes_allocated; + size_t next_block_size; + size_t max_block_size; + + /* Linked list of blocks. Points to an arena_block, defined in env.c */ + void *block_head; + + /* Cleanup entries. Pointer to a cleanup_ent, defined in env.c */ + void *cleanup_head; + + /* For future expansion, since the size of this struct is exposed to users. */ + void *future1; + void *future2; +}; + + +/* upb::Environment ***********************************************************/ + +/* A upb::Environment provides a means for injecting malloc and an + * error-reporting callback into encoders/decoders. This allows them to be + * independent of nearly all assumptions about their actual environment. + * + * It is also a container for allocating the encoders/decoders themselves that + * insulates clients from knowing their actual size. This provides ABI + * compatibility even if the size of the objects change. And this allows the + * structure definitions to be in the .c files instead of the .h files, making + * the .h files smaller and more readable. + * + * We might want to consider renaming this to "Pipeline" if/when the concept of + * a pipeline element becomes more formalized. */ +UPB_DECLARE_TYPE(upb::Environment, upb_env) + +/* A function that receives an error report from an encoder or decoder. The + * callback can return true to request that the error should be recovered, but + * if the error is not recoverable this has no effect. */ +typedef bool upb_error_func(void *ud, const upb_status *status); + +UPB_BEGIN_EXTERN_C + +void upb_env_init(upb_env *e); +void upb_env_init2(upb_env *e, void *mem, size_t n, upb_alloc *alloc); +void upb_env_uninit(upb_env *e); + +void upb_env_initonly(upb_env *e); + +upb_arena *upb_env_arena(upb_env *e); +bool upb_env_ok(const upb_env *e); +void upb_env_seterrorfunc(upb_env *e, upb_error_func *func, void *ud); + +/* Convenience wrappers around the methods of the contained arena. */ +void upb_env_reporterrorsto(upb_env *e, upb_status *s); +bool upb_env_reporterror(upb_env *e, const upb_status *s); +void *upb_env_malloc(upb_env *e, size_t size); +void *upb_env_realloc(upb_env *e, void *ptr, size_t oldsize, size_t size); +void upb_env_free(upb_env *e, void *ptr); +bool upb_env_addcleanup(upb_env *e, upb_cleanup_func *func, void *ud); +size_t upb_env_bytesallocated(const upb_env *e); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +class upb::Environment { + public: + /* The given Arena must outlive this environment. */ + Environment() { upb_env_initonly(this); } + + Environment(void *mem, size_t len, Allocator *a) : arena_(mem, len, a) { + upb_env_initonly(this); + } + + Arena* arena() { return upb_env_arena(this); } + + /* Set a custom error reporting function. */ + void SetErrorFunction(upb_error_func* func, void* ud) { + upb_env_seterrorfunc(this, func, ud); + } + + /* Set the error reporting function to simply copy the status to the given + * status and abort. */ + void ReportErrorsTo(Status* status) { upb_env_reporterrorsto(this, status); } + + /* Returns true if all allocations and AddCleanup() calls have succeeded, + * and no errors were reported with ReportError() (except ones that recovered + * successfully). */ + bool ok() const { return upb_env_ok(this); } + + /* Reports an error to this environment's callback, returning true if + * the caller should try to recover. */ + bool ReportError(const Status* status) { + return upb_env_reporterror(this, status); + } + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Environment) + +#else +struct upb_env { +#endif /* __cplusplus */ + upb_arena arena_; + upb_error_func *error_func_; + void *error_ud_; + bool ok_; +}; + + +/* upb::InlinedArena **********************************************************/ +/* upb::InlinedEnvironment ****************************************************/ + +/* upb::InlinedArena and upb::InlinedEnvironment seed their arenas with a + * predefined amount of memory. No heap memory will be allocated until the + * initial block is exceeded. + * + * These types only exist in C++ */ + +#ifdef __cplusplus + +template <int N> class upb::InlinedArena : public upb::Arena { + public: + InlinedArena() : Arena(initial_block_, N, NULL) {} + explicit InlinedArena(Allocator* a) : Arena(initial_block_, N, a) {} + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(InlinedArena) + + char initial_block_[N + UPB_ARENA_BLOCK_OVERHEAD]; +}; + +template <int N> class upb::InlinedEnvironment : public upb::Environment { + public: + InlinedEnvironment() : Environment(initial_block_, N, NULL) {} + explicit InlinedEnvironment(Allocator *a) + : Environment(initial_block_, N, a) {} + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(InlinedEnvironment) + + char initial_block_[N + UPB_ARENA_BLOCK_OVERHEAD]; +}; + +#endif /* __cplusplus */ + + + +#endif /* UPB_H_ */ + +#ifdef __cplusplus +extern "C" { +#endif + + +/* upb_value ******************************************************************/ + +/* A tagged union (stored untagged inside the table) so that we can check that + * clients calling table accessors are correctly typed without having to have + * an explosion of accessors. */ +typedef enum { + UPB_CTYPE_INT32 = 1, + UPB_CTYPE_INT64 = 2, + UPB_CTYPE_UINT32 = 3, + UPB_CTYPE_UINT64 = 4, + UPB_CTYPE_BOOL = 5, + UPB_CTYPE_CSTR = 6, + UPB_CTYPE_PTR = 7, + UPB_CTYPE_CONSTPTR = 8, + UPB_CTYPE_FPTR = 9 +} upb_ctype_t; + +typedef struct { + uint64_t val; +#ifndef NDEBUG + /* In debug mode we carry the value type around also so we can check accesses + * to be sure the right member is being read. */ + upb_ctype_t ctype; +#endif +} upb_value; + +#ifdef NDEBUG +#define SET_TYPE(dest, val) UPB_UNUSED(val) +#else +#define SET_TYPE(dest, val) dest = val +#endif + +/* Like strdup(), which isn't always available since it's not ANSI C. */ +char *upb_strdup(const char *s, upb_alloc *a); +/* Variant that works with a length-delimited rather than NULL-delimited string, + * as supported by strtable. */ +char *upb_strdup2(const char *s, size_t len, upb_alloc *a); + +UPB_INLINE char *upb_gstrdup(const char *s) { + return upb_strdup(s, &upb_alloc_global); +} + +UPB_INLINE void _upb_value_setval(upb_value *v, uint64_t val, + upb_ctype_t ctype) { + v->val = val; + SET_TYPE(v->ctype, ctype); +} + +UPB_INLINE upb_value _upb_value_val(uint64_t val, upb_ctype_t ctype) { + upb_value ret; + _upb_value_setval(&ret, val, ctype); + return ret; +} + +/* For each value ctype, define the following set of functions: + * + * // Get/set an int32 from a upb_value. + * int32_t upb_value_getint32(upb_value val); + * void upb_value_setint32(upb_value *val, int32_t cval); + * + * // Construct a new upb_value from an int32. + * upb_value upb_value_int32(int32_t val); */ +#define FUNCS(name, membername, type_t, converter, proto_type) \ + UPB_INLINE void upb_value_set ## name(upb_value *val, type_t cval) { \ + val->val = (converter)cval; \ + SET_TYPE(val->ctype, proto_type); \ + } \ + UPB_INLINE upb_value upb_value_ ## name(type_t val) { \ + upb_value ret; \ + upb_value_set ## name(&ret, val); \ + return ret; \ + } \ + UPB_INLINE type_t upb_value_get ## name(upb_value val) { \ + assert(val.ctype == proto_type); \ + return (type_t)(converter)val.val; \ + } + +FUNCS(int32, int32, int32_t, int32_t, UPB_CTYPE_INT32) +FUNCS(int64, int64, int64_t, int64_t, UPB_CTYPE_INT64) +FUNCS(uint32, uint32, uint32_t, uint32_t, UPB_CTYPE_UINT32) +FUNCS(uint64, uint64, uint64_t, uint64_t, UPB_CTYPE_UINT64) +FUNCS(bool, _bool, bool, bool, UPB_CTYPE_BOOL) +FUNCS(cstr, cstr, char*, uintptr_t, UPB_CTYPE_CSTR) +FUNCS(ptr, ptr, void*, uintptr_t, UPB_CTYPE_PTR) +FUNCS(constptr, constptr, const void*, uintptr_t, UPB_CTYPE_CONSTPTR) +FUNCS(fptr, fptr, upb_func*, uintptr_t, UPB_CTYPE_FPTR) + +#undef FUNCS +#undef SET_TYPE + + +/* upb_tabkey *****************************************************************/ + +/* Either: + * 1. an actual integer key, or + * 2. a pointer to a string prefixed by its uint32_t length, owned by us. + * + * ...depending on whether this is a string table or an int table. We would + * make this a union of those two types, but C89 doesn't support statically + * initializing a non-first union member. */ +typedef uintptr_t upb_tabkey; + +#define UPB_TABKEY_NUM(n) n +#define UPB_TABKEY_NONE 0 +/* The preprocessor isn't quite powerful enough to turn the compile-time string + * length into a byte-wise string representation, so code generation needs to + * help it along. + * + * "len1" is the low byte and len4 is the high byte. */ +#ifdef UPB_BIG_ENDIAN +#define UPB_TABKEY_STR(len1, len2, len3, len4, strval) \ + (uintptr_t)(len4 len3 len2 len1 strval) +#else +#define UPB_TABKEY_STR(len1, len2, len3, len4, strval) \ + (uintptr_t)(len1 len2 len3 len4 strval) +#endif + +UPB_INLINE char *upb_tabstr(upb_tabkey key, uint32_t *len) { + char* mem = (char*)key; + if (len) memcpy(len, mem, sizeof(*len)); + return mem + sizeof(*len); +} + + +/* upb_tabval *****************************************************************/ + +#ifdef __cplusplus + +/* Status initialization not supported. + * + * This separate definition is necessary because in C++, UINTPTR_MAX isn't + * reliably available. */ +typedef struct { + uint64_t val; +} upb_tabval; + +#else + +/* C -- supports static initialization, but to support static initialization of + * both integers and points for both 32 and 64 bit targets, it takes a little + * bit of doing. */ + +#if UINTPTR_MAX == 0xffffffffffffffffULL +#define UPB_PTR_IS_64BITS +#elif UINTPTR_MAX != 0xffffffff +#error Could not determine how many bits pointers are. +#endif + +typedef union { + /* For static initialization. + * + * Unfortunately this ugliness is necessary -- it is the only way that we can, + * with -std=c89 -pedantic, statically initialize this to either a pointer or + * an integer on 32-bit platforms. */ + struct { +#ifdef UPB_PTR_IS_64BITS + uintptr_t val; +#else + uintptr_t val1; + uintptr_t val2; +#endif + } staticinit; + + /* The normal accessor that we use for everything at runtime. */ + uint64_t val; +} upb_tabval; + +#ifdef UPB_PTR_IS_64BITS +#define UPB_TABVALUE_INT_INIT(v) {{v}} +#define UPB_TABVALUE_EMPTY_INIT {{-1}} +#else + +/* 32-bit pointers */ + +#ifdef UPB_BIG_ENDIAN +#define UPB_TABVALUE_INT_INIT(v) {{0, v}} +#define UPB_TABVALUE_EMPTY_INIT {{-1, -1}} +#else +#define UPB_TABVALUE_INT_INIT(v) {{v, 0}} +#define UPB_TABVALUE_EMPTY_INIT {{-1, -1}} +#endif + +#endif + +#define UPB_TABVALUE_PTR_INIT(v) UPB_TABVALUE_INT_INIT((uintptr_t)v) + +#undef UPB_PTR_IS_64BITS + +#endif /* __cplusplus */ + + +/* upb_table ******************************************************************/ + +typedef struct _upb_tabent { + upb_tabkey key; + upb_tabval val; + + /* Internal chaining. This is const so we can create static initializers for + * tables. We cast away const sometimes, but *only* when the containing + * upb_table is known to be non-const. This requires a bit of care, but + * the subtlety is confined to table.c. */ + const struct _upb_tabent *next; +} upb_tabent; + +typedef struct { + size_t count; /* Number of entries in the hash part. */ + size_t mask; /* Mask to turn hash value -> bucket. */ + upb_ctype_t ctype; /* Type of all values. */ + uint8_t size_lg2; /* Size of the hashtable part is 2^size_lg2 entries. */ + + /* Hash table entries. + * Making this const isn't entirely accurate; what we really want is for it to + * have the same const-ness as the table it's inside. But there's no way to + * declare that in C. So we have to make it const so that we can statically + * initialize const hash tables. Then we cast away const when we have to. + */ + const upb_tabent *entries; + +#ifndef NDEBUG + /* This table's allocator. We make the user pass it in to every relevant + * function and only use this to check it in debug mode. We do this solely + * to keep upb_table as small as possible. This might seem slightly paranoid + * but the plan is to use upb_table for all map fields and extension sets in + * a forthcoming message representation, so there could be a lot of these. + * If this turns out to be too annoying later, we can change it (since this + * is an internal-only header file). */ + upb_alloc *alloc; +#endif +} upb_table; + +#ifdef NDEBUG +# define UPB_TABLE_INIT(count, mask, ctype, size_lg2, entries) \ + {count, mask, ctype, size_lg2, entries} +#else +# ifdef UPB_DEBUG_REFS +/* At the moment the only mutable tables we statically initialize are debug + * ref tables. */ +# define UPB_TABLE_INIT(count, mask, ctype, size_lg2, entries) \ + {count, mask, ctype, size_lg2, entries, &upb_alloc_debugrefs} +# else +# define UPB_TABLE_INIT(count, mask, ctype, size_lg2, entries) \ + {count, mask, ctype, size_lg2, entries, NULL} +# endif +#endif + +typedef struct { + upb_table t; +} upb_strtable; + +#define UPB_STRTABLE_INIT(count, mask, ctype, size_lg2, entries) \ + {UPB_TABLE_INIT(count, mask, ctype, size_lg2, entries)} + +#define UPB_EMPTY_STRTABLE_INIT(ctype) \ + UPB_STRTABLE_INIT(0, 0, ctype, 0, NULL) + +typedef struct { + upb_table t; /* For entries that don't fit in the array part. */ + const upb_tabval *array; /* Array part of the table. See const note above. */ + size_t array_size; /* Array part size. */ + size_t array_count; /* Array part number of elements. */ +} upb_inttable; + +#define UPB_INTTABLE_INIT(count, mask, ctype, size_lg2, ent, a, asize, acount) \ + {UPB_TABLE_INIT(count, mask, ctype, size_lg2, ent), a, asize, acount} + +#define UPB_EMPTY_INTTABLE_INIT(ctype) \ + UPB_INTTABLE_INIT(0, 0, ctype, 0, NULL, NULL, 0, 0) + +#define UPB_ARRAY_EMPTYENT -1 + +UPB_INLINE size_t upb_table_size(const upb_table *t) { + if (t->size_lg2 == 0) + return 0; + else + return 1 << t->size_lg2; +} + +/* Internal-only functions, in .h file only out of necessity. */ +UPB_INLINE bool upb_tabent_isempty(const upb_tabent *e) { + return e->key == 0; +} + +/* Used by some of the unit tests for generic hashing functionality. */ +uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed); + +UPB_INLINE uintptr_t upb_intkey(uintptr_t key) { + return key; +} + +UPB_INLINE uint32_t upb_inthash(uintptr_t key) { + return (uint32_t)key; +} + +static const upb_tabent *upb_getentry(const upb_table *t, uint32_t hash) { + return t->entries + (hash & t->mask); +} + +UPB_INLINE bool upb_arrhas(upb_tabval key) { + return key.val != (uint64_t)-1; +} + +/* Initialize and uninitialize a table, respectively. If memory allocation + * failed, false is returned that the table is uninitialized. */ +bool upb_inttable_init2(upb_inttable *table, upb_ctype_t ctype, upb_alloc *a); +bool upb_strtable_init2(upb_strtable *table, upb_ctype_t ctype, upb_alloc *a); +void upb_inttable_uninit2(upb_inttable *table, upb_alloc *a); +void upb_strtable_uninit2(upb_strtable *table, upb_alloc *a); + +UPB_INLINE bool upb_inttable_init(upb_inttable *table, upb_ctype_t ctype) { + return upb_inttable_init2(table, ctype, &upb_alloc_global); +} + +UPB_INLINE bool upb_strtable_init(upb_strtable *table, upb_ctype_t ctype) { + return upb_strtable_init2(table, ctype, &upb_alloc_global); +} + +UPB_INLINE void upb_inttable_uninit(upb_inttable *table) { + upb_inttable_uninit2(table, &upb_alloc_global); +} + +UPB_INLINE void upb_strtable_uninit(upb_strtable *table) { + upb_strtable_uninit2(table, &upb_alloc_global); +} + +/* Returns the number of values in the table. */ +size_t upb_inttable_count(const upb_inttable *t); +UPB_INLINE size_t upb_strtable_count(const upb_strtable *t) { + return t->t.count; +} + +/* Inserts the given key into the hashtable with the given value. The key must + * not already exist in the hash table. For string tables, the key must be + * NULL-terminated, and the table will make an internal copy of the key. + * Inttables must not insert a value of UINTPTR_MAX. + * + * If a table resize was required but memory allocation failed, false is + * returned and the table is unchanged. */ +bool upb_inttable_insert2(upb_inttable *t, uintptr_t key, upb_value val, + upb_alloc *a); +bool upb_strtable_insert3(upb_strtable *t, const char *key, size_t len, + upb_value val, upb_alloc *a); + +UPB_INLINE bool upb_inttable_insert(upb_inttable *t, uintptr_t key, + upb_value val) { + return upb_inttable_insert2(t, key, val, &upb_alloc_global); +} + +UPB_INLINE bool upb_strtable_insert2(upb_strtable *t, const char *key, + size_t len, upb_value val) { + return upb_strtable_insert3(t, key, len, val, &upb_alloc_global); +} + +/* For NULL-terminated strings. */ +UPB_INLINE bool upb_strtable_insert(upb_strtable *t, const char *key, + upb_value val) { + return upb_strtable_insert2(t, key, strlen(key), val); +} + +/* Looks up key in this table, returning "true" if the key was found. + * If v is non-NULL, copies the value for this key into *v. */ +bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v); +bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len, + upb_value *v); + +/* For NULL-terminated strings. */ +UPB_INLINE bool upb_strtable_lookup(const upb_strtable *t, const char *key, + upb_value *v) { + return upb_strtable_lookup2(t, key, strlen(key), v); +} + +/* Removes an item from the table. Returns true if the remove was successful, + * and stores the removed item in *val if non-NULL. */ +bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val); +bool upb_strtable_remove3(upb_strtable *t, const char *key, size_t len, + upb_value *val, upb_alloc *alloc); + +UPB_INLINE bool upb_strtable_remove2(upb_strtable *t, const char *key, + size_t len, upb_value *val) { + return upb_strtable_remove3(t, key, len, val, &upb_alloc_global); +} + +/* For NULL-terminated strings. */ +UPB_INLINE bool upb_strtable_remove(upb_strtable *t, const char *key, + upb_value *v) { + return upb_strtable_remove2(t, key, strlen(key), v); +} + +/* Updates an existing entry in an inttable. If the entry does not exist, + * returns false and does nothing. Unlike insert/remove, this does not + * invalidate iterators. */ +bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val); + +/* Handy routines for treating an inttable like a stack. May not be mixed with + * other insert/remove calls. */ +bool upb_inttable_push2(upb_inttable *t, upb_value val, upb_alloc *a); +upb_value upb_inttable_pop(upb_inttable *t); + +UPB_INLINE bool upb_inttable_push(upb_inttable *t, upb_value val) { + return upb_inttable_push2(t, val, &upb_alloc_global); +} + +/* Convenience routines for inttables with pointer keys. */ +bool upb_inttable_insertptr2(upb_inttable *t, const void *key, upb_value val, + upb_alloc *a); +bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val); +bool upb_inttable_lookupptr( + const upb_inttable *t, const void *key, upb_value *val); + +UPB_INLINE bool upb_inttable_insertptr(upb_inttable *t, const void *key, + upb_value val) { + return upb_inttable_insertptr2(t, key, val, &upb_alloc_global); +} + +/* Optimizes the table for the current set of entries, for both memory use and + * lookup time. Client should call this after all entries have been inserted; + * inserting more entries is legal, but will likely require a table resize. */ +void upb_inttable_compact2(upb_inttable *t, upb_alloc *a); + +UPB_INLINE void upb_inttable_compact(upb_inttable *t) { + upb_inttable_compact2(t, &upb_alloc_global); +} + +/* A special-case inlinable version of the lookup routine for 32-bit + * integers. */ +UPB_INLINE bool upb_inttable_lookup32(const upb_inttable *t, uint32_t key, + upb_value *v) { + *v = upb_value_int32(0); /* Silence compiler warnings. */ + if (key < t->array_size) { + upb_tabval arrval = t->array[key]; + if (upb_arrhas(arrval)) { + _upb_value_setval(v, arrval.val, t->t.ctype); + return true; + } else { + return false; + } + } else { + const upb_tabent *e; + if (t->t.entries == NULL) return false; + for (e = upb_getentry(&t->t, upb_inthash(key)); true; e = e->next) { + if ((uint32_t)e->key == key) { + _upb_value_setval(v, e->val.val, t->t.ctype); + return true; + } + if (e->next == NULL) return false; + } + } +} + +/* Exposed for testing only. */ +bool upb_strtable_resize(upb_strtable *t, size_t size_lg2, upb_alloc *a); + +/* Iterators ******************************************************************/ + +/* Iterators for int and string tables. We are subject to some kind of unusual + * design constraints: + * + * For high-level languages: + * - we must be able to guarantee that we don't crash or corrupt memory even if + * the program accesses an invalidated iterator. + * + * For C++11 range-based for: + * - iterators must be copyable + * - iterators must be comparable + * - it must be possible to construct an "end" value. + * + * Iteration order is undefined. + * + * Modifying the table invalidates iterators. upb_{str,int}table_done() is + * guaranteed to work even on an invalidated iterator, as long as the table it + * is iterating over has not been freed. Calling next() or accessing data from + * an invalidated iterator yields unspecified elements from the table, but it is + * guaranteed not to crash and to return real table elements (except when done() + * is true). */ + + +/* upb_strtable_iter **********************************************************/ + +/* upb_strtable_iter i; + * upb_strtable_begin(&i, t); + * for(; !upb_strtable_done(&i); upb_strtable_next(&i)) { + * const char *key = upb_strtable_iter_key(&i); + * const upb_value val = upb_strtable_iter_value(&i); + * // ... + * } + */ + +typedef struct { + const upb_strtable *t; + size_t index; +} upb_strtable_iter; + +void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t); +void upb_strtable_next(upb_strtable_iter *i); +bool upb_strtable_done(const upb_strtable_iter *i); +const char *upb_strtable_iter_key(const upb_strtable_iter *i); +size_t upb_strtable_iter_keylength(const upb_strtable_iter *i); +upb_value upb_strtable_iter_value(const upb_strtable_iter *i); +void upb_strtable_iter_setdone(upb_strtable_iter *i); +bool upb_strtable_iter_isequal(const upb_strtable_iter *i1, + const upb_strtable_iter *i2); + + +/* upb_inttable_iter **********************************************************/ + +/* upb_inttable_iter i; + * upb_inttable_begin(&i, t); + * for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + * uintptr_t key = upb_inttable_iter_key(&i); + * upb_value val = upb_inttable_iter_value(&i); + * // ... + * } + */ + +typedef struct { + const upb_inttable *t; + size_t index; + bool array_part; +} upb_inttable_iter; + +void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t); +void upb_inttable_next(upb_inttable_iter *i); +bool upb_inttable_done(const upb_inttable_iter *i); +uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i); +upb_value upb_inttable_iter_value(const upb_inttable_iter *i); +void upb_inttable_iter_setdone(upb_inttable_iter *i); +bool upb_inttable_iter_isequal(const upb_inttable_iter *i1, + const upb_inttable_iter *i2); + + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_TABLE_H_ */ + +/* Reference tracking will check ref()/unref() operations to make sure the + * ref ownership is correct. Where possible it will also make tools like + * Valgrind attribute ref leaks to the code that took the leaked ref, not + * the code that originally created the object. + * + * Enabling this requires the application to define upb_lock()/upb_unlock() + * functions that acquire/release a global mutex (or #define UPB_THREAD_UNSAFE). + * For this reason we don't enable it by default, even in debug builds. + */ + +/* #define UPB_DEBUG_REFS */ + +#ifdef __cplusplus +namespace upb { +class RefCounted; +template <class T> class reffed_ptr; +} +#endif + +UPB_DECLARE_TYPE(upb::RefCounted, upb_refcounted) + +struct upb_refcounted_vtbl; + +#ifdef __cplusplus + +class upb::RefCounted { + public: + /* Returns true if the given object is frozen. */ + bool IsFrozen() const; + + /* Increases the ref count, the new ref is owned by "owner" which must not + * already own a ref (and should not itself be a refcounted object if the ref + * could possibly be circular; see below). + * Thread-safe iff "this" is frozen. */ + void Ref(const void *owner) const; + + /* Release a ref that was acquired from upb_refcounted_ref() and collects any + * objects it can. */ + void Unref(const void *owner) const; + + /* Moves an existing ref from "from" to "to", without changing the overall + * ref count. DonateRef(foo, NULL, owner) is the same as Ref(foo, owner), + * but "to" may not be NULL. */ + void DonateRef(const void *from, const void *to) const; + + /* Verifies that a ref to the given object is currently held by the given + * owner. Only effective in UPB_DEBUG_REFS builds. */ + void CheckRef(const void *owner) const; + + private: + UPB_DISALLOW_POD_OPS(RefCounted, upb::RefCounted) +#else +struct upb_refcounted { +#endif + /* TODO(haberman): move the actual structure definition to structdefs.int.h. + * The only reason they are here is because inline functions need to see the + * definition of upb_handlers, which needs to see this definition. But we + * can change the upb_handlers inline functions to deal in raw offsets + * instead. + */ + + /* A single reference count shared by all objects in the group. */ + uint32_t *group; + + /* A singly-linked list of all objects in the group. */ + upb_refcounted *next; + + /* Table of function pointers for this type. */ + const struct upb_refcounted_vtbl *vtbl; + + /* Maintained only when mutable, this tracks the number of refs (but not + * ref2's) to this object. *group should be the sum of all individual_count + * in the group. */ + uint32_t individual_count; + + bool is_frozen; + +#ifdef UPB_DEBUG_REFS + upb_inttable *refs; /* Maps owner -> trackedref for incoming refs. */ + upb_inttable *ref2s; /* Set of targets for outgoing ref2s. */ +#endif +}; + +#ifdef UPB_DEBUG_REFS +extern upb_alloc upb_alloc_debugrefs; +#define UPB_REFCOUNT_INIT(vtbl, refs, ref2s) \ + {&static_refcount, NULL, vtbl, 0, true, refs, ref2s} +#else +#define UPB_REFCOUNT_INIT(vtbl, refs, ref2s) \ + {&static_refcount, NULL, vtbl, 0, true} +#endif + +UPB_BEGIN_EXTERN_C + +/* It is better to use tracked refs when possible, for the extra debugging + * capability. But if this is not possible (because you don't have easy access + * to a stable pointer value that is associated with the ref), you can pass + * UPB_UNTRACKED_REF instead. */ +extern const void *UPB_UNTRACKED_REF; + +/* Native C API. */ +bool upb_refcounted_isfrozen(const upb_refcounted *r); +void upb_refcounted_ref(const upb_refcounted *r, const void *owner); +void upb_refcounted_unref(const upb_refcounted *r, const void *owner); +void upb_refcounted_donateref( + const upb_refcounted *r, const void *from, const void *to); +void upb_refcounted_checkref(const upb_refcounted *r, const void *owner); + +#define UPB_REFCOUNTED_CMETHODS(type, upcastfunc) \ + UPB_INLINE bool type ## _isfrozen(const type *v) { \ + return upb_refcounted_isfrozen(upcastfunc(v)); \ + } \ + UPB_INLINE void type ## _ref(const type *v, const void *owner) { \ + upb_refcounted_ref(upcastfunc(v), owner); \ + } \ + UPB_INLINE void type ## _unref(const type *v, const void *owner) { \ + upb_refcounted_unref(upcastfunc(v), owner); \ + } \ + UPB_INLINE void type ## _donateref(const type *v, const void *from, const void *to) { \ + upb_refcounted_donateref(upcastfunc(v), from, to); \ + } \ + UPB_INLINE void type ## _checkref(const type *v, const void *owner) { \ + upb_refcounted_checkref(upcastfunc(v), owner); \ + } + +#define UPB_REFCOUNTED_CPPMETHODS \ + bool IsFrozen() const { \ + return upb::upcast_to<const upb::RefCounted>(this)->IsFrozen(); \ + } \ + void Ref(const void *owner) const { \ + return upb::upcast_to<const upb::RefCounted>(this)->Ref(owner); \ + } \ + void Unref(const void *owner) const { \ + return upb::upcast_to<const upb::RefCounted>(this)->Unref(owner); \ + } \ + void DonateRef(const void *from, const void *to) const { \ + return upb::upcast_to<const upb::RefCounted>(this)->DonateRef(from, to); \ + } \ + void CheckRef(const void *owner) const { \ + return upb::upcast_to<const upb::RefCounted>(this)->CheckRef(owner); \ + } + +/* Internal-to-upb Interface **************************************************/ + +typedef void upb_refcounted_visit(const upb_refcounted *r, + const upb_refcounted *subobj, + void *closure); + +struct upb_refcounted_vtbl { + /* Must visit all subobjects that are currently ref'd via upb_refcounted_ref2. + * Must be longjmp()-safe. */ + void (*visit)(const upb_refcounted *r, upb_refcounted_visit *visit, void *c); + + /* Must free the object and release all references to other objects. */ + void (*free)(upb_refcounted *r); +}; + +/* Initializes the refcounted with a single ref for the given owner. Returns + * false if memory could not be allocated. */ +bool upb_refcounted_init(upb_refcounted *r, + const struct upb_refcounted_vtbl *vtbl, + const void *owner); + +/* Adds a ref from one refcounted object to another ("from" must not already + * own a ref). These refs may be circular; cycles will be collected correctly + * (if conservatively). These refs do not need to be freed in from's free() + * function. */ +void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from); + +/* Removes a ref that was acquired from upb_refcounted_ref2(), and collects any + * object it can. This is only necessary when "from" no longer points to "r", + * and not from from's "free" function. */ +void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from); + +#define upb_ref2(r, from) \ + upb_refcounted_ref2((const upb_refcounted*)r, (upb_refcounted*)from) +#define upb_unref2(r, from) \ + upb_refcounted_unref2((const upb_refcounted*)r, (upb_refcounted*)from) + +/* Freezes all mutable object reachable by ref2() refs from the given roots. + * This will split refcounting groups into precise SCC groups, so that + * refcounting of frozen objects can be more aggressive. If memory allocation + * fails, or if more than 2**31 mutable objects are reachable from "roots", or + * if the maximum depth of the graph exceeds "maxdepth", false is returned and + * the objects are unchanged. + * + * After this operation succeeds, the objects are frozen/const, and may not be + * used through non-const pointers. In particular, they may not be passed as + * the second parameter of upb_refcounted_{ref,unref}2(). On the upside, all + * operations on frozen refcounteds are threadsafe, and objects will be freed + * at the precise moment that they become unreachable. + * + * Caller must own refs on each object in the "roots" list. */ +bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth); + +/* Shared by all compiled-in refcounted objects. */ +extern uint32_t static_refcount; + +UPB_END_EXTERN_C + +#ifdef __cplusplus +/* C++ Wrappers. */ +namespace upb { +inline bool RefCounted::IsFrozen() const { + return upb_refcounted_isfrozen(this); +} +inline void RefCounted::Ref(const void *owner) const { + upb_refcounted_ref(this, owner); +} +inline void RefCounted::Unref(const void *owner) const { + upb_refcounted_unref(this, owner); +} +inline void RefCounted::DonateRef(const void *from, const void *to) const { + upb_refcounted_donateref(this, from, to); +} +inline void RefCounted::CheckRef(const void *owner) const { + upb_refcounted_checkref(this, owner); +} +} /* namespace upb */ +#endif + + +/* upb::reffed_ptr ************************************************************/ + +#ifdef __cplusplus + +#include <algorithm> /* For std::swap(). */ + +/* Provides RAII semantics for upb refcounted objects. Each reffed_ptr owns a + * ref on whatever object it points to (if any). */ +template <class T> class upb::reffed_ptr { + public: + reffed_ptr() : ptr_(NULL) {} + + /* If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. */ + template <class U> + reffed_ptr(U* val, const void* ref_donor = NULL) + : ptr_(upb::upcast(val)) { + if (ref_donor) { + assert(ptr_); + ptr_->DonateRef(ref_donor, this); + } else if (ptr_) { + ptr_->Ref(this); + } + } + + template <class U> + reffed_ptr(const reffed_ptr<U>& other) + : ptr_(upb::upcast(other.get())) { + if (ptr_) ptr_->Ref(this); + } + + reffed_ptr(const reffed_ptr& other) + : ptr_(upb::upcast(other.get())) { + if (ptr_) ptr_->Ref(this); + } + + ~reffed_ptr() { if (ptr_) ptr_->Unref(this); } + + template <class U> + reffed_ptr& operator=(const reffed_ptr<U>& other) { + reset(other.get()); + return *this; + } + + reffed_ptr& operator=(const reffed_ptr& other) { + reset(other.get()); + return *this; + } + + /* TODO(haberman): add C++11 move construction/assignment for greater + * efficiency. */ + + void swap(reffed_ptr& other) { + if (ptr_ == other.ptr_) { + return; + } + + if (ptr_) ptr_->DonateRef(this, &other); + if (other.ptr_) other.ptr_->DonateRef(&other, this); + std::swap(ptr_, other.ptr_); + } + + T& operator*() const { + assert(ptr_); + return *ptr_; + } + + T* operator->() const { + assert(ptr_); + return ptr_; + } + + T* get() const { return ptr_; } + + /* If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. */ + template <class U> + void reset(U* ptr = NULL, const void* ref_donor = NULL) { + reffed_ptr(ptr, ref_donor).swap(*this); + } + + template <class U> + reffed_ptr<U> down_cast() { + return reffed_ptr<U>(upb::down_cast<U*>(get())); + } + + template <class U> + reffed_ptr<U> dyn_cast() { + return reffed_ptr<U>(upb::dyn_cast<U*>(get())); + } + + /* Plain release() is unsafe; if we were the only owner, it would leak the + * object. Instead we provide this: */ + T* ReleaseTo(const void* new_owner) { + T* ret = NULL; + ptr_->DonateRef(this, new_owner); + std::swap(ret, ptr_); + return ret; + } + + private: + T* ptr_; +}; + +#endif /* __cplusplus */ + +#endif /* UPB_REFCOUNT_H_ */ + +#ifdef __cplusplus +#include <cstring> +#include <string> +#include <vector> + +namespace upb { +class Def; +class EnumDef; +class FieldDef; +class FileDef; +class MessageDef; +class OneofDef; +} +#endif + +UPB_DECLARE_DERIVED_TYPE(upb::Def, upb::RefCounted, upb_def, upb_refcounted) +UPB_DECLARE_DERIVED_TYPE(upb::OneofDef, upb::RefCounted, upb_oneofdef, + upb_refcounted) +UPB_DECLARE_DERIVED_TYPE(upb::FileDef, upb::RefCounted, upb_filedef, + upb_refcounted) + +/* The maximum message depth that the type graph can have. This is a resource + * limit for the C stack since we sometimes need to recursively traverse the + * graph. Cycles are ok; the traversal will stop when it detects a cycle, but + * we must hit the cycle before the maximum depth is reached. + * + * If having a single static limit is too inflexible, we can add another variant + * of Def::Freeze that allows specifying this as a parameter. */ +#define UPB_MAX_MESSAGE_DEPTH 64 + + +/* upb::Def: base class for top-level defs ***********************************/ + +/* All the different kind of defs that can be defined at the top-level and put + * in a SymbolTable or appear in a FileDef::defs() list. This excludes some + * defs (like oneofs and files). It only includes fields because they can be + * defined as extensions. */ +typedef enum { + UPB_DEF_MSG, + UPB_DEF_FIELD, + UPB_DEF_ENUM, + UPB_DEF_SERVICE, /* Not yet implemented. */ + UPB_DEF_ANY = -1 /* Wildcard for upb_symtab_get*() */ +} upb_deftype_t; + +#ifdef __cplusplus + +/* The base class of all defs. Its base is upb::RefCounted (use upb::upcast() + * to convert). */ +class upb::Def { + public: + typedef upb_deftype_t Type; + + Def* Dup(const void *owner) const; + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + Type def_type() const; + + /* "fullname" is the def's fully-qualified name (eg. foo.bar.Message). */ + const char *full_name() const; + + /* The final part of a def's name (eg. Message). */ + const char *name() const; + + /* The def must be mutable. Caller retains ownership of fullname. Defs are + * not required to have a name; if a def has no name when it is frozen, it + * will remain an anonymous def. On failure, returns false and details in "s" + * if non-NULL. */ + bool set_full_name(const char* fullname, upb::Status* s); + bool set_full_name(const std::string &fullname, upb::Status* s); + + /* The file in which this def appears. It is not necessary to add a def to a + * file (and consequently the accessor may return NULL). Set this by calling + * file->Add(def). */ + FileDef* file() const; + + /* Freezes the given defs; this validates all constraints and marks the defs + * as frozen (read-only). "defs" may not contain any fielddefs, but fields + * of any msgdefs will be frozen. + * + * Symbolic references to sub-types and enum defaults must have already been + * resolved. Any mutable defs reachable from any of "defs" must also be in + * the list; more formally, "defs" must be a transitive closure of mutable + * defs. + * + * After this operation succeeds, the finalized defs must only be accessed + * through a const pointer! */ + static bool Freeze(Def* const* defs, size_t n, Status* status); + static bool Freeze(const std::vector<Def*>& defs, Status* status); + + private: + UPB_DISALLOW_POD_OPS(Def, upb::Def) +}; + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ +upb_def *upb_def_dup(const upb_def *def, const void *owner); + +/* Include upb_refcounted methods like upb_def_ref()/upb_def_unref(). */ +UPB_REFCOUNTED_CMETHODS(upb_def, upb_def_upcast) + +upb_deftype_t upb_def_type(const upb_def *d); +const char *upb_def_fullname(const upb_def *d); +const char *upb_def_name(const upb_def *d); +const upb_filedef *upb_def_file(const upb_def *d); +bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s); +bool upb_def_freeze(upb_def *const *defs, size_t n, upb_status *s); + +/* Temporary API: for internal use only. */ +bool _upb_def_validate(upb_def *const*defs, size_t n, upb_status *s); + +UPB_END_EXTERN_C + + +/* upb::Def casts *************************************************************/ + +#ifdef __cplusplus +#define UPB_CPP_CASTS(cname, cpptype) \ + namespace upb { \ + template <> \ + inline cpptype *down_cast<cpptype *, Def>(Def * def) { \ + return upb_downcast_##cname##_mutable(def); \ + } \ + template <> \ + inline cpptype *dyn_cast<cpptype *, Def>(Def * def) { \ + return upb_dyncast_##cname##_mutable(def); \ + } \ + template <> \ + inline const cpptype *down_cast<const cpptype *, const Def>( \ + const Def *def) { \ + return upb_downcast_##cname(def); \ + } \ + template <> \ + inline const cpptype *dyn_cast<const cpptype *, const Def>(const Def *def) { \ + return upb_dyncast_##cname(def); \ + } \ + template <> \ + inline const cpptype *down_cast<const cpptype *, Def>(Def * def) { \ + return upb_downcast_##cname(def); \ + } \ + template <> \ + inline const cpptype *dyn_cast<const cpptype *, Def>(Def * def) { \ + return upb_dyncast_##cname(def); \ + } \ + } /* namespace upb */ +#else +#define UPB_CPP_CASTS(cname, cpptype) +#endif /* __cplusplus */ + +/* Dynamic casts, for determining if a def is of a particular type at runtime. + * Downcasts, for when some wants to assert that a def is of a particular type. + * These are only checked if we are building debug. */ +#define UPB_DEF_CASTS(lower, upper, cpptype) \ + UPB_INLINE const upb_##lower *upb_dyncast_##lower(const upb_def *def) { \ + if (upb_def_type(def) != UPB_DEF_##upper) return NULL; \ + return (upb_##lower *)def; \ + } \ + UPB_INLINE const upb_##lower *upb_downcast_##lower(const upb_def *def) { \ + assert(upb_def_type(def) == UPB_DEF_##upper); \ + return (const upb_##lower *)def; \ + } \ + UPB_INLINE upb_##lower *upb_dyncast_##lower##_mutable(upb_def *def) { \ + return (upb_##lower *)upb_dyncast_##lower(def); \ + } \ + UPB_INLINE upb_##lower *upb_downcast_##lower##_mutable(upb_def *def) { \ + return (upb_##lower *)upb_downcast_##lower(def); \ + } \ + UPB_CPP_CASTS(lower, cpptype) + +#define UPB_DEFINE_DEF(cppname, lower, upper, cppmethods, members) \ + UPB_DEFINE_CLASS2(cppname, upb::Def, upb::RefCounted, cppmethods, \ + members) \ + UPB_DEF_CASTS(lower, upper, cppname) + +#define UPB_DECLARE_DEF_TYPE(cppname, lower, upper) \ + UPB_DECLARE_DERIVED_TYPE2(cppname, upb::Def, upb::RefCounted, \ + upb_ ## lower, upb_def, upb_refcounted) \ + UPB_DEF_CASTS(lower, upper, cppname) + +UPB_DECLARE_DEF_TYPE(upb::FieldDef, fielddef, FIELD) +UPB_DECLARE_DEF_TYPE(upb::MessageDef, msgdef, MSG) +UPB_DECLARE_DEF_TYPE(upb::EnumDef, enumdef, ENUM) + +#undef UPB_DECLARE_DEF_TYPE +#undef UPB_DEF_CASTS +#undef UPB_CPP_CASTS + + +/* upb::FieldDef **************************************************************/ + +/* The types a field can have. Note that this list is not identical to the + * types defined in descriptor.proto, which gives INT32 and SINT32 separate + * types (we distinguish the two with the "integer encoding" enum below). */ +typedef enum { + UPB_TYPE_FLOAT = 1, + UPB_TYPE_DOUBLE = 2, + UPB_TYPE_BOOL = 3, + UPB_TYPE_STRING = 4, + UPB_TYPE_BYTES = 5, + UPB_TYPE_MESSAGE = 6, + UPB_TYPE_ENUM = 7, /* Enum values are int32. */ + UPB_TYPE_INT32 = 8, + UPB_TYPE_UINT32 = 9, + UPB_TYPE_INT64 = 10, + UPB_TYPE_UINT64 = 11 +} upb_fieldtype_t; + +/* The repeated-ness of each field; this matches descriptor.proto. */ +typedef enum { + UPB_LABEL_OPTIONAL = 1, + UPB_LABEL_REQUIRED = 2, + UPB_LABEL_REPEATED = 3 +} upb_label_t; + +/* How integers should be encoded in serializations that offer multiple + * integer encoding methods. */ +typedef enum { + UPB_INTFMT_VARIABLE = 1, + UPB_INTFMT_FIXED = 2, + UPB_INTFMT_ZIGZAG = 3 /* Only for signed types (INT32/INT64). */ +} upb_intfmt_t; + +/* Descriptor types, as defined in descriptor.proto. */ +typedef enum { + UPB_DESCRIPTOR_TYPE_DOUBLE = 1, + UPB_DESCRIPTOR_TYPE_FLOAT = 2, + UPB_DESCRIPTOR_TYPE_INT64 = 3, + UPB_DESCRIPTOR_TYPE_UINT64 = 4, + UPB_DESCRIPTOR_TYPE_INT32 = 5, + UPB_DESCRIPTOR_TYPE_FIXED64 = 6, + UPB_DESCRIPTOR_TYPE_FIXED32 = 7, + UPB_DESCRIPTOR_TYPE_BOOL = 8, + UPB_DESCRIPTOR_TYPE_STRING = 9, + UPB_DESCRIPTOR_TYPE_GROUP = 10, + UPB_DESCRIPTOR_TYPE_MESSAGE = 11, + UPB_DESCRIPTOR_TYPE_BYTES = 12, + UPB_DESCRIPTOR_TYPE_UINT32 = 13, + UPB_DESCRIPTOR_TYPE_ENUM = 14, + UPB_DESCRIPTOR_TYPE_SFIXED32 = 15, + UPB_DESCRIPTOR_TYPE_SFIXED64 = 16, + UPB_DESCRIPTOR_TYPE_SINT32 = 17, + UPB_DESCRIPTOR_TYPE_SINT64 = 18 +} upb_descriptortype_t; + +typedef enum { + UPB_SYNTAX_PROTO2 = 2, + UPB_SYNTAX_PROTO3 = 3 +} upb_syntax_t; + +/* Maximum field number allowed for FieldDefs. This is an inherent limit of the + * protobuf wire format. */ +#define UPB_MAX_FIELDNUMBER ((1 << 29) - 1) + +#ifdef __cplusplus + +/* A upb_fielddef describes a single field in a message. It is most often + * found as a part of a upb_msgdef, but can also stand alone to represent + * an extension. + * + * Its base class is upb::Def (use upb::upcast() to convert). */ +class upb::FieldDef { + public: + typedef upb_fieldtype_t Type; + typedef upb_label_t Label; + typedef upb_intfmt_t IntegerFormat; + typedef upb_descriptortype_t DescriptorType; + + /* These return true if the given value is a valid member of the enumeration. */ + static bool CheckType(int32_t val); + static bool CheckLabel(int32_t val); + static bool CheckDescriptorType(int32_t val); + static bool CheckIntegerFormat(int32_t val); + + /* These convert to the given enumeration; they require that the value is + * valid. */ + static Type ConvertType(int32_t val); + static Label ConvertLabel(int32_t val); + static DescriptorType ConvertDescriptorType(int32_t val); + static IntegerFormat ConvertIntegerFormat(int32_t val); + + /* Returns NULL if memory allocation failed. */ + static reffed_ptr<FieldDef> New(); + + /* Duplicates the given field, returning NULL if memory allocation failed. + * When a fielddef is duplicated, the subdef (if any) is made symbolic if it + * wasn't already. If the subdef is set but has no name (which is possible + * since msgdefs are not required to have a name) the new fielddef's subdef + * will be unset. */ + FieldDef* Dup(const void* owner) const; + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + /* Functionality from upb::Def. */ + const char* full_name() const; + + bool type_is_set() const; /* set_[descriptor_]type() has been called? */ + Type type() const; /* Requires that type_is_set() == true. */ + Label label() const; /* Defaults to UPB_LABEL_OPTIONAL. */ + const char* name() const; /* NULL if uninitialized. */ + uint32_t number() const; /* Returns 0 if uninitialized. */ + bool is_extension() const; + + /* Copies the JSON name for this field into the given buffer. Returns the + * actual size of the JSON name, including the NULL terminator. If the + * return value is 0, the JSON name is unset. If the return value is + * greater than len, the JSON name was truncated. The buffer is always + * NULL-terminated if len > 0. + * + * The JSON name always defaults to a camelCased version of the regular + * name. However if the regular name is unset, the JSON name will be unset + * also. + */ + size_t GetJsonName(char* buf, size_t len) const; + + /* Convenience version of the above function which copies the JSON name + * into the given string, returning false if the name is not set. */ + template <class T> + bool GetJsonName(T* str) { + str->resize(GetJsonName(NULL, 0)); + GetJsonName(&(*str)[0], str->size()); + return str->size() > 0; + } + + /* For UPB_TYPE_MESSAGE fields only where is_tag_delimited() == false, + * indicates whether this field should have lazy parsing handlers that yield + * the unparsed string for the submessage. + * + * TODO(haberman): I think we want to move this into a FieldOptions container + * when we add support for custom options (the FieldOptions struct will + * contain both regular FieldOptions like "lazy" *and* custom options). */ + bool lazy() const; + + /* For non-string, non-submessage fields, this indicates whether binary + * protobufs are encoded in packed or non-packed format. + * + * TODO(haberman): see note above about putting options like this into a + * FieldOptions container. */ + bool packed() const; + + /* An integer that can be used as an index into an array of fields for + * whatever message this field belongs to. Guaranteed to be less than + * f->containing_type()->field_count(). May only be accessed once the def has + * been finalized. */ + uint32_t index() const; + + /* The MessageDef to which this field belongs. + * + * If this field has been added to a MessageDef, that message can be retrieved + * directly (this is always the case for frozen FieldDefs). + * + * If the field has not yet been added to a MessageDef, you can set the name + * of the containing type symbolically instead. This is mostly useful for + * extensions, where the extension is declared separately from the message. */ + const MessageDef* containing_type() const; + const char* containing_type_name(); + + /* The OneofDef to which this field belongs, or NULL if this field is not part + * of a oneof. */ + const OneofDef* containing_oneof() const; + + /* The field's type according to the enum in descriptor.proto. This is not + * the same as UPB_TYPE_*, because it distinguishes between (for example) + * INT32 and SINT32, whereas our "type" enum does not. This return of + * descriptor_type() is a function of type(), integer_format(), and + * is_tag_delimited(). Likewise set_descriptor_type() sets all three + * appropriately. */ + DescriptorType descriptor_type() const; + + /* Convenient field type tests. */ + bool IsSubMessage() const; + bool IsString() const; + bool IsSequence() const; + bool IsPrimitive() const; + bool IsMap() const; + + /* Whether this field must be able to explicitly represent presence: + * + * * This is always false for repeated fields (an empty repeated field is + * equivalent to a repeated field with zero entries). + * + * * This is always true for submessages. + * + * * For other fields, it depends on the message (see + * MessageDef::SetPrimitivesHavePresence()) + */ + bool HasPresence() const; + + /* How integers are encoded. Only meaningful for integer types. + * Defaults to UPB_INTFMT_VARIABLE, and is reset when "type" changes. */ + IntegerFormat integer_format() const; + + /* Whether a submessage field is tag-delimited or not (if false, then + * length-delimited). May only be set when type() == UPB_TYPE_MESSAGE. */ + bool is_tag_delimited() const; + + /* Returns the non-string default value for this fielddef, which may either + * be something the client set explicitly or the "default default" (0 for + * numbers, empty for strings). The field's type indicates the type of the + * returned value, except for enum fields that are still mutable. + * + * Requires that the given function matches the field's current type. */ + int64_t default_int64() const; + int32_t default_int32() const; + uint64_t default_uint64() const; + uint32_t default_uint32() const; + bool default_bool() const; + float default_float() const; + double default_double() const; + + /* The resulting string is always NULL-terminated. If non-NULL, the length + * will be stored in *len. */ + const char *default_string(size_t* len) const; + + /* For frozen UPB_TYPE_ENUM fields, enum defaults can always be read as either + * string or int32, and both of these methods will always return true. + * + * For mutable UPB_TYPE_ENUM fields, the story is a bit more complicated. + * Enum defaults are unusual. They can be specified either as string or int32, + * but to be valid the enum must have that value as a member. And if no + * default is specified, the "default default" comes from the EnumDef. + * + * We allow reading the default as either an int32 or a string, but only if + * we have a meaningful value to report. We have a meaningful value if it was + * set explicitly, or if we could get the "default default" from the EnumDef. + * Also if you explicitly set the name and we find the number in the EnumDef */ + bool EnumHasStringDefault() const; + bool EnumHasInt32Default() const; + + /* Submessage and enum fields must reference a "subdef", which is the + * upb::MessageDef or upb::EnumDef that defines their type. Note that when + * the FieldDef is mutable it may not have a subdef *yet*, but this function + * still returns true to indicate that the field's type requires a subdef. */ + bool HasSubDef() const; + + /* Returns the enum or submessage def for this field, if any. The field's + * type must match (ie. you may only call enum_subdef() for fields where + * type() == UPB_TYPE_ENUM). Returns NULL if the subdef has not been set or + * is currently set symbolically. */ + const EnumDef* enum_subdef() const; + const MessageDef* message_subdef() const; + + /* Returns the generic subdef for this field. Requires that HasSubDef() (ie. + * only works for UPB_TYPE_ENUM and UPB_TYPE_MESSAGE fields). */ + const Def* subdef() const; + + /* Returns the symbolic name of the subdef. If the subdef is currently set + * unresolved (ie. set symbolically) returns the symbolic name. If it has + * been resolved to a specific subdef, returns the name from that subdef. */ + const char* subdef_name() const; + + /* Setters (non-const methods), only valid for mutable FieldDefs! ***********/ + + bool set_full_name(const char* fullname, upb::Status* s); + bool set_full_name(const std::string& fullname, upb::Status* s); + + /* This may only be called if containing_type() == NULL (ie. the field has not + * been added to a message yet). */ + bool set_containing_type_name(const char *name, Status* status); + bool set_containing_type_name(const std::string& name, Status* status); + + /* Defaults to false. When we freeze, we ensure that this can only be true + * for length-delimited message fields. Prior to freezing this can be true or + * false with no restrictions. */ + void set_lazy(bool lazy); + + /* Defaults to true. Sets whether this field is encoded in packed format. */ + void set_packed(bool packed); + + /* "type" or "descriptor_type" MUST be set explicitly before the fielddef is + * finalized. These setters require that the enum value is valid; if the + * value did not come directly from an enum constant, the caller should + * validate it first with the functions above (CheckFieldType(), etc). */ + void set_type(Type type); + void set_label(Label label); + void set_descriptor_type(DescriptorType type); + void set_is_extension(bool is_extension); + + /* "number" and "name" must be set before the FieldDef is added to a + * MessageDef, and may not be set after that. + * + * "name" is the same as full_name()/set_full_name(), but since fielddefs + * most often use simple, non-qualified names, we provide this accessor + * also. Generally only extensions will want to think of this name as + * fully-qualified. */ + bool set_number(uint32_t number, upb::Status* s); + bool set_name(const char* name, upb::Status* s); + bool set_name(const std::string& name, upb::Status* s); + + /* Sets the JSON name to the given string. */ + /* TODO(haberman): implement. Right now only default json_name (camelCase) + * is supported. */ + bool set_json_name(const char* json_name, upb::Status* s); + bool set_json_name(const std::string& name, upb::Status* s); + + /* Clears the JSON name. This will make it revert to its default, which is + * a camelCased version of the regular field name. */ + void clear_json_name(); + + void set_integer_format(IntegerFormat format); + bool set_tag_delimited(bool tag_delimited, upb::Status* s); + + /* Sets default value for the field. The call must exactly match the type + * of the field. Enum fields may use either setint32 or setstring to set + * the default numerically or symbolically, respectively, but symbolic + * defaults must be resolved before finalizing (see ResolveEnumDefault()). + * + * Changing the type of a field will reset its default. */ + void set_default_int64(int64_t val); + void set_default_int32(int32_t val); + void set_default_uint64(uint64_t val); + void set_default_uint32(uint32_t val); + void set_default_bool(bool val); + void set_default_float(float val); + void set_default_double(double val); + bool set_default_string(const void *str, size_t len, Status *s); + bool set_default_string(const std::string &str, Status *s); + void set_default_cstr(const char *str, Status *s); + + /* Before a fielddef is frozen, its subdef may be set either directly (with a + * upb::Def*) or symbolically. Symbolic refs must be resolved before the + * containing msgdef can be frozen (see upb_resolve() above). upb always + * guarantees that any def reachable from a live def will also be kept alive. + * + * Both methods require that upb_hassubdef(f) (so the type must be set prior + * to calling these methods). Returns false if this is not the case, or if + * the given subdef is not of the correct type. The subdef is reset if the + * field's type is changed. The subdef can be set to NULL to clear it. */ + bool set_subdef(const Def* subdef, Status* s); + bool set_enum_subdef(const EnumDef* subdef, Status* s); + bool set_message_subdef(const MessageDef* subdef, Status* s); + bool set_subdef_name(const char* name, Status* s); + bool set_subdef_name(const std::string &name, Status* s); + + private: + UPB_DISALLOW_POD_OPS(FieldDef, upb::FieldDef) +}; + +# endif /* defined(__cplusplus) */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ +upb_fielddef *upb_fielddef_new(const void *owner); +upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner); + +/* Include upb_refcounted methods like upb_fielddef_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_fielddef, upb_fielddef_upcast2) + +/* Methods from upb_def. */ +const char *upb_fielddef_fullname(const upb_fielddef *f); +bool upb_fielddef_setfullname(upb_fielddef *f, const char *fullname, + upb_status *s); + +bool upb_fielddef_typeisset(const upb_fielddef *f); +upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f); +upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f); +upb_label_t upb_fielddef_label(const upb_fielddef *f); +uint32_t upb_fielddef_number(const upb_fielddef *f); +const char *upb_fielddef_name(const upb_fielddef *f); +bool upb_fielddef_isextension(const upb_fielddef *f); +bool upb_fielddef_lazy(const upb_fielddef *f); +bool upb_fielddef_packed(const upb_fielddef *f); +size_t upb_fielddef_getjsonname(const upb_fielddef *f, char *buf, size_t len); +const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f); +const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f); +upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f); +const char *upb_fielddef_containingtypename(upb_fielddef *f); +upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f); +uint32_t upb_fielddef_index(const upb_fielddef *f); +bool upb_fielddef_istagdelim(const upb_fielddef *f); +bool upb_fielddef_issubmsg(const upb_fielddef *f); +bool upb_fielddef_isstring(const upb_fielddef *f); +bool upb_fielddef_isseq(const upb_fielddef *f); +bool upb_fielddef_isprimitive(const upb_fielddef *f); +bool upb_fielddef_ismap(const upb_fielddef *f); +bool upb_fielddef_haspresence(const upb_fielddef *f); +int64_t upb_fielddef_defaultint64(const upb_fielddef *f); +int32_t upb_fielddef_defaultint32(const upb_fielddef *f); +uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f); +uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f); +bool upb_fielddef_defaultbool(const upb_fielddef *f); +float upb_fielddef_defaultfloat(const upb_fielddef *f); +double upb_fielddef_defaultdouble(const upb_fielddef *f); +const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len); +bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f); +bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f); +bool upb_fielddef_hassubdef(const upb_fielddef *f); +const upb_def *upb_fielddef_subdef(const upb_fielddef *f); +const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f); +const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f); +const char *upb_fielddef_subdefname(const upb_fielddef *f); + +void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type); +void upb_fielddef_setdescriptortype(upb_fielddef *f, int type); +void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label); +bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s); +bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s); +bool upb_fielddef_setjsonname(upb_fielddef *f, const char *name, upb_status *s); +bool upb_fielddef_clearjsonname(upb_fielddef *f); +bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name, + upb_status *s); +void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension); +void upb_fielddef_setlazy(upb_fielddef *f, bool lazy); +void upb_fielddef_setpacked(upb_fielddef *f, bool packed); +void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt); +void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim); +void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t val); +void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t val); +void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t val); +void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t val); +void upb_fielddef_setdefaultbool(upb_fielddef *f, bool val); +void upb_fielddef_setdefaultfloat(upb_fielddef *f, float val); +void upb_fielddef_setdefaultdouble(upb_fielddef *f, double val); +bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len, + upb_status *s); +void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str, + upb_status *s); +bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef, + upb_status *s); +bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef, + upb_status *s); +bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef, + upb_status *s); +bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name, + upb_status *s); + +bool upb_fielddef_checklabel(int32_t label); +bool upb_fielddef_checktype(int32_t type); +bool upb_fielddef_checkdescriptortype(int32_t type); +bool upb_fielddef_checkintfmt(int32_t fmt); + +UPB_END_EXTERN_C + + +/* upb::MessageDef ************************************************************/ + +typedef upb_inttable_iter upb_msg_field_iter; +typedef upb_strtable_iter upb_msg_oneof_iter; + +/* Well-known field tag numbers for map-entry messages. */ +#define UPB_MAPENTRY_KEY 1 +#define UPB_MAPENTRY_VALUE 2 + +#ifdef __cplusplus + +/* Structure that describes a single .proto message type. + * + * Its base class is upb::Def (use upb::upcast() to convert). */ +class upb::MessageDef { + public: + /* Returns NULL if memory allocation failed. */ + static reffed_ptr<MessageDef> New(); + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + /* Functionality from upb::Def. */ + const char* full_name() const; + const char* name() const; + bool set_full_name(const char* fullname, Status* s); + bool set_full_name(const std::string& fullname, Status* s); + + /* Call to freeze this MessageDef. + * WARNING: this will fail if this message has any unfrozen submessages! + * Messages with cycles must be frozen as a batch using upb::Def::Freeze(). */ + bool Freeze(Status* s); + + /* The number of fields that belong to the MessageDef. */ + int field_count() const; + + /* The number of oneofs that belong to the MessageDef. */ + int oneof_count() const; + + /* Adds a field (upb_fielddef object) to a msgdef. Requires that the msgdef + * and the fielddefs are mutable. The fielddef's name and number must be + * set, and the message may not already contain any field with this name or + * number, and this fielddef may not be part of another message. In error + * cases false is returned and the msgdef is unchanged. + * + * If the given field is part of a oneof, this call succeeds if and only if + * that oneof is already part of this msgdef. (Note that adding a oneof to a + * msgdef automatically adds all of its fields to the msgdef at the time that + * the oneof is added, so it is usually more idiomatic to add the oneof's + * fields first then add the oneof to the msgdef. This case is supported for + * convenience.) + * + * If |f| is already part of this MessageDef, this method performs no action + * and returns true (success). Thus, this method is idempotent. */ + bool AddField(FieldDef* f, Status* s); + bool AddField(const reffed_ptr<FieldDef>& f, Status* s); + + /* Adds a oneof (upb_oneofdef object) to a msgdef. Requires that the msgdef, + * oneof, and any fielddefs are mutable, that the fielddefs contained in the + * oneof do not have any name or number conflicts with existing fields in the + * msgdef, and that the oneof's name is unique among all oneofs in the msgdef. + * If the oneof is added successfully, all of its fields will be added + * directly to the msgdef as well. In error cases, false is returned and the + * msgdef is unchanged. */ + bool AddOneof(OneofDef* o, Status* s); + bool AddOneof(const reffed_ptr<OneofDef>& o, Status* s); + + upb_syntax_t syntax() const; + + /* Returns false if we don't support this syntax value. */ + bool set_syntax(upb_syntax_t syntax); + + /* Set this to false to indicate that primitive fields should not have + * explicit presence information associated with them. This will affect all + * fields added to this message. Defaults to true. */ + void SetPrimitivesHavePresence(bool have_presence); + + /* These return NULL if the field is not found. */ + FieldDef* FindFieldByNumber(uint32_t number); + FieldDef* FindFieldByName(const char *name, size_t len); + const FieldDef* FindFieldByNumber(uint32_t number) const; + const FieldDef* FindFieldByName(const char* name, size_t len) const; + + + FieldDef* FindFieldByName(const char *name) { + return FindFieldByName(name, strlen(name)); + } + const FieldDef* FindFieldByName(const char *name) const { + return FindFieldByName(name, strlen(name)); + } + + template <class T> + FieldDef* FindFieldByName(const T& str) { + return FindFieldByName(str.c_str(), str.size()); + } + template <class T> + const FieldDef* FindFieldByName(const T& str) const { + return FindFieldByName(str.c_str(), str.size()); + } + + OneofDef* FindOneofByName(const char* name, size_t len); + const OneofDef* FindOneofByName(const char* name, size_t len) const; + + OneofDef* FindOneofByName(const char* name) { + return FindOneofByName(name, strlen(name)); + } + const OneofDef* FindOneofByName(const char* name) const { + return FindOneofByName(name, strlen(name)); + } + + template<class T> + OneofDef* FindOneofByName(const T& str) { + return FindOneofByName(str.c_str(), str.size()); + } + template<class T> + const OneofDef* FindOneofByName(const T& str) const { + return FindOneofByName(str.c_str(), str.size()); + } + + /* Returns a new msgdef that is a copy of the given msgdef (and a copy of all + * the fields) but with any references to submessages broken and replaced + * with just the name of the submessage. Returns NULL if memory allocation + * failed. + * + * TODO(haberman): which is more useful, keeping fields resolved or + * unresolving them? If there's no obvious answer, Should this functionality + * just be moved into symtab.c? */ + MessageDef* Dup(const void* owner) const; + + /* Is this message a map entry? */ + void setmapentry(bool map_entry); + bool mapentry() const; + + /* Iteration over fields. The order is undefined. */ + class field_iterator + : public std::iterator<std::forward_iterator_tag, FieldDef*> { + public: + explicit field_iterator(MessageDef* md); + static field_iterator end(MessageDef* md); + + void operator++(); + FieldDef* operator*() const; + bool operator!=(const field_iterator& other) const; + bool operator==(const field_iterator& other) const; + + private: + upb_msg_field_iter iter_; + }; + + class const_field_iterator + : public std::iterator<std::forward_iterator_tag, const FieldDef*> { + public: + explicit const_field_iterator(const MessageDef* md); + static const_field_iterator end(const MessageDef* md); + + void operator++(); + const FieldDef* operator*() const; + bool operator!=(const const_field_iterator& other) const; + bool operator==(const const_field_iterator& other) const; + + private: + upb_msg_field_iter iter_; + }; + + /* Iteration over oneofs. The order is undefined. */ + class oneof_iterator + : public std::iterator<std::forward_iterator_tag, FieldDef*> { + public: + explicit oneof_iterator(MessageDef* md); + static oneof_iterator end(MessageDef* md); + + void operator++(); + OneofDef* operator*() const; + bool operator!=(const oneof_iterator& other) const; + bool operator==(const oneof_iterator& other) const; + + private: + upb_msg_oneof_iter iter_; + }; + + class const_oneof_iterator + : public std::iterator<std::forward_iterator_tag, const FieldDef*> { + public: + explicit const_oneof_iterator(const MessageDef* md); + static const_oneof_iterator end(const MessageDef* md); + + void operator++(); + const OneofDef* operator*() const; + bool operator!=(const const_oneof_iterator& other) const; + bool operator==(const const_oneof_iterator& other) const; + + private: + upb_msg_oneof_iter iter_; + }; + + class FieldAccessor { + public: + explicit FieldAccessor(MessageDef* msg) : msg_(msg) {} + field_iterator begin() { return msg_->field_begin(); } + field_iterator end() { return msg_->field_end(); } + private: + MessageDef* msg_; + }; + + class ConstFieldAccessor { + public: + explicit ConstFieldAccessor(const MessageDef* msg) : msg_(msg) {} + const_field_iterator begin() { return msg_->field_begin(); } + const_field_iterator end() { return msg_->field_end(); } + private: + const MessageDef* msg_; + }; + + class OneofAccessor { + public: + explicit OneofAccessor(MessageDef* msg) : msg_(msg) {} + oneof_iterator begin() { return msg_->oneof_begin(); } + oneof_iterator end() { return msg_->oneof_end(); } + private: + MessageDef* msg_; + }; + + class ConstOneofAccessor { + public: + explicit ConstOneofAccessor(const MessageDef* msg) : msg_(msg) {} + const_oneof_iterator begin() { return msg_->oneof_begin(); } + const_oneof_iterator end() { return msg_->oneof_end(); } + private: + const MessageDef* msg_; + }; + + field_iterator field_begin(); + field_iterator field_end(); + const_field_iterator field_begin() const; + const_field_iterator field_end() const; + + oneof_iterator oneof_begin(); + oneof_iterator oneof_end(); + const_oneof_iterator oneof_begin() const; + const_oneof_iterator oneof_end() const; + + FieldAccessor fields() { return FieldAccessor(this); } + ConstFieldAccessor fields() const { return ConstFieldAccessor(this); } + OneofAccessor oneofs() { return OneofAccessor(this); } + ConstOneofAccessor oneofs() const { return ConstOneofAccessor(this); } + + private: + UPB_DISALLOW_POD_OPS(MessageDef, upb::MessageDef) +}; + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Returns NULL if memory allocation failed. */ +upb_msgdef *upb_msgdef_new(const void *owner); + +/* Include upb_refcounted methods like upb_msgdef_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_msgdef, upb_msgdef_upcast2) + +bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status); + +upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner); +const char *upb_msgdef_fullname(const upb_msgdef *m); +const char *upb_msgdef_name(const upb_msgdef *m); +int upb_msgdef_numoneofs(const upb_msgdef *m); +upb_syntax_t upb_msgdef_syntax(const upb_msgdef *m); + +bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor, + upb_status *s); +bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor, + upb_status *s); +bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname, upb_status *s); +void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry); +bool upb_msgdef_mapentry(const upb_msgdef *m); +bool upb_msgdef_setsyntax(upb_msgdef *m, upb_syntax_t syntax); + +/* Field lookup in a couple of different variations: + * - itof = int to field + * - ntof = name to field + * - ntofz = name to field, null-terminated string. */ +const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i); +const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name, + size_t len); +int upb_msgdef_numfields(const upb_msgdef *m); + +UPB_INLINE const upb_fielddef *upb_msgdef_ntofz(const upb_msgdef *m, + const char *name) { + return upb_msgdef_ntof(m, name, strlen(name)); +} + +UPB_INLINE upb_fielddef *upb_msgdef_itof_mutable(upb_msgdef *m, uint32_t i) { + return (upb_fielddef*)upb_msgdef_itof(m, i); +} + +UPB_INLINE upb_fielddef *upb_msgdef_ntof_mutable(upb_msgdef *m, + const char *name, size_t len) { + return (upb_fielddef *)upb_msgdef_ntof(m, name, len); +} + +/* Oneof lookup: + * - ntoo = name to oneof + * - ntooz = name to oneof, null-terminated string. */ +const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name, + size_t len); +int upb_msgdef_numoneofs(const upb_msgdef *m); + +UPB_INLINE const upb_oneofdef *upb_msgdef_ntooz(const upb_msgdef *m, + const char *name) { + return upb_msgdef_ntoo(m, name, strlen(name)); +} + +UPB_INLINE upb_oneofdef *upb_msgdef_ntoo_mutable(upb_msgdef *m, + const char *name, size_t len) { + return (upb_oneofdef *)upb_msgdef_ntoo(m, name, len); +} + +/* Lookup of either field or oneof by name. Returns whether either was found. + * If the return is true, then the found def will be set, and the non-found + * one set to NULL. */ +bool upb_msgdef_lookupname(const upb_msgdef *m, const char *name, size_t len, + const upb_fielddef **f, const upb_oneofdef **o); + +UPB_INLINE bool upb_msgdef_lookupnamez(const upb_msgdef *m, const char *name, + const upb_fielddef **f, + const upb_oneofdef **o) { + return upb_msgdef_lookupname(m, name, strlen(name), f, o); +} + +/* Iteration over fields and oneofs. For example: + * + * upb_msg_field_iter i; + * for(upb_msg_field_begin(&i, m); + * !upb_msg_field_done(&i); + * upb_msg_field_next(&i)) { + * upb_fielddef *f = upb_msg_iter_field(&i); + * // ... + * } + * + * For C we don't have separate iterators for const and non-const. + * It is the caller's responsibility to cast the upb_fielddef* to + * const if the upb_msgdef* is const. */ +void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m); +void upb_msg_field_next(upb_msg_field_iter *iter); +bool upb_msg_field_done(const upb_msg_field_iter *iter); +upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter); +void upb_msg_field_iter_setdone(upb_msg_field_iter *iter); + +/* Similar to above, we also support iterating through the oneofs in a + * msgdef. */ +void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m); +void upb_msg_oneof_next(upb_msg_oneof_iter *iter); +bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter); +upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter); +void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter); + +UPB_END_EXTERN_C + + +/* upb::EnumDef ***************************************************************/ + +typedef upb_strtable_iter upb_enum_iter; + +#ifdef __cplusplus + +/* Class that represents an enum. Its base class is upb::Def (convert with + * upb::upcast()). */ +class upb::EnumDef { + public: + /* Returns NULL if memory allocation failed. */ + static reffed_ptr<EnumDef> New(); + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + /* Functionality from upb::Def. */ + const char* full_name() const; + const char* name() const; + bool set_full_name(const char* fullname, Status* s); + bool set_full_name(const std::string& fullname, Status* s); + + /* Call to freeze this EnumDef. */ + bool Freeze(Status* s); + + /* The value that is used as the default when no field default is specified. + * If not set explicitly, the first value that was added will be used. + * The default value must be a member of the enum. + * Requires that value_count() > 0. */ + int32_t default_value() const; + + /* Sets the default value. If this value is not valid, returns false and an + * error message in status. */ + bool set_default_value(int32_t val, Status* status); + + /* Returns the number of values currently defined in the enum. Note that + * multiple names can refer to the same number, so this may be greater than + * the total number of unique numbers. */ + int value_count() const; + + /* Adds a single name/number pair to the enum. Fails if this name has + * already been used by another value. */ + bool AddValue(const char* name, int32_t num, Status* status); + bool AddValue(const std::string& name, int32_t num, Status* status); + + /* Lookups from name to integer, returning true if found. */ + bool FindValueByName(const char* name, int32_t* num) const; + + /* Finds the name corresponding to the given number, or NULL if none was + * found. If more than one name corresponds to this number, returns the + * first one that was added. */ + const char* FindValueByNumber(int32_t num) const; + + /* Returns a new EnumDef with all the same values. The new EnumDef will be + * owned by the given owner. */ + EnumDef* Dup(const void* owner) const; + + /* Iteration over name/value pairs. The order is undefined. + * Adding an enum val invalidates any iterators. + * + * TODO: make compatible with range-for, with elements as pairs? */ + class Iterator { + public: + explicit Iterator(const EnumDef*); + + int32_t number(); + const char *name(); + bool Done(); + void Next(); + + private: + upb_enum_iter iter_; + }; + + private: + UPB_DISALLOW_POD_OPS(EnumDef, upb::EnumDef) +}; + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ +upb_enumdef *upb_enumdef_new(const void *owner); +upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner); + +/* Include upb_refcounted methods like upb_enumdef_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_enumdef, upb_enumdef_upcast2) + +bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status); + +/* From upb_def. */ +const char *upb_enumdef_fullname(const upb_enumdef *e); +const char *upb_enumdef_name(const upb_enumdef *e); +bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname, + upb_status *s); + +int32_t upb_enumdef_default(const upb_enumdef *e); +bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s); +int upb_enumdef_numvals(const upb_enumdef *e); +bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num, + upb_status *status); + +/* Enum lookups: + * - ntoi: look up a name with specified length. + * - ntoiz: look up a name provided as a null-terminated string. + * - iton: look up an integer, returning the name as a null-terminated + * string. */ +bool upb_enumdef_ntoi(const upb_enumdef *e, const char *name, size_t len, + int32_t *num); +UPB_INLINE bool upb_enumdef_ntoiz(const upb_enumdef *e, + const char *name, int32_t *num) { + return upb_enumdef_ntoi(e, name, strlen(name), num); +} +const char *upb_enumdef_iton(const upb_enumdef *e, int32_t num); + +/* upb_enum_iter i; + * for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) { + * // ... + * } + */ +void upb_enum_begin(upb_enum_iter *iter, const upb_enumdef *e); +void upb_enum_next(upb_enum_iter *iter); +bool upb_enum_done(upb_enum_iter *iter); +const char *upb_enum_iter_name(upb_enum_iter *iter); +int32_t upb_enum_iter_number(upb_enum_iter *iter); + +UPB_END_EXTERN_C + +/* upb::OneofDef **************************************************************/ + +typedef upb_inttable_iter upb_oneof_iter; + +#ifdef __cplusplus + +/* Class that represents a oneof. */ +class upb::OneofDef { + public: + /* Returns NULL if memory allocation failed. */ + static reffed_ptr<OneofDef> New(); + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + /* Returns the MessageDef that owns this OneofDef. */ + const MessageDef* containing_type() const; + + /* Returns the name of this oneof. This is the name used to look up the oneof + * by name once added to a message def. */ + const char* name() const; + bool set_name(const char* name, Status* s); + bool set_name(const std::string& name, Status* s); + + /* Returns the number of fields currently defined in the oneof. */ + int field_count() const; + + /* Adds a field to the oneof. The field must not have been added to any other + * oneof or msgdef. If the oneof is not yet part of a msgdef, then when the + * oneof is eventually added to a msgdef, all fields added to the oneof will + * also be added to the msgdef at that time. If the oneof is already part of a + * msgdef, the field must either be a part of that msgdef already, or must not + * be a part of any msgdef; in the latter case, the field is added to the + * msgdef as a part of this operation. + * + * The field may only have an OPTIONAL label, never REQUIRED or REPEATED. + * + * If |f| is already part of this MessageDef, this method performs no action + * and returns true (success). Thus, this method is idempotent. */ + bool AddField(FieldDef* field, Status* s); + bool AddField(const reffed_ptr<FieldDef>& field, Status* s); + + /* Looks up by name. */ + const FieldDef* FindFieldByName(const char* name, size_t len) const; + FieldDef* FindFieldByName(const char* name, size_t len); + const FieldDef* FindFieldByName(const char* name) const { + return FindFieldByName(name, strlen(name)); + } + FieldDef* FindFieldByName(const char* name) { + return FindFieldByName(name, strlen(name)); + } + + template <class T> + FieldDef* FindFieldByName(const T& str) { + return FindFieldByName(str.c_str(), str.size()); + } + template <class T> + const FieldDef* FindFieldByName(const T& str) const { + return FindFieldByName(str.c_str(), str.size()); + } + + /* Looks up by tag number. */ + const FieldDef* FindFieldByNumber(uint32_t num) const; + + /* Returns a new OneofDef with all the same fields. The OneofDef will be owned + * by the given owner. */ + OneofDef* Dup(const void* owner) const; + + /* Iteration over fields. The order is undefined. */ + class iterator : public std::iterator<std::forward_iterator_tag, FieldDef*> { + public: + explicit iterator(OneofDef* md); + static iterator end(OneofDef* md); + + void operator++(); + FieldDef* operator*() const; + bool operator!=(const iterator& other) const; + bool operator==(const iterator& other) const; + + private: + upb_oneof_iter iter_; + }; + + class const_iterator + : public std::iterator<std::forward_iterator_tag, const FieldDef*> { + public: + explicit const_iterator(const OneofDef* md); + static const_iterator end(const OneofDef* md); + + void operator++(); + const FieldDef* operator*() const; + bool operator!=(const const_iterator& other) const; + bool operator==(const const_iterator& other) const; + + private: + upb_oneof_iter iter_; + }; + + iterator begin(); + iterator end(); + const_iterator begin() const; + const_iterator end() const; + + private: + UPB_DISALLOW_POD_OPS(OneofDef, upb::OneofDef) +}; + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ +upb_oneofdef *upb_oneofdef_new(const void *owner); +upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner); + +/* Include upb_refcounted methods like upb_oneofdef_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_oneofdef, upb_oneofdef_upcast) + +const char *upb_oneofdef_name(const upb_oneofdef *o); +bool upb_oneofdef_setname(upb_oneofdef *o, const char *name, upb_status *s); + +const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o); +int upb_oneofdef_numfields(const upb_oneofdef *o); +bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f, + const void *ref_donor, + upb_status *s); + +/* Oneof lookups: + * - ntof: look up a field by name. + * - ntofz: look up a field by name (as a null-terminated string). + * - itof: look up a field by number. */ +const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o, + const char *name, size_t length); +UPB_INLINE const upb_fielddef *upb_oneofdef_ntofz(const upb_oneofdef *o, + const char *name) { + return upb_oneofdef_ntof(o, name, strlen(name)); +} +const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num); + +/* upb_oneof_iter i; + * for(upb_oneof_begin(&i, e); !upb_oneof_done(&i); upb_oneof_next(&i)) { + * // ... + * } + */ +void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o); +void upb_oneof_next(upb_oneof_iter *iter); +bool upb_oneof_done(upb_oneof_iter *iter); +upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter); +void upb_oneof_iter_setdone(upb_oneof_iter *iter); + +UPB_END_EXTERN_C + + +/* upb::FileDef ***************************************************************/ + +#ifdef __cplusplus + +/* Class that represents a .proto file with some things defined in it. + * + * Many users won't care about FileDefs, but they are necessary if you want to + * read the values of file-level options. */ +class upb::FileDef { + public: + /* Returns NULL if memory allocation failed. */ + static reffed_ptr<FileDef> New(); + + /* upb::RefCounted methods like Ref()/Unref(). */ + UPB_REFCOUNTED_CPPMETHODS + + /* Get/set name of the file (eg. "foo/bar.proto"). */ + const char* name() const; + bool set_name(const char* name, Status* s); + bool set_name(const std::string& name, Status* s); + + /* Package name for definitions inside the file (eg. "foo.bar"). */ + const char* package() const; + bool set_package(const char* package, Status* s); + + /* Syntax for the file. Defaults to proto2. */ + upb_syntax_t syntax() const; + void set_syntax(upb_syntax_t syntax); + + /* Get the list of defs from the file. These are returned in the order that + * they were added to the FileDef. */ + int def_count() const; + const Def* def(int index) const; + Def* def(int index); + + /* Get the list of dependencies from the file. These are returned in the + * order that they were added to the FileDef. */ + int dependency_count() const; + const FileDef* dependency(int index) const; + + /* Adds defs to this file. The def must not already belong to another + * file. + * + * Note: this does *not* ensure that this def's name is unique in this file! + * Use a SymbolTable if you want to check this property. Especially since + * properly checking uniqueness would require a check across *all* files + * (including dependencies). */ + bool AddDef(Def* def, Status* s); + bool AddMessage(MessageDef* m, Status* s); + bool AddEnum(EnumDef* e, Status* s); + bool AddExtension(FieldDef* f, Status* s); + + /* Adds a dependency of this file. */ + bool AddDependency(const FileDef* file); + + /* Freezes this FileDef and all messages/enums under it. All subdefs must be + * resolved and all messages/enums must validate. Returns true if this + * succeeded. + * + * TODO(haberman): should we care whether the file's dependencies are frozen + * already? */ + bool Freeze(Status* s); + + private: + UPB_DISALLOW_POD_OPS(FileDef, upb::FileDef) +}; + +#endif + +UPB_BEGIN_EXTERN_C + +upb_filedef *upb_filedef_new(const void *owner); + +/* Include upb_refcounted methods like upb_msgdef_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_filedef, upb_filedef_upcast) + +const char *upb_filedef_name(const upb_filedef *f); +const char *upb_filedef_package(const upb_filedef *f); +upb_syntax_t upb_filedef_syntax(const upb_filedef *f); +size_t upb_filedef_defcount(const upb_filedef *f); +size_t upb_filedef_depcount(const upb_filedef *f); +const upb_def *upb_filedef_def(const upb_filedef *f, size_t i); +const upb_filedef *upb_filedef_dep(const upb_filedef *f, size_t i); + +bool upb_filedef_freeze(upb_filedef *f, upb_status *s); +bool upb_filedef_setname(upb_filedef *f, const char *name, upb_status *s); +bool upb_filedef_setpackage(upb_filedef *f, const char *package, upb_status *s); +bool upb_filedef_setsyntax(upb_filedef *f, upb_syntax_t syntax, upb_status *s); + +bool upb_filedef_adddef(upb_filedef *f, upb_def *def, const void *ref_donor, + upb_status *s); +bool upb_filedef_adddep(upb_filedef *f, const upb_filedef *dep); + +UPB_INLINE bool upb_filedef_addmsg(upb_filedef *f, upb_msgdef *m, + const void *ref_donor, upb_status *s) { + return upb_filedef_adddef(f, upb_msgdef_upcast_mutable(m), ref_donor, s); +} + +UPB_INLINE bool upb_filedef_addenum(upb_filedef *f, upb_enumdef *e, + const void *ref_donor, upb_status *s) { + return upb_filedef_adddef(f, upb_enumdef_upcast_mutable(e), ref_donor, s); +} + +UPB_INLINE bool upb_filedef_addext(upb_filedef *file, upb_fielddef *f, + const void *ref_donor, upb_status *s) { + return upb_filedef_adddef(file, upb_fielddef_upcast_mutable(f), ref_donor, s); +} +UPB_INLINE upb_def *upb_filedef_mutabledef(upb_filedef *f, int i) { + return (upb_def*)upb_filedef_def(f, i); +} + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +UPB_INLINE const char* upb_safecstr(const std::string& str) { + assert(str.size() == std::strlen(str.c_str())); + return str.c_str(); +} + +/* Inline C++ wrappers. */ +namespace upb { + +inline Def* Def::Dup(const void* owner) const { + return upb_def_dup(this, owner); +} +inline Def::Type Def::def_type() const { return upb_def_type(this); } +inline const char* Def::full_name() const { return upb_def_fullname(this); } +inline const char* Def::name() const { return upb_def_name(this); } +inline bool Def::set_full_name(const char* fullname, Status* s) { + return upb_def_setfullname(this, fullname, s); +} +inline bool Def::set_full_name(const std::string& fullname, Status* s) { + return upb_def_setfullname(this, upb_safecstr(fullname), s); +} +inline bool Def::Freeze(Def* const* defs, size_t n, Status* status) { + return upb_def_freeze(defs, n, status); +} +inline bool Def::Freeze(const std::vector<Def*>& defs, Status* status) { + return upb_def_freeze((Def* const*)&defs[0], defs.size(), status); +} + +inline bool FieldDef::CheckType(int32_t val) { + return upb_fielddef_checktype(val); +} +inline bool FieldDef::CheckLabel(int32_t val) { + return upb_fielddef_checklabel(val); +} +inline bool FieldDef::CheckDescriptorType(int32_t val) { + return upb_fielddef_checkdescriptortype(val); +} +inline bool FieldDef::CheckIntegerFormat(int32_t val) { + return upb_fielddef_checkintfmt(val); +} +inline FieldDef::Type FieldDef::ConvertType(int32_t val) { + assert(CheckType(val)); + return static_cast<FieldDef::Type>(val); +} +inline FieldDef::Label FieldDef::ConvertLabel(int32_t val) { + assert(CheckLabel(val)); + return static_cast<FieldDef::Label>(val); +} +inline FieldDef::DescriptorType FieldDef::ConvertDescriptorType(int32_t val) { + assert(CheckDescriptorType(val)); + return static_cast<FieldDef::DescriptorType>(val); +} +inline FieldDef::IntegerFormat FieldDef::ConvertIntegerFormat(int32_t val) { + assert(CheckIntegerFormat(val)); + return static_cast<FieldDef::IntegerFormat>(val); +} + +inline reffed_ptr<FieldDef> FieldDef::New() { + upb_fielddef *f = upb_fielddef_new(&f); + return reffed_ptr<FieldDef>(f, &f); +} +inline FieldDef* FieldDef::Dup(const void* owner) const { + return upb_fielddef_dup(this, owner); +} +inline const char* FieldDef::full_name() const { + return upb_fielddef_fullname(this); +} +inline bool FieldDef::set_full_name(const char* fullname, Status* s) { + return upb_fielddef_setfullname(this, fullname, s); +} +inline bool FieldDef::set_full_name(const std::string& fullname, Status* s) { + return upb_fielddef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool FieldDef::type_is_set() const { + return upb_fielddef_typeisset(this); +} +inline FieldDef::Type FieldDef::type() const { return upb_fielddef_type(this); } +inline FieldDef::DescriptorType FieldDef::descriptor_type() const { + return upb_fielddef_descriptortype(this); +} +inline FieldDef::Label FieldDef::label() const { + return upb_fielddef_label(this); +} +inline uint32_t FieldDef::number() const { return upb_fielddef_number(this); } +inline const char* FieldDef::name() const { return upb_fielddef_name(this); } +inline bool FieldDef::is_extension() const { + return upb_fielddef_isextension(this); +} +inline size_t FieldDef::GetJsonName(char* buf, size_t len) const { + return upb_fielddef_getjsonname(this, buf, len); +} +inline bool FieldDef::lazy() const { + return upb_fielddef_lazy(this); +} +inline void FieldDef::set_lazy(bool lazy) { + upb_fielddef_setlazy(this, lazy); +} +inline bool FieldDef::packed() const { + return upb_fielddef_packed(this); +} +inline uint32_t FieldDef::index() const { + return upb_fielddef_index(this); +} +inline void FieldDef::set_packed(bool packed) { + upb_fielddef_setpacked(this, packed); +} +inline const MessageDef* FieldDef::containing_type() const { + return upb_fielddef_containingtype(this); +} +inline const OneofDef* FieldDef::containing_oneof() const { + return upb_fielddef_containingoneof(this); +} +inline const char* FieldDef::containing_type_name() { + return upb_fielddef_containingtypename(this); +} +inline bool FieldDef::set_number(uint32_t number, Status* s) { + return upb_fielddef_setnumber(this, number, s); +} +inline bool FieldDef::set_name(const char *name, Status* s) { + return upb_fielddef_setname(this, name, s); +} +inline bool FieldDef::set_name(const std::string& name, Status* s) { + return upb_fielddef_setname(this, upb_safecstr(name), s); +} +inline bool FieldDef::set_json_name(const char *name, Status* s) { + return upb_fielddef_setjsonname(this, name, s); +} +inline bool FieldDef::set_json_name(const std::string& name, Status* s) { + return upb_fielddef_setjsonname(this, upb_safecstr(name), s); +} +inline void FieldDef::clear_json_name() { + upb_fielddef_clearjsonname(this); +} +inline bool FieldDef::set_containing_type_name(const char *name, Status* s) { + return upb_fielddef_setcontainingtypename(this, name, s); +} +inline bool FieldDef::set_containing_type_name(const std::string &name, + Status *s) { + return upb_fielddef_setcontainingtypename(this, upb_safecstr(name), s); +} +inline void FieldDef::set_type(upb_fieldtype_t type) { + upb_fielddef_settype(this, type); +} +inline void FieldDef::set_is_extension(bool is_extension) { + upb_fielddef_setisextension(this, is_extension); +} +inline void FieldDef::set_descriptor_type(FieldDef::DescriptorType type) { + upb_fielddef_setdescriptortype(this, type); +} +inline void FieldDef::set_label(upb_label_t label) { + upb_fielddef_setlabel(this, label); +} +inline bool FieldDef::IsSubMessage() const { + return upb_fielddef_issubmsg(this); +} +inline bool FieldDef::IsString() const { return upb_fielddef_isstring(this); } +inline bool FieldDef::IsSequence() const { return upb_fielddef_isseq(this); } +inline bool FieldDef::IsMap() const { return upb_fielddef_ismap(this); } +inline int64_t FieldDef::default_int64() const { + return upb_fielddef_defaultint64(this); +} +inline int32_t FieldDef::default_int32() const { + return upb_fielddef_defaultint32(this); +} +inline uint64_t FieldDef::default_uint64() const { + return upb_fielddef_defaultuint64(this); +} +inline uint32_t FieldDef::default_uint32() const { + return upb_fielddef_defaultuint32(this); +} +inline bool FieldDef::default_bool() const { + return upb_fielddef_defaultbool(this); +} +inline float FieldDef::default_float() const { + return upb_fielddef_defaultfloat(this); +} +inline double FieldDef::default_double() const { + return upb_fielddef_defaultdouble(this); +} +inline const char* FieldDef::default_string(size_t* len) const { + return upb_fielddef_defaultstr(this, len); +} +inline void FieldDef::set_default_int64(int64_t value) { + upb_fielddef_setdefaultint64(this, value); +} +inline void FieldDef::set_default_int32(int32_t value) { + upb_fielddef_setdefaultint32(this, value); +} +inline void FieldDef::set_default_uint64(uint64_t value) { + upb_fielddef_setdefaultuint64(this, value); +} +inline void FieldDef::set_default_uint32(uint32_t value) { + upb_fielddef_setdefaultuint32(this, value); +} +inline void FieldDef::set_default_bool(bool value) { + upb_fielddef_setdefaultbool(this, value); +} +inline void FieldDef::set_default_float(float value) { + upb_fielddef_setdefaultfloat(this, value); +} +inline void FieldDef::set_default_double(double value) { + upb_fielddef_setdefaultdouble(this, value); +} +inline bool FieldDef::set_default_string(const void *str, size_t len, + Status *s) { + return upb_fielddef_setdefaultstr(this, str, len, s); +} +inline bool FieldDef::set_default_string(const std::string& str, Status* s) { + return upb_fielddef_setdefaultstr(this, str.c_str(), str.size(), s); +} +inline void FieldDef::set_default_cstr(const char* str, Status* s) { + return upb_fielddef_setdefaultcstr(this, str, s); +} +inline bool FieldDef::HasSubDef() const { return upb_fielddef_hassubdef(this); } +inline const Def* FieldDef::subdef() const { return upb_fielddef_subdef(this); } +inline const MessageDef *FieldDef::message_subdef() const { + return upb_fielddef_msgsubdef(this); +} +inline const EnumDef *FieldDef::enum_subdef() const { + return upb_fielddef_enumsubdef(this); +} +inline const char* FieldDef::subdef_name() const { + return upb_fielddef_subdefname(this); +} +inline bool FieldDef::set_subdef(const Def* subdef, Status* s) { + return upb_fielddef_setsubdef(this, subdef, s); +} +inline bool FieldDef::set_enum_subdef(const EnumDef* subdef, Status* s) { + return upb_fielddef_setenumsubdef(this, subdef, s); +} +inline bool FieldDef::set_message_subdef(const MessageDef* subdef, Status* s) { + return upb_fielddef_setmsgsubdef(this, subdef, s); +} +inline bool FieldDef::set_subdef_name(const char* name, Status* s) { + return upb_fielddef_setsubdefname(this, name, s); +} +inline bool FieldDef::set_subdef_name(const std::string& name, Status* s) { + return upb_fielddef_setsubdefname(this, upb_safecstr(name), s); +} + +inline reffed_ptr<MessageDef> MessageDef::New() { + upb_msgdef *m = upb_msgdef_new(&m); + return reffed_ptr<MessageDef>(m, &m); +} +inline const char *MessageDef::full_name() const { + return upb_msgdef_fullname(this); +} +inline const char *MessageDef::name() const { + return upb_msgdef_name(this); +} +inline upb_syntax_t MessageDef::syntax() const { + return upb_msgdef_syntax(this); +} +inline bool MessageDef::set_full_name(const char* fullname, Status* s) { + return upb_msgdef_setfullname(this, fullname, s); +} +inline bool MessageDef::set_full_name(const std::string& fullname, Status* s) { + return upb_msgdef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool MessageDef::set_syntax(upb_syntax_t syntax) { + return upb_msgdef_setsyntax(this, syntax); +} +inline bool MessageDef::Freeze(Status* status) { + return upb_msgdef_freeze(this, status); +} +inline int MessageDef::field_count() const { + return upb_msgdef_numfields(this); +} +inline int MessageDef::oneof_count() const { + return upb_msgdef_numoneofs(this); +} +inline bool MessageDef::AddField(upb_fielddef* f, Status* s) { + return upb_msgdef_addfield(this, f, NULL, s); +} +inline bool MessageDef::AddField(const reffed_ptr<FieldDef>& f, Status* s) { + return upb_msgdef_addfield(this, f.get(), NULL, s); +} +inline bool MessageDef::AddOneof(upb_oneofdef* o, Status* s) { + return upb_msgdef_addoneof(this, o, NULL, s); +} +inline bool MessageDef::AddOneof(const reffed_ptr<OneofDef>& o, Status* s) { + return upb_msgdef_addoneof(this, o.get(), NULL, s); +} +inline FieldDef* MessageDef::FindFieldByNumber(uint32_t number) { + return upb_msgdef_itof_mutable(this, number); +} +inline FieldDef* MessageDef::FindFieldByName(const char* name, size_t len) { + return upb_msgdef_ntof_mutable(this, name, len); +} +inline const FieldDef* MessageDef::FindFieldByNumber(uint32_t number) const { + return upb_msgdef_itof(this, number); +} +inline const FieldDef *MessageDef::FindFieldByName(const char *name, + size_t len) const { + return upb_msgdef_ntof(this, name, len); +} +inline OneofDef* MessageDef::FindOneofByName(const char* name, size_t len) { + return upb_msgdef_ntoo_mutable(this, name, len); +} +inline const OneofDef* MessageDef::FindOneofByName(const char* name, + size_t len) const { + return upb_msgdef_ntoo(this, name, len); +} +inline MessageDef* MessageDef::Dup(const void *owner) const { + return upb_msgdef_dup(this, owner); +} +inline void MessageDef::setmapentry(bool map_entry) { + upb_msgdef_setmapentry(this, map_entry); +} +inline bool MessageDef::mapentry() const { + return upb_msgdef_mapentry(this); +} +inline MessageDef::field_iterator MessageDef::field_begin() { + return field_iterator(this); +} +inline MessageDef::field_iterator MessageDef::field_end() { + return field_iterator::end(this); +} +inline MessageDef::const_field_iterator MessageDef::field_begin() const { + return const_field_iterator(this); +} +inline MessageDef::const_field_iterator MessageDef::field_end() const { + return const_field_iterator::end(this); +} + +inline MessageDef::oneof_iterator MessageDef::oneof_begin() { + return oneof_iterator(this); +} +inline MessageDef::oneof_iterator MessageDef::oneof_end() { + return oneof_iterator::end(this); +} +inline MessageDef::const_oneof_iterator MessageDef::oneof_begin() const { + return const_oneof_iterator(this); +} +inline MessageDef::const_oneof_iterator MessageDef::oneof_end() const { + return const_oneof_iterator::end(this); +} + +inline MessageDef::field_iterator::field_iterator(MessageDef* md) { + upb_msg_field_begin(&iter_, md); +} +inline MessageDef::field_iterator MessageDef::field_iterator::end( + MessageDef* md) { + MessageDef::field_iterator iter(md); + upb_msg_field_iter_setdone(&iter.iter_); + return iter; +} +inline FieldDef* MessageDef::field_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void MessageDef::field_iterator::operator++() { + return upb_msg_field_next(&iter_); +} +inline bool MessageDef::field_iterator::operator==( + const field_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::field_iterator::operator!=( + const field_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::const_field_iterator::const_field_iterator( + const MessageDef* md) { + upb_msg_field_begin(&iter_, md); +} +inline MessageDef::const_field_iterator MessageDef::const_field_iterator::end( + const MessageDef *md) { + MessageDef::const_field_iterator iter(md); + upb_msg_field_iter_setdone(&iter.iter_); + return iter; +} +inline const FieldDef* MessageDef::const_field_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void MessageDef::const_field_iterator::operator++() { + return upb_msg_field_next(&iter_); +} +inline bool MessageDef::const_field_iterator::operator==( + const const_field_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::const_field_iterator::operator!=( + const const_field_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::oneof_iterator::oneof_iterator(MessageDef* md) { + upb_msg_oneof_begin(&iter_, md); +} +inline MessageDef::oneof_iterator MessageDef::oneof_iterator::end( + MessageDef* md) { + MessageDef::oneof_iterator iter(md); + upb_msg_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline OneofDef* MessageDef::oneof_iterator::operator*() const { + return upb_msg_iter_oneof(&iter_); +} +inline void MessageDef::oneof_iterator::operator++() { + return upb_msg_oneof_next(&iter_); +} +inline bool MessageDef::oneof_iterator::operator==( + const oneof_iterator &other) const { + return upb_strtable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::oneof_iterator::operator!=( + const oneof_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::const_oneof_iterator::const_oneof_iterator( + const MessageDef* md) { + upb_msg_oneof_begin(&iter_, md); +} +inline MessageDef::const_oneof_iterator MessageDef::const_oneof_iterator::end( + const MessageDef *md) { + MessageDef::const_oneof_iterator iter(md); + upb_msg_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline const OneofDef* MessageDef::const_oneof_iterator::operator*() const { + return upb_msg_iter_oneof(&iter_); +} +inline void MessageDef::const_oneof_iterator::operator++() { + return upb_msg_oneof_next(&iter_); +} +inline bool MessageDef::const_oneof_iterator::operator==( + const const_oneof_iterator &other) const { + return upb_strtable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::const_oneof_iterator::operator!=( + const const_oneof_iterator &other) const { + return !(*this == other); +} + +inline reffed_ptr<EnumDef> EnumDef::New() { + upb_enumdef *e = upb_enumdef_new(&e); + return reffed_ptr<EnumDef>(e, &e); +} +inline const char* EnumDef::full_name() const { + return upb_enumdef_fullname(this); +} +inline const char* EnumDef::name() const { + return upb_enumdef_name(this); +} +inline bool EnumDef::set_full_name(const char* fullname, Status* s) { + return upb_enumdef_setfullname(this, fullname, s); +} +inline bool EnumDef::set_full_name(const std::string& fullname, Status* s) { + return upb_enumdef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool EnumDef::Freeze(Status* status) { + return upb_enumdef_freeze(this, status); +} +inline int32_t EnumDef::default_value() const { + return upb_enumdef_default(this); +} +inline bool EnumDef::set_default_value(int32_t val, Status* status) { + return upb_enumdef_setdefault(this, val, status); +} +inline int EnumDef::value_count() const { return upb_enumdef_numvals(this); } +inline bool EnumDef::AddValue(const char* name, int32_t num, Status* status) { + return upb_enumdef_addval(this, name, num, status); +} +inline bool EnumDef::AddValue(const std::string& name, int32_t num, + Status* status) { + return upb_enumdef_addval(this, upb_safecstr(name), num, status); +} +inline bool EnumDef::FindValueByName(const char* name, int32_t *num) const { + return upb_enumdef_ntoiz(this, name, num); +} +inline const char* EnumDef::FindValueByNumber(int32_t num) const { + return upb_enumdef_iton(this, num); +} +inline EnumDef* EnumDef::Dup(const void* owner) const { + return upb_enumdef_dup(this, owner); +} + +inline EnumDef::Iterator::Iterator(const EnumDef* e) { + upb_enum_begin(&iter_, e); +} +inline int32_t EnumDef::Iterator::number() { + return upb_enum_iter_number(&iter_); +} +inline const char* EnumDef::Iterator::name() { + return upb_enum_iter_name(&iter_); +} +inline bool EnumDef::Iterator::Done() { return upb_enum_done(&iter_); } +inline void EnumDef::Iterator::Next() { return upb_enum_next(&iter_); } + +inline reffed_ptr<OneofDef> OneofDef::New() { + upb_oneofdef *o = upb_oneofdef_new(&o); + return reffed_ptr<OneofDef>(o, &o); +} + +inline const MessageDef* OneofDef::containing_type() const { + return upb_oneofdef_containingtype(this); +} +inline const char* OneofDef::name() const { + return upb_oneofdef_name(this); +} +inline bool OneofDef::set_name(const char* name, Status* s) { + return upb_oneofdef_setname(this, name, s); +} +inline bool OneofDef::set_name(const std::string& name, Status* s) { + return upb_oneofdef_setname(this, upb_safecstr(name), s); +} +inline int OneofDef::field_count() const { + return upb_oneofdef_numfields(this); +} +inline bool OneofDef::AddField(FieldDef* field, Status* s) { + return upb_oneofdef_addfield(this, field, NULL, s); +} +inline bool OneofDef::AddField(const reffed_ptr<FieldDef>& field, Status* s) { + return upb_oneofdef_addfield(this, field.get(), NULL, s); +} +inline const FieldDef* OneofDef::FindFieldByName(const char* name, + size_t len) const { + return upb_oneofdef_ntof(this, name, len); +} +inline const FieldDef* OneofDef::FindFieldByNumber(uint32_t num) const { + return upb_oneofdef_itof(this, num); +} +inline OneofDef::iterator OneofDef::begin() { return iterator(this); } +inline OneofDef::iterator OneofDef::end() { return iterator::end(this); } +inline OneofDef::const_iterator OneofDef::begin() const { + return const_iterator(this); +} +inline OneofDef::const_iterator OneofDef::end() const { + return const_iterator::end(this); +} + +inline OneofDef::iterator::iterator(OneofDef* o) { + upb_oneof_begin(&iter_, o); +} +inline OneofDef::iterator OneofDef::iterator::end(OneofDef* o) { + OneofDef::iterator iter(o); + upb_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline FieldDef* OneofDef::iterator::operator*() const { + return upb_oneof_iter_field(&iter_); +} +inline void OneofDef::iterator::operator++() { return upb_oneof_next(&iter_); } +inline bool OneofDef::iterator::operator==(const iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool OneofDef::iterator::operator!=(const iterator &other) const { + return !(*this == other); +} + +inline OneofDef::const_iterator::const_iterator(const OneofDef* md) { + upb_oneof_begin(&iter_, md); +} +inline OneofDef::const_iterator OneofDef::const_iterator::end( + const OneofDef *md) { + OneofDef::const_iterator iter(md); + upb_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline const FieldDef* OneofDef::const_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void OneofDef::const_iterator::operator++() { + return upb_oneof_next(&iter_); +} +inline bool OneofDef::const_iterator::operator==( + const const_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool OneofDef::const_iterator::operator!=( + const const_iterator &other) const { + return !(*this == other); +} + +inline reffed_ptr<FileDef> FileDef::New() { + upb_filedef *f = upb_filedef_new(&f); + return reffed_ptr<FileDef>(f, &f); +} + +inline const char* FileDef::name() const { + return upb_filedef_name(this); +} +inline bool FileDef::set_name(const char* name, Status* s) { + return upb_filedef_setname(this, name, s); +} +inline bool FileDef::set_name(const std::string& name, Status* s) { + return upb_filedef_setname(this, upb_safecstr(name), s); +} +inline const char* FileDef::package() const { + return upb_filedef_package(this); +} +inline bool FileDef::set_package(const char* package, Status* s) { + return upb_filedef_setpackage(this, package, s); +} +inline int FileDef::def_count() const { + return upb_filedef_defcount(this); +} +inline const Def* FileDef::def(int index) const { + return upb_filedef_def(this, index); +} +inline Def* FileDef::def(int index) { + return const_cast<Def*>(upb_filedef_def(this, index)); +} +inline int FileDef::dependency_count() const { + return upb_filedef_depcount(this); +} +inline const FileDef* FileDef::dependency(int index) const { + return upb_filedef_dep(this, index); +} +inline bool FileDef::AddDef(Def* def, Status* s) { + return upb_filedef_adddef(this, def, NULL, s); +} +inline bool FileDef::AddMessage(MessageDef* m, Status* s) { + return upb_filedef_addmsg(this, m, NULL, s); +} +inline bool FileDef::AddEnum(EnumDef* e, Status* s) { + return upb_filedef_addenum(this, e, NULL, s); +} +inline bool FileDef::AddExtension(FieldDef* f, Status* s) { + return upb_filedef_addext(this, f, NULL, s); +} +inline bool FileDef::AddDependency(const FileDef* file) { + return upb_filedef_adddep(this, file); +} + +} /* namespace upb */ +#endif + +#endif /* UPB_DEF_H_ */ +/* +** This file contains definitions of structs that should be considered private +** and NOT stable across versions of upb. +** +** The only reason they are declared here and not in .c files is to allow upb +** and the application (if desired) to embed statically-initialized instances +** of structures like defs. +** +** If you include this file, all guarantees of ABI compatibility go out the +** window! Any code that includes this file needs to recompile against the +** exact same version of upb that they are linking against. +** +** You also need to recompile if you change the value of the UPB_DEBUG_REFS +** flag. +*/ + + +#ifndef UPB_STATICINIT_H_ +#define UPB_STATICINIT_H_ + +#ifdef __cplusplus +/* Because of how we do our typedefs, this header can't be included from C++. */ +#error This file cannot be included from C++ +#endif + +/* upb_refcounted *************************************************************/ + + +/* upb_def ********************************************************************/ + +struct upb_def { + upb_refcounted base; + + const char *fullname; + const upb_filedef* file; + char type; /* A upb_deftype_t (char to save space) */ + + /* Used as a flag during the def's mutable stage. Must be false unless + * it is currently being used by a function on the stack. This allows + * us to easily determine which defs were passed into the function's + * current invocation. */ + bool came_from_user; +}; + +#define UPB_DEF_INIT(name, type, vtbl, refs, ref2s) \ + { UPB_REFCOUNT_INIT(vtbl, refs, ref2s), name, NULL, type, false } + + +/* upb_fielddef ***************************************************************/ + +struct upb_fielddef { + upb_def base; + + union { + int64_t sint; + uint64_t uint; + double dbl; + float flt; + void *bytes; + } defaultval; + union { + const upb_msgdef *def; /* If !msg_is_symbolic. */ + char *name; /* If msg_is_symbolic. */ + } msg; + union { + const upb_def *def; /* If !subdef_is_symbolic. */ + char *name; /* If subdef_is_symbolic. */ + } sub; /* The msgdef or enumdef for this field, if upb_hassubdef(f). */ + bool subdef_is_symbolic; + bool msg_is_symbolic; + const upb_oneofdef *oneof; + bool default_is_string; + bool type_is_set_; /* False until type is explicitly set. */ + bool is_extension_; + bool lazy_; + bool packed_; + upb_intfmt_t intfmt; + bool tagdelim; + upb_fieldtype_t type_; + upb_label_t label_; + uint32_t number_; + uint32_t selector_base; /* Used to index into a upb::Handlers table. */ + uint32_t index_; +}; + +extern const struct upb_refcounted_vtbl upb_fielddef_vtbl; + +#define UPB_FIELDDEF_INIT(label, type, intfmt, tagdelim, is_extension, lazy, \ + packed, name, num, msgdef, subdef, selector_base, \ + index, defaultval, refs, ref2s) \ + { \ + UPB_DEF_INIT(name, UPB_DEF_FIELD, &upb_fielddef_vtbl, refs, ref2s), \ + defaultval, {msgdef}, {subdef}, NULL, false, false, \ + type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES, true, is_extension, \ + lazy, packed, intfmt, tagdelim, type, label, num, selector_base, index \ + } + + +/* upb_msgdef *****************************************************************/ + +struct upb_msgdef { + upb_def base; + + size_t selector_count; + uint32_t submsg_field_count; + + /* Tables for looking up fields by number and name. */ + upb_inttable itof; /* int to field */ + upb_strtable ntof; /* name to field/oneof */ + + /* Is this a map-entry message? */ + bool map_entry; + + /* Whether this message has proto2 or proto3 semantics. */ + upb_syntax_t syntax; + + /* TODO(haberman): proper extension ranges (there can be multiple). */ +}; + +extern const struct upb_refcounted_vtbl upb_msgdef_vtbl; + +/* TODO: also support static initialization of the oneofs table. This will be + * needed if we compile in descriptors that contain oneofs. */ +#define UPB_MSGDEF_INIT(name, selector_count, submsg_field_count, itof, ntof, \ + map_entry, syntax, refs, ref2s) \ + { \ + UPB_DEF_INIT(name, UPB_DEF_MSG, &upb_fielddef_vtbl, refs, ref2s), \ + selector_count, submsg_field_count, itof, ntof, map_entry, syntax \ + } + + +/* upb_enumdef ****************************************************************/ + +struct upb_enumdef { + upb_def base; + + upb_strtable ntoi; + upb_inttable iton; + int32_t defaultval; +}; + +extern const struct upb_refcounted_vtbl upb_enumdef_vtbl; + +#define UPB_ENUMDEF_INIT(name, ntoi, iton, defaultval, refs, ref2s) \ + { UPB_DEF_INIT(name, UPB_DEF_ENUM, &upb_enumdef_vtbl, refs, ref2s), ntoi, \ + iton, defaultval } + + +/* upb_oneofdef ***************************************************************/ + +struct upb_oneofdef { + upb_refcounted base; + + const char *name; + upb_strtable ntof; + upb_inttable itof; + const upb_msgdef *parent; +}; + +extern const struct upb_refcounted_vtbl upb_oneofdef_vtbl; + +#define UPB_ONEOFDEF_INIT(name, ntof, itof, refs, ref2s) \ + { UPB_REFCOUNT_INIT(&upb_oneofdef_vtbl, refs, ref2s), name, ntof, itof } + + +/* upb_symtab *****************************************************************/ + +struct upb_symtab { + upb_refcounted base; + + upb_strtable symtab; +}; + +struct upb_filedef { + upb_refcounted base; + + const char *name; + const char *package; + upb_syntax_t syntax; + + upb_inttable defs; + upb_inttable deps; +}; + +extern const struct upb_refcounted_vtbl upb_filedef_vtbl; + +#endif /* UPB_STATICINIT_H_ */ +/* +** upb::Handlers (upb_handlers) +** +** A upb_handlers is like a virtual table for a upb_msgdef. Each field of the +** message can have associated functions that will be called when we are +** parsing or visiting a stream of data. This is similar to how handlers work +** in SAX (the Simple API for XML). +** +** The handlers have no idea where the data is coming from, so a single set of +** handlers could be used with two completely different data sources (for +** example, a parser and a visitor over in-memory objects). This decoupling is +** the most important feature of upb, because it allows parsers and serializers +** to be highly reusable. +** +** This is a mixed C/C++ interface that offers a full API to both languages. +** See the top-level README for more information. +*/ + +#ifndef UPB_HANDLERS_H +#define UPB_HANDLERS_H + + +#ifdef __cplusplus +namespace upb { +class BufferHandle; +class BytesHandler; +class HandlerAttributes; +class Handlers; +template <class T> class Handler; +template <class T> struct CanonicalType; +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::BufferHandle, upb_bufhandle) +UPB_DECLARE_TYPE(upb::BytesHandler, upb_byteshandler) +UPB_DECLARE_TYPE(upb::HandlerAttributes, upb_handlerattr) +UPB_DECLARE_DERIVED_TYPE(upb::Handlers, upb::RefCounted, + upb_handlers, upb_refcounted) + +/* The maximum depth that the handler graph can have. This is a resource limit + * for the C stack since we sometimes need to recursively traverse the graph. + * Cycles are ok; the traversal will stop when it detects a cycle, but we must + * hit the cycle before the maximum depth is reached. + * + * If having a single static limit is too inflexible, we can add another variant + * of Handlers::Freeze that allows specifying this as a parameter. */ +#define UPB_MAX_HANDLER_DEPTH 64 + +/* All the different types of handlers that can be registered. + * Only needed for the advanced functions in upb::Handlers. */ +typedef enum { + UPB_HANDLER_INT32, + UPB_HANDLER_INT64, + UPB_HANDLER_UINT32, + UPB_HANDLER_UINT64, + UPB_HANDLER_FLOAT, + UPB_HANDLER_DOUBLE, + UPB_HANDLER_BOOL, + UPB_HANDLER_STARTSTR, + UPB_HANDLER_STRING, + UPB_HANDLER_ENDSTR, + UPB_HANDLER_STARTSUBMSG, + UPB_HANDLER_ENDSUBMSG, + UPB_HANDLER_STARTSEQ, + UPB_HANDLER_ENDSEQ +} upb_handlertype_t; + +#define UPB_HANDLER_MAX (UPB_HANDLER_ENDSEQ+1) + +#define UPB_BREAK NULL + +/* A convenient definition for when no closure is needed. */ +extern char _upb_noclosure; +#define UPB_NO_CLOSURE &_upb_noclosure + +/* A selector refers to a specific field handler in the Handlers object + * (for example: the STARTSUBMSG handler for field "field15"). */ +typedef int32_t upb_selector_t; + +UPB_BEGIN_EXTERN_C + +/* Forward-declares for C inline accessors. We need to declare these here + * so we can "friend" them in the class declarations in C++. */ +UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h, + upb_selector_t s); +UPB_INLINE const void *upb_handlerattr_handlerdata(const upb_handlerattr *attr); +UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h, + upb_selector_t s); + +UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h); +UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj, + const void *type); +UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf, + size_t ofs); +UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h); +UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h); +UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h); + +UPB_END_EXTERN_C + + +/* Static selectors for upb::Handlers. */ +#define UPB_STARTMSG_SELECTOR 0 +#define UPB_ENDMSG_SELECTOR 1 +#define UPB_STATIC_SELECTOR_COUNT 2 + +/* Static selectors for upb::BytesHandler. */ +#define UPB_STARTSTR_SELECTOR 0 +#define UPB_STRING_SELECTOR 1 +#define UPB_ENDSTR_SELECTOR 2 + +typedef void upb_handlerfree(void *d); + +#ifdef __cplusplus + +/* A set of attributes that accompanies a handler's function pointer. */ +class upb::HandlerAttributes { + public: + HandlerAttributes(); + ~HandlerAttributes(); + + /* Sets the handler data that will be passed as the second parameter of the + * handler. To free this pointer when the handlers are freed, call + * Handlers::AddCleanup(). */ + bool SetHandlerData(const void *handler_data); + const void* handler_data() const; + + /* Use this to specify the type of the closure. This will be checked against + * all other closure types for handler that use the same closure. + * Registration will fail if this does not match all other non-NULL closure + * types. */ + bool SetClosureType(const void *closure_type); + const void* closure_type() const; + + /* Use this to specify the type of the returned closure. Only used for + * Start*{String,SubMessage,Sequence} handlers. This must match the closure + * type of any handlers that use it (for example, the StringBuf handler must + * match the closure returned from StartString). */ + bool SetReturnClosureType(const void *return_closure_type); + const void* return_closure_type() const; + + /* Set to indicate that the handler always returns "ok" (either "true" or a + * non-NULL closure). This is a hint that can allow code generators to + * generate more efficient code. */ + bool SetAlwaysOk(bool always_ok); + bool always_ok() const; + + private: + friend UPB_INLINE const void * ::upb_handlerattr_handlerdata( + const upb_handlerattr *attr); +#else +struct upb_handlerattr { +#endif + const void *handler_data_; + const void *closure_type_; + const void *return_closure_type_; + bool alwaysok_; +}; + +#define UPB_HANDLERATTR_INITIALIZER {NULL, NULL, NULL, false} + +typedef struct { + upb_func *func; + + /* It is wasteful to include the entire attributes here: + * + * * Some of the information is redundant (like storing the closure type + * separately for each handler that must match). + * * Some of the info is only needed prior to freeze() (like closure types). + * * alignment padding wastes a lot of space for alwaysok_. + * + * If/when the size and locality of handlers is an issue, we can optimize this + * not to store the entire attr like this. We do not expose the table's + * layout to allow this optimization in the future. */ + upb_handlerattr attr; +} upb_handlers_tabent; + +#ifdef __cplusplus + +/* Extra information about a buffer that is passed to a StringBuf handler. + * TODO(haberman): allow the handle to be pinned so that it will outlive + * the handler invocation. */ +class upb::BufferHandle { + public: + BufferHandle(); + ~BufferHandle(); + + /* The beginning of the buffer. This may be different than the pointer + * passed to a StringBuf handler because the handler may receive data + * that is from the middle or end of a larger buffer. */ + const char* buffer() const; + + /* The offset within the attached object where this buffer begins. Only + * meaningful if there is an attached object. */ + size_t object_offset() const; + + /* Note that object_offset is the offset of "buf" within the attached + * object. */ + void SetBuffer(const char* buf, size_t object_offset); + + /* The BufferHandle can have an "attached object", which can be used to + * tunnel through a pointer to the buffer's underlying representation. */ + template <class T> + void SetAttachedObject(const T* obj); + + /* Returns NULL if the attached object is not of this type. */ + template <class T> + const T* GetAttachedObject() const; + + private: + friend UPB_INLINE void ::upb_bufhandle_init(upb_bufhandle *h); + friend UPB_INLINE void ::upb_bufhandle_setobj(upb_bufhandle *h, + const void *obj, + const void *type); + friend UPB_INLINE void ::upb_bufhandle_setbuf(upb_bufhandle *h, + const char *buf, size_t ofs); + friend UPB_INLINE const void* ::upb_bufhandle_obj(const upb_bufhandle *h); + friend UPB_INLINE const void* ::upb_bufhandle_objtype( + const upb_bufhandle *h); + friend UPB_INLINE const char* ::upb_bufhandle_buf(const upb_bufhandle *h); +#else +struct upb_bufhandle { +#endif + const char *buf_; + const void *obj_; + const void *objtype_; + size_t objofs_; +}; + +#ifdef __cplusplus + +/* A upb::Handlers object represents the set of handlers associated with a + * message in the graph of messages. You can think of it as a big virtual + * table with functions corresponding to all the events that can fire while + * parsing or visiting a message of a specific type. + * + * Any handlers that are not set behave as if they had successfully consumed + * the value. Any unset Start* handlers will propagate their closure to the + * inner frame. + * + * The easiest way to create the *Handler objects needed by the Set* methods is + * with the UpbBind() and UpbMakeHandler() macros; see below. */ +class upb::Handlers { + public: + typedef upb_selector_t Selector; + typedef upb_handlertype_t Type; + + typedef Handler<void *(*)(void *, const void *)> StartFieldHandler; + typedef Handler<bool (*)(void *, const void *)> EndFieldHandler; + typedef Handler<bool (*)(void *, const void *)> StartMessageHandler; + typedef Handler<bool (*)(void *, const void *, Status*)> EndMessageHandler; + typedef Handler<void *(*)(void *, const void *, size_t)> StartStringHandler; + typedef Handler<size_t (*)(void *, const void *, const char *, size_t, + const BufferHandle *)> StringHandler; + + template <class T> struct ValueHandler { + typedef Handler<bool(*)(void *, const void *, T)> H; + }; + + typedef ValueHandler<int32_t>::H Int32Handler; + typedef ValueHandler<int64_t>::H Int64Handler; + typedef ValueHandler<uint32_t>::H UInt32Handler; + typedef ValueHandler<uint64_t>::H UInt64Handler; + typedef ValueHandler<float>::H FloatHandler; + typedef ValueHandler<double>::H DoubleHandler; + typedef ValueHandler<bool>::H BoolHandler; + + /* Any function pointer can be converted to this and converted back to its + * correct type. */ + typedef void GenericFunction(); + + typedef void HandlersCallback(const void *closure, upb_handlers *h); + + /* Returns a new handlers object for the given frozen msgdef. + * Returns NULL if memory allocation failed. */ + static reffed_ptr<Handlers> New(const MessageDef *m); + + /* Convenience function for registering a graph of handlers that mirrors the + * graph of msgdefs for some message. For "m" and all its children a new set + * of handlers will be created and the given callback will be invoked, + * allowing the client to register handlers for this message. Note that any + * subhandlers set by the callback will be overwritten. */ + static reffed_ptr<const Handlers> NewFrozen(const MessageDef *m, + HandlersCallback *callback, + const void *closure); + + /* Functionality from upb::RefCounted. */ + UPB_REFCOUNTED_CPPMETHODS + + /* All handler registration functions return bool to indicate success or + * failure; details about failures are stored in this status object. If a + * failure does occur, it must be cleared before the Handlers are frozen, + * otherwise the freeze() operation will fail. The functions may *only* be + * used while the Handlers are mutable. */ + const Status* status(); + void ClearError(); + + /* Call to freeze these Handlers. Requires that any SubHandlers are already + * frozen. For cycles, you must use the static version below and freeze the + * whole graph at once. */ + bool Freeze(Status* s); + + /* Freezes the given set of handlers. You may not freeze a handler without + * also freezing any handlers they point to. */ + static bool Freeze(Handlers*const* handlers, int n, Status* s); + static bool Freeze(const std::vector<Handlers*>& handlers, Status* s); + + /* Returns the msgdef associated with this handlers object. */ + const MessageDef* message_def() const; + + /* Adds the given pointer and function to the list of cleanup functions that + * will be run when these handlers are freed. If this pointer has previously + * been registered, the function returns false and does nothing. */ + bool AddCleanup(void *ptr, upb_handlerfree *cleanup); + + /* Sets the startmsg handler for the message, which is defined as follows: + * + * bool startmsg(MyType* closure) { + * // Called when the message begins. Returns true if processing should + * // continue. + * return true; + * } + */ + bool SetStartMessageHandler(const StartMessageHandler& handler); + + /* Sets the endmsg handler for the message, which is defined as follows: + * + * bool endmsg(MyType* closure, upb_status *status) { + * // Called when processing of this message ends, whether in success or + * // failure. "status" indicates the final status of processing, and + * // can also be modified in-place to update the final status. + * } + */ + bool SetEndMessageHandler(const EndMessageHandler& handler); + + /* Sets the value handler for the given field, which is defined as follows + * (this is for an int32 field; other field types will pass their native + * C/C++ type for "val"): + * + * bool OnValue(MyClosure* c, const MyHandlerData* d, int32_t val) { + * // Called when the field's value is encountered. "d" contains + * // whatever data was bound to this field when it was registered. + * // Returns true if processing should continue. + * return true; + * } + * + * handers->SetInt32Handler(f, UpbBind(OnValue, new MyHandlerData(...))); + * + * The value type must exactly match f->type(). + * For example, a handler that takes an int32_t parameter may only be used for + * fields of type UPB_TYPE_INT32 and UPB_TYPE_ENUM. + * + * Returns false if the handler failed to register; in this case the cleanup + * handler (if any) will be called immediately. + */ + bool SetInt32Handler (const FieldDef* f, const Int32Handler& h); + bool SetInt64Handler (const FieldDef* f, const Int64Handler& h); + bool SetUInt32Handler(const FieldDef* f, const UInt32Handler& h); + bool SetUInt64Handler(const FieldDef* f, const UInt64Handler& h); + bool SetFloatHandler (const FieldDef* f, const FloatHandler& h); + bool SetDoubleHandler(const FieldDef* f, const DoubleHandler& h); + bool SetBoolHandler (const FieldDef* f, const BoolHandler& h); + + /* Like the previous, but templated on the type on the value (ie. int32). + * This is mostly useful to call from other templates. To call this you must + * specify the template parameter explicitly, ie: + * h->SetValueHandler<T>(f, UpbBind(MyHandler<T>, MyData)); */ + template <class T> + bool SetValueHandler( + const FieldDef *f, + const typename ValueHandler<typename CanonicalType<T>::Type>::H& handler); + + /* Sets handlers for a string field, which are defined as follows: + * + * MySubClosure* startstr(MyClosure* c, const MyHandlerData* d, + * size_t size_hint) { + * // Called when a string value begins. The return value indicates the + * // closure for the string. "size_hint" indicates the size of the + * // string if it is known, however if the string is length-delimited + * // and the end-of-string is not available size_hint will be zero. + * // This case is indistinguishable from the case where the size is + * // known to be zero. + * // + * // TODO(haberman): is it important to distinguish these cases? + * // If we had ssize_t as a type we could make -1 "unknown", but + * // ssize_t is POSIX (not ANSI) and therefore less portable. + * // In practice I suspect it won't be important to distinguish. + * return closure; + * } + * + * size_t str(MyClosure* closure, const MyHandlerData* d, + * const char *str, size_t len) { + * // Called for each buffer of string data; the multiple physical buffers + * // are all part of the same logical string. The return value indicates + * // how many bytes were consumed. If this number is less than "len", + * // this will also indicate that processing should be halted for now, + * // like returning false or UPB_BREAK from any other callback. If + * // number is greater than "len", the excess bytes will be skipped over + * // and not passed to the callback. + * return len; + * } + * + * bool endstr(MyClosure* c, const MyHandlerData* d) { + * // Called when a string value ends. Return value indicates whether + * // processing should continue. + * return true; + * } + */ + bool SetStartStringHandler(const FieldDef* f, const StartStringHandler& h); + bool SetStringHandler(const FieldDef* f, const StringHandler& h); + bool SetEndStringHandler(const FieldDef* f, const EndFieldHandler& h); + + /* Sets the startseq handler, which is defined as follows: + * + * MySubClosure *startseq(MyClosure* c, const MyHandlerData* d) { + * // Called when a sequence (repeated field) begins. The returned + * // pointer indicates the closure for the sequence (or UPB_BREAK + * // to interrupt processing). + * return closure; + * } + * + * h->SetStartSequenceHandler(f, UpbBind(startseq, new MyHandlerData(...))); + * + * Returns "false" if "f" does not belong to this message or is not a + * repeated field. + */ + bool SetStartSequenceHandler(const FieldDef* f, const StartFieldHandler& h); + + /* Sets the startsubmsg handler for the given field, which is defined as + * follows: + * + * MySubClosure* startsubmsg(MyClosure* c, const MyHandlerData* d) { + * // Called when a submessage begins. The returned pointer indicates the + * // closure for the sequence (or UPB_BREAK to interrupt processing). + * return closure; + * } + * + * h->SetStartSubMessageHandler(f, UpbBind(startsubmsg, + * new MyHandlerData(...))); + * + * Returns "false" if "f" does not belong to this message or is not a + * submessage/group field. + */ + bool SetStartSubMessageHandler(const FieldDef* f, const StartFieldHandler& h); + + /* Sets the endsubmsg handler for the given field, which is defined as + * follows: + * + * bool endsubmsg(MyClosure* c, const MyHandlerData* d) { + * // Called when a submessage ends. Returns true to continue processing. + * return true; + * } + * + * Returns "false" if "f" does not belong to this message or is not a + * submessage/group field. + */ + bool SetEndSubMessageHandler(const FieldDef *f, const EndFieldHandler &h); + + /* Starts the endsubseq handler for the given field, which is defined as + * follows: + * + * bool endseq(MyClosure* c, const MyHandlerData* d) { + * // Called when a sequence ends. Returns true continue processing. + * return true; + * } + * + * Returns "false" if "f" does not belong to this message or is not a + * repeated field. + */ + bool SetEndSequenceHandler(const FieldDef* f, const EndFieldHandler& h); + + /* Sets or gets the object that specifies handlers for the given field, which + * must be a submessage or group. Returns NULL if no handlers are set. */ + bool SetSubHandlers(const FieldDef* f, const Handlers* sub); + const Handlers* GetSubHandlers(const FieldDef* f) const; + + /* Equivalent to GetSubHandlers, but takes the STARTSUBMSG selector for the + * field. */ + const Handlers* GetSubHandlers(Selector startsubmsg) const; + + /* A selector refers to a specific field handler in the Handlers object + * (for example: the STARTSUBMSG handler for field "field15"). + * On success, returns true and stores the selector in "s". + * If the FieldDef or Type are invalid, returns false. + * The returned selector is ONLY valid for Handlers whose MessageDef + * contains this FieldDef. */ + static bool GetSelector(const FieldDef* f, Type type, Selector* s); + + /* Given a START selector of any kind, returns the corresponding END selector. */ + static Selector GetEndSelector(Selector start_selector); + + /* Returns the function pointer for this handler. It is the client's + * responsibility to cast to the correct function type before calling it. */ + GenericFunction* GetHandler(Selector selector); + + /* Sets the given attributes to the attributes for this selector. */ + bool GetAttributes(Selector selector, HandlerAttributes* attr); + + /* Returns the handler data that was registered with this handler. */ + const void* GetHandlerData(Selector selector); + + /* Could add any of the following functions as-needed, with some minor + * implementation changes: + * + * const FieldDef* GetFieldDef(Selector selector); + * static bool IsSequence(Selector selector); */ + + private: + UPB_DISALLOW_POD_OPS(Handlers, upb::Handlers) + + friend UPB_INLINE GenericFunction *::upb_handlers_gethandler( + const upb_handlers *h, upb_selector_t s); + friend UPB_INLINE const void *::upb_handlers_gethandlerdata( + const upb_handlers *h, upb_selector_t s); +#else +struct upb_handlers { +#endif + upb_refcounted base; + + const upb_msgdef *msg; + const upb_handlers **sub; + const void *top_closure_type; + upb_inttable cleanup_; + upb_status status_; /* Used only when mutable. */ + upb_handlers_tabent table[1]; /* Dynamically-sized field handler array. */ +}; + +#ifdef __cplusplus + +namespace upb { + +/* Convenience macros for creating a Handler object that is wrapped with a + * type-safe wrapper function that converts the "void*" parameters/returns + * of the underlying C API into nice C++ function. + * + * Sample usage: + * void OnValue1(MyClosure* c, const MyHandlerData* d, int32_t val) { + * // do stuff ... + * } + * + * // Handler that doesn't need any data bound to it. + * void OnValue2(MyClosure* c, int32_t val) { + * // do stuff ... + * } + * + * // Handler that returns bool so it can return failure if necessary. + * bool OnValue3(MyClosure* c, int32_t val) { + * // do stuff ... + * return ok; + * } + * + * // Member function handler. + * class MyClosure { + * public: + * void OnValue(int32_t val) { + * // do stuff ... + * } + * }; + * + * // Takes ownership of the MyHandlerData. + * handlers->SetInt32Handler(f1, UpbBind(OnValue1, new MyHandlerData(...))); + * handlers->SetInt32Handler(f2, UpbMakeHandler(OnValue2)); + * handlers->SetInt32Handler(f1, UpbMakeHandler(OnValue3)); + * handlers->SetInt32Handler(f2, UpbMakeHandler(&MyClosure::OnValue)); + */ + +#ifdef UPB_CXX11 + +/* In C++11, the "template" disambiguator can appear even outside templates, + * so all calls can safely use this pair of macros. */ + +#define UpbMakeHandler(f) upb::MatchFunc(f).template GetFunc<f>() + +/* We have to be careful to only evaluate "d" once. */ +#define UpbBind(f, d) upb::MatchFunc(f).template GetFunc<f>((d)) + +#else + +/* Prior to C++11, the "template" disambiguator may only appear inside a + * template, so the regular macro must not use "template" */ + +#define UpbMakeHandler(f) upb::MatchFunc(f).GetFunc<f>() + +#define UpbBind(f, d) upb::MatchFunc(f).GetFunc<f>((d)) + +#endif /* UPB_CXX11 */ + +/* This macro must be used in C++98 for calls from inside a template. But we + * define this variant in all cases; code that wants to be compatible with both + * C++98 and C++11 should always use this macro when calling from a template. */ +#define UpbMakeHandlerT(f) upb::MatchFunc(f).template GetFunc<f>() + +/* We have to be careful to only evaluate "d" once. */ +#define UpbBindT(f, d) upb::MatchFunc(f).template GetFunc<f>((d)) + +/* Handler: a struct that contains the (handler, data, deleter) tuple that is + * used to register all handlers. Users can Make() these directly but it's + * more convenient to use the UpbMakeHandler/UpbBind macros above. */ +template <class T> class Handler { + public: + /* The underlying, handler function signature that upb uses internally. */ + typedef T FuncPtr; + + /* Intentionally implicit. */ + template <class F> Handler(F func); + ~Handler(); + + private: + void AddCleanup(Handlers* h) const { + if (cleanup_func_) { + bool ok = h->AddCleanup(cleanup_data_, cleanup_func_); + UPB_ASSERT_VAR(ok, ok); + } + } + + UPB_DISALLOW_COPY_AND_ASSIGN(Handler) + friend class Handlers; + FuncPtr handler_; + mutable HandlerAttributes attr_; + mutable bool registered_; + void *cleanup_data_; + upb_handlerfree *cleanup_func_; +}; + +} /* namespace upb */ + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ + +/* Handler function typedefs. */ +typedef bool upb_startmsg_handlerfunc(void *c, const void*); +typedef bool upb_endmsg_handlerfunc(void *c, const void *, upb_status *status); +typedef void* upb_startfield_handlerfunc(void *c, const void *hd); +typedef bool upb_endfield_handlerfunc(void *c, const void *hd); +typedef bool upb_int32_handlerfunc(void *c, const void *hd, int32_t val); +typedef bool upb_int64_handlerfunc(void *c, const void *hd, int64_t val); +typedef bool upb_uint32_handlerfunc(void *c, const void *hd, uint32_t val); +typedef bool upb_uint64_handlerfunc(void *c, const void *hd, uint64_t val); +typedef bool upb_float_handlerfunc(void *c, const void *hd, float val); +typedef bool upb_double_handlerfunc(void *c, const void *hd, double val); +typedef bool upb_bool_handlerfunc(void *c, const void *hd, bool val); +typedef void *upb_startstr_handlerfunc(void *c, const void *hd, + size_t size_hint); +typedef size_t upb_string_handlerfunc(void *c, const void *hd, const char *buf, + size_t n, const upb_bufhandle* handle); + +/* upb_bufhandle */ +size_t upb_bufhandle_objofs(const upb_bufhandle *h); + +/* upb_handlerattr */ +void upb_handlerattr_init(upb_handlerattr *attr); +void upb_handlerattr_uninit(upb_handlerattr *attr); + +bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd); +bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type); +const void *upb_handlerattr_closuretype(const upb_handlerattr *attr); +bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr, + const void *type); +const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr); +bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok); +bool upb_handlerattr_alwaysok(const upb_handlerattr *attr); + +UPB_INLINE const void *upb_handlerattr_handlerdata( + const upb_handlerattr *attr) { + return attr->handler_data_; +} + +/* upb_handlers */ +typedef void upb_handlers_callback(const void *closure, upb_handlers *h); +upb_handlers *upb_handlers_new(const upb_msgdef *m, + const void *owner); +const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m, + const void *owner, + upb_handlers_callback *callback, + const void *closure); + +/* Include refcounted methods like upb_handlers_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_handlers, upb_handlers_upcast) + +const upb_status *upb_handlers_status(upb_handlers *h); +void upb_handlers_clearerr(upb_handlers *h); +const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h); +bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *hfree); + +bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setint32(upb_handlers *h, const upb_fielddef *f, + upb_int32_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setint64(upb_handlers *h, const upb_fielddef *f, + upb_int64_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setuint32(upb_handlers *h, const upb_fielddef *f, + upb_uint32_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setuint64(upb_handlers *h, const upb_fielddef *f, + upb_uint64_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setfloat(upb_handlers *h, const upb_fielddef *f, + upb_float_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setdouble(upb_handlers *h, const upb_fielddef *f, + upb_double_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setbool(upb_handlers *h, const upb_fielddef *f, + upb_bool_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartstr(upb_handlers *h, const upb_fielddef *f, + upb_startstr_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstring(upb_handlers *h, const upb_fielddef *f, + upb_string_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendstr(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartseq(upb_handlers *h, const upb_fielddef *f, + upb_startfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartsubmsg(upb_handlers *h, const upb_fielddef *f, + upb_startfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendsubmsg(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendseq(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); + +bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f, + const upb_handlers *sub); +const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h, + const upb_fielddef *f); +const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h, + upb_selector_t sel); + +UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h, + upb_selector_t s) { + return (upb_func *)h->table[s].func; +} + +bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t s, + upb_handlerattr *attr); + +UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h, + upb_selector_t s) { + return upb_handlerattr_handlerdata(&h->table[s].attr); +} + +#ifdef __cplusplus + +/* Handler types for single fields. + * Right now we only have one for TYPE_BYTES but ones for other types + * should follow. + * + * These follow the same handlers protocol for fields of a message. */ +class upb::BytesHandler { + public: + BytesHandler(); + ~BytesHandler(); +#else +struct upb_byteshandler { +#endif + upb_handlers_tabent table[3]; +}; + +void upb_byteshandler_init(upb_byteshandler *h); + +/* Caller must ensure that "d" outlives the handlers. + * TODO(haberman): should this have a "freeze" operation? It's not necessary + * for memory management, but could be useful to force immutability and provide + * a convenient moment to verify that all registration succeeded. */ +bool upb_byteshandler_setstartstr(upb_byteshandler *h, + upb_startstr_handlerfunc *func, void *d); +bool upb_byteshandler_setstring(upb_byteshandler *h, + upb_string_handlerfunc *func, void *d); +bool upb_byteshandler_setendstr(upb_byteshandler *h, + upb_endfield_handlerfunc *func, void *d); + +/* "Static" methods */ +bool upb_handlers_freeze(upb_handlers *const *handlers, int n, upb_status *s); +upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f); +bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type, + upb_selector_t *s); +UPB_INLINE upb_selector_t upb_handlers_getendselector(upb_selector_t start) { + return start + 1; +} + +/* Internal-only. */ +uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f); +uint32_t upb_handlers_selectorcount(const upb_fielddef *f); + +UPB_END_EXTERN_C + +/* +** Inline definitions for handlers.h, which are particularly long and a bit +** tricky. +*/ + +#ifndef UPB_HANDLERS_INL_H_ +#define UPB_HANDLERS_INL_H_ + +#include <limits.h> + +/* C inline methods. */ + +/* upb_bufhandle */ +UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h) { + h->obj_ = NULL; + h->objtype_ = NULL; + h->buf_ = NULL; + h->objofs_ = 0; +} +UPB_INLINE void upb_bufhandle_uninit(upb_bufhandle *h) { + UPB_UNUSED(h); +} +UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj, + const void *type) { + h->obj_ = obj; + h->objtype_ = type; +} +UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf, + size_t ofs) { + h->buf_ = buf; + h->objofs_ = ofs; +} +UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h) { + return h->obj_; +} +UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h) { + return h->objtype_; +} +UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h) { + return h->buf_; +} + + +#ifdef __cplusplus + +/* Type detection and typedefs for integer types. + * For platforms where there are multiple 32-bit or 64-bit types, we need to be + * able to enumerate them so we can properly create overloads for all variants. + * + * If any platform existed where there were three integer types with the same + * size, this would have to become more complicated. For example, short, int, + * and long could all be 32-bits. Even more diabolically, short, int, long, + * and long long could all be 64 bits and still be standard-compliant. + * However, few platforms are this strange, and it's unlikely that upb will be + * used on the strangest ones. */ + +/* Can't count on stdint.h limits like INT32_MAX, because in C++ these are + * only defined when __STDC_LIMIT_MACROS are defined before the *first* include + * of stdint.h. We can't guarantee that someone else didn't include these first + * without defining __STDC_LIMIT_MACROS. */ +#define UPB_INT32_MAX 0x7fffffffLL +#define UPB_INT32_MIN (-UPB_INT32_MAX - 1) +#define UPB_INT64_MAX 0x7fffffffffffffffLL +#define UPB_INT64_MIN (-UPB_INT64_MAX - 1) + +#if INT_MAX == UPB_INT32_MAX && INT_MIN == UPB_INT32_MIN +#define UPB_INT_IS_32BITS 1 +#endif + +#if LONG_MAX == UPB_INT32_MAX && LONG_MIN == UPB_INT32_MIN +#define UPB_LONG_IS_32BITS 1 +#endif + +#if LONG_MAX == UPB_INT64_MAX && LONG_MIN == UPB_INT64_MIN +#define UPB_LONG_IS_64BITS 1 +#endif + +#if LLONG_MAX == UPB_INT64_MAX && LLONG_MIN == UPB_INT64_MIN +#define UPB_LLONG_IS_64BITS 1 +#endif + +/* We use macros instead of typedefs so we can undefine them later and avoid + * leaking them outside this header file. */ +#if UPB_INT_IS_32BITS +#define UPB_INT32_T int +#define UPB_UINT32_T unsigned int + +#if UPB_LONG_IS_32BITS +#define UPB_TWO_32BIT_TYPES 1 +#define UPB_INT32ALT_T long +#define UPB_UINT32ALT_T unsigned long +#endif /* UPB_LONG_IS_32BITS */ + +#elif UPB_LONG_IS_32BITS /* && !UPB_INT_IS_32BITS */ +#define UPB_INT32_T long +#define UPB_UINT32_T unsigned long +#endif /* UPB_INT_IS_32BITS */ + + +#if UPB_LONG_IS_64BITS +#define UPB_INT64_T long +#define UPB_UINT64_T unsigned long + +#if UPB_LLONG_IS_64BITS +#define UPB_TWO_64BIT_TYPES 1 +#define UPB_INT64ALT_T long long +#define UPB_UINT64ALT_T unsigned long long +#endif /* UPB_LLONG_IS_64BITS */ + +#elif UPB_LLONG_IS_64BITS /* && !UPB_LONG_IS_64BITS */ +#define UPB_INT64_T long long +#define UPB_UINT64_T unsigned long long +#endif /* UPB_LONG_IS_64BITS */ + +#undef UPB_INT32_MAX +#undef UPB_INT32_MIN +#undef UPB_INT64_MAX +#undef UPB_INT64_MIN +#undef UPB_INT_IS_32BITS +#undef UPB_LONG_IS_32BITS +#undef UPB_LONG_IS_64BITS +#undef UPB_LLONG_IS_64BITS + + +namespace upb { + +typedef void CleanupFunc(void *ptr); + +/* Template to remove "const" from "const T*" and just return "T*". + * + * We define a nonsense default because otherwise it will fail to instantiate as + * a function parameter type even in cases where we don't expect any caller to + * actually match the overload. */ +class CouldntRemoveConst {}; +template <class T> struct remove_constptr { typedef CouldntRemoveConst type; }; +template <class T> struct remove_constptr<const T *> { typedef T *type; }; + +/* Template that we use below to remove a template specialization from + * consideration if it matches a specific type. */ +template <class T, class U> struct disable_if_same { typedef void Type; }; +template <class T> struct disable_if_same<T, T> {}; + +template <class T> void DeletePointer(void *p) { delete static_cast<T>(p); } + +template <class T1, class T2> +struct FirstUnlessVoidOrBool { + typedef T1 value; +}; + +template <class T2> +struct FirstUnlessVoidOrBool<void, T2> { + typedef T2 value; +}; + +template <class T2> +struct FirstUnlessVoidOrBool<bool, T2> { + typedef T2 value; +}; + +template<class T, class U> +struct is_same { + static bool value; +}; + +template<class T> +struct is_same<T, T> { + static bool value; +}; + +template<class T, class U> +bool is_same<T, U>::value = false; + +template<class T> +bool is_same<T, T>::value = true; + +/* FuncInfo *******************************************************************/ + +/* Info about the user's original, pre-wrapped function. */ +template <class C, class R = void> +struct FuncInfo { + /* The type of the closure that the function takes (its first param). */ + typedef C Closure; + + /* The return type. */ + typedef R Return; +}; + +/* Func ***********************************************************************/ + +/* Func1, Func2, Func3: Template classes representing a function and its + * signature. + * + * Since the function is a template parameter, calling the function can be + * inlined at compile-time and does not require a function pointer at runtime. + * These functions are not bound to a handler data so have no data or cleanup + * handler. */ +struct UnboundFunc { + CleanupFunc *GetCleanup() { return NULL; } + void *GetData() { return NULL; } +}; + +template <class R, class P1, R F(P1), class I> +struct Func1 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1) { return F(p1); } +}; + +template <class R, class P1, class P2, R F(P1, P2), class I> +struct Func2 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2) { return F(p1, p2); } +}; + +template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I> +struct Func3 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3) { return F(p1, p2, p3); } +}; + +template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4), + class I> +struct Func4 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3, P4 p4) { return F(p1, p2, p3, p4); } +}; + +template <class R, class P1, class P2, class P3, class P4, class P5, + R F(P1, P2, P3, P4, P5), class I> +struct Func5 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3, P4 p4, P5 p5) { + return F(p1, p2, p3, p4, p5); + } +}; + +/* BoundFunc ******************************************************************/ + +/* BoundFunc2, BoundFunc3: Like Func2/Func3 except also contains a value that + * shall be bound to the function's second parameter. + * + * Note that the second parameter is a const pointer, but our stored bound value + * is non-const so we can free it when the handlers are destroyed. */ +template <class T> +struct BoundFunc { + typedef typename remove_constptr<T>::type MutableP2; + explicit BoundFunc(MutableP2 data_) : data(data_) {} + CleanupFunc *GetCleanup() { return &DeletePointer<MutableP2>; } + MutableP2 GetData() { return data; } + MutableP2 data; +}; + +template <class R, class P1, class P2, R F(P1, P2), class I> +struct BoundFunc2 : public BoundFunc<P2> { + typedef BoundFunc<P2> Base; + typedef I FuncInfo; + explicit BoundFunc2(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I> +struct BoundFunc3 : public BoundFunc<P2> { + typedef BoundFunc<P2> Base; + typedef I FuncInfo; + explicit BoundFunc3(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4), + class I> +struct BoundFunc4 : public BoundFunc<P2> { + typedef BoundFunc<P2> Base; + typedef I FuncInfo; + explicit BoundFunc4(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template <class R, class P1, class P2, class P3, class P4, class P5, + R F(P1, P2, P3, P4, P5), class I> +struct BoundFunc5 : public BoundFunc<P2> { + typedef BoundFunc<P2> Base; + typedef I FuncInfo; + explicit BoundFunc5(typename Base::MutableP2 arg) : Base(arg) {} +}; + +/* FuncSig ********************************************************************/ + +/* FuncSig1, FuncSig2, FuncSig3: template classes reflecting a function + * *signature*, but without a specific function attached. + * + * These classes contain member functions that can be invoked with a + * specific function to return a Func/BoundFunc class. */ +template <class R, class P1> +struct FuncSig1 { + template <R F(P1)> + Func1<R, P1, F, FuncInfo<P1, R> > GetFunc() { + return Func1<R, P1, F, FuncInfo<P1, R> >(); + } +}; + +template <class R, class P1, class P2> +struct FuncSig2 { + template <R F(P1, P2)> + Func2<R, P1, P2, F, FuncInfo<P1, R> > GetFunc() { + return Func2<R, P1, P2, F, FuncInfo<P1, R> >(); + } + + template <R F(P1, P2)> + BoundFunc2<R, P1, P2, F, FuncInfo<P1, R> > GetFunc( + typename remove_constptr<P2>::type param2) { + return BoundFunc2<R, P1, P2, F, FuncInfo<P1, R> >(param2); + } +}; + +template <class R, class P1, class P2, class P3> +struct FuncSig3 { + template <R F(P1, P2, P3)> + Func3<R, P1, P2, P3, F, FuncInfo<P1, R> > GetFunc() { + return Func3<R, P1, P2, P3, F, FuncInfo<P1, R> >(); + } + + template <R F(P1, P2, P3)> + BoundFunc3<R, P1, P2, P3, F, FuncInfo<P1, R> > GetFunc( + typename remove_constptr<P2>::type param2) { + return BoundFunc3<R, P1, P2, P3, F, FuncInfo<P1, R> >(param2); + } +}; + +template <class R, class P1, class P2, class P3, class P4> +struct FuncSig4 { + template <R F(P1, P2, P3, P4)> + Func4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> > GetFunc() { + return Func4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> >(); + } + + template <R F(P1, P2, P3, P4)> + BoundFunc4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> > GetFunc( + typename remove_constptr<P2>::type param2) { + return BoundFunc4<R, P1, P2, P3, P4, F, FuncInfo<P1, R> >(param2); + } +}; + +template <class R, class P1, class P2, class P3, class P4, class P5> +struct FuncSig5 { + template <R F(P1, P2, P3, P4, P5)> + Func5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> > GetFunc() { + return Func5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> >(); + } + + template <R F(P1, P2, P3, P4, P5)> + BoundFunc5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> > GetFunc( + typename remove_constptr<P2>::type param2) { + return BoundFunc5<R, P1, P2, P3, P4, P5, F, FuncInfo<P1, R> >(param2); + } +}; + +/* Overloaded template function that can construct the appropriate FuncSig* + * class given a function pointer by deducing the template parameters. */ +template <class R, class P1> +inline FuncSig1<R, P1> MatchFunc(R (*f)(P1)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return FuncSig1<R, P1>(); +} + +template <class R, class P1, class P2> +inline FuncSig2<R, P1, P2> MatchFunc(R (*f)(P1, P2)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return FuncSig2<R, P1, P2>(); +} + +template <class R, class P1, class P2, class P3> +inline FuncSig3<R, P1, P2, P3> MatchFunc(R (*f)(P1, P2, P3)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return FuncSig3<R, P1, P2, P3>(); +} + +template <class R, class P1, class P2, class P3, class P4> +inline FuncSig4<R, P1, P2, P3, P4> MatchFunc(R (*f)(P1, P2, P3, P4)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return FuncSig4<R, P1, P2, P3, P4>(); +} + +template <class R, class P1, class P2, class P3, class P4, class P5> +inline FuncSig5<R, P1, P2, P3, P4, P5> MatchFunc(R (*f)(P1, P2, P3, P4, P5)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return FuncSig5<R, P1, P2, P3, P4, P5>(); +} + +/* MethodSig ******************************************************************/ + +/* CallMethod*: a function template that calls a given method. */ +template <class R, class C, R (C::*F)()> +R CallMethod0(C *obj) { + return ((*obj).*F)(); +} + +template <class R, class C, class P1, R (C::*F)(P1)> +R CallMethod1(C *obj, P1 arg1) { + return ((*obj).*F)(arg1); +} + +template <class R, class C, class P1, class P2, R (C::*F)(P1, P2)> +R CallMethod2(C *obj, P1 arg1, P2 arg2) { + return ((*obj).*F)(arg1, arg2); +} + +template <class R, class C, class P1, class P2, class P3, R (C::*F)(P1, P2, P3)> +R CallMethod3(C *obj, P1 arg1, P2 arg2, P3 arg3) { + return ((*obj).*F)(arg1, arg2, arg3); +} + +template <class R, class C, class P1, class P2, class P3, class P4, + R (C::*F)(P1, P2, P3, P4)> +R CallMethod4(C *obj, P1 arg1, P2 arg2, P3 arg3, P4 arg4) { + return ((*obj).*F)(arg1, arg2, arg3, arg4); +} + +/* MethodSig: like FuncSig, but for member functions. + * + * GetFunc() returns a normal FuncN object, so after calling GetFunc() no + * more logic is required to special-case methods. */ +template <class R, class C> +struct MethodSig0 { + template <R (C::*F)()> + Func1<R, C *, CallMethod0<R, C, F>, FuncInfo<C *, R> > GetFunc() { + return Func1<R, C *, CallMethod0<R, C, F>, FuncInfo<C *, R> >(); + } +}; + +template <class R, class C, class P1> +struct MethodSig1 { + template <R (C::*F)(P1)> + Func2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> > GetFunc() { + return Func2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> >(); + } + + template <R (C::*F)(P1)> + BoundFunc2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> > GetFunc( + typename remove_constptr<P1>::type param1) { + return BoundFunc2<R, C *, P1, CallMethod1<R, C, P1, F>, FuncInfo<C *, R> >( + param1); + } +}; + +template <class R, class C, class P1, class P2> +struct MethodSig2 { + template <R (C::*F)(P1, P2)> + Func3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, FuncInfo<C *, R> > + GetFunc() { + return Func3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, + FuncInfo<C *, R> >(); + } + + template <R (C::*F)(P1, P2)> + BoundFunc3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, FuncInfo<C *, R> > + GetFunc(typename remove_constptr<P1>::type param1) { + return BoundFunc3<R, C *, P1, P2, CallMethod2<R, C, P1, P2, F>, + FuncInfo<C *, R> >(param1); + } +}; + +template <class R, class C, class P1, class P2, class P3> +struct MethodSig3 { + template <R (C::*F)(P1, P2, P3)> + Func4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>, FuncInfo<C *, R> > + GetFunc() { + return Func4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>, + FuncInfo<C *, R> >(); + } + + template <R (C::*F)(P1, P2, P3)> + BoundFunc4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>, + FuncInfo<C *, R> > + GetFunc(typename remove_constptr<P1>::type param1) { + return BoundFunc4<R, C *, P1, P2, P3, CallMethod3<R, C, P1, P2, P3, F>, + FuncInfo<C *, R> >(param1); + } +}; + +template <class R, class C, class P1, class P2, class P3, class P4> +struct MethodSig4 { + template <R (C::*F)(P1, P2, P3, P4)> + Func5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>, + FuncInfo<C *, R> > + GetFunc() { + return Func5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>, + FuncInfo<C *, R> >(); + } + + template <R (C::*F)(P1, P2, P3, P4)> + BoundFunc5<R, C *, P1, P2, P3, P4, CallMethod4<R, C, P1, P2, P3, P4, F>, + FuncInfo<C *, R> > + GetFunc(typename remove_constptr<P1>::type param1) { + return BoundFunc5<R, C *, P1, P2, P3, P4, + CallMethod4<R, C, P1, P2, P3, P4, F>, FuncInfo<C *, R> >( + param1); + } +}; + +template <class R, class C> +inline MethodSig0<R, C> MatchFunc(R (C::*f)()) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return MethodSig0<R, C>(); +} + +template <class R, class C, class P1> +inline MethodSig1<R, C, P1> MatchFunc(R (C::*f)(P1)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return MethodSig1<R, C, P1>(); +} + +template <class R, class C, class P1, class P2> +inline MethodSig2<R, C, P1, P2> MatchFunc(R (C::*f)(P1, P2)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return MethodSig2<R, C, P1, P2>(); +} + +template <class R, class C, class P1, class P2, class P3> +inline MethodSig3<R, C, P1, P2, P3> MatchFunc(R (C::*f)(P1, P2, P3)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return MethodSig3<R, C, P1, P2, P3>(); +} + +template <class R, class C, class P1, class P2, class P3, class P4> +inline MethodSig4<R, C, P1, P2, P3, P4> MatchFunc(R (C::*f)(P1, P2, P3, P4)) { + UPB_UNUSED(f); /* Only used for template parameter deduction. */ + return MethodSig4<R, C, P1, P2, P3, P4>(); +} + +/* MaybeWrapReturn ************************************************************/ + +/* Template class that attempts to wrap the return value of the function so it + * matches the expected type. There are two main adjustments it may make: + * + * 1. If the function returns void, make it return the expected type and with + * a value that always indicates success. + * 2. If the function returns bool, make it return the expected type with a + * value that indicates success or failure. + * + * The "expected type" for return is: + * 1. void* for start handlers. If the closure parameter has a different type + * we will cast it to void* for the return in the success case. + * 2. size_t for string buffer handlers. + * 3. bool for everything else. */ + +/* Template parameters are FuncN type and desired return type. */ +template <class F, class R, class Enable = void> +struct MaybeWrapReturn; + +/* If the return type matches, return the given function unwrapped. */ +template <class F> +struct MaybeWrapReturn<F, typename F::Return> { + typedef F Func; +}; + +/* Function wrapper that munges the return value from void to (bool)true. */ +template <class P1, class P2, void F(P1, P2)> +bool ReturnTrue2(P1 p1, P2 p2) { + F(p1, p2); + return true; +} + +template <class P1, class P2, class P3, void F(P1, P2, P3)> +bool ReturnTrue3(P1 p1, P2 p2, P3 p3) { + F(p1, p2, p3); + return true; +} + +/* Function wrapper that munges the return value from void to (void*)arg1 */ +template <class P1, class P2, void F(P1, P2)> +void *ReturnClosure2(P1 p1, P2 p2) { + F(p1, p2); + return p1; +} + +template <class P1, class P2, class P3, void F(P1, P2, P3)> +void *ReturnClosure3(P1 p1, P2 p2, P3 p3) { + F(p1, p2, p3); + return p1; +} + +/* Function wrapper that munges the return value from R to void*. */ +template <class R, class P1, class P2, R F(P1, P2)> +void *CastReturnToVoidPtr2(P1 p1, P2 p2) { + return F(p1, p2); +} + +template <class R, class P1, class P2, class P3, R F(P1, P2, P3)> +void *CastReturnToVoidPtr3(P1 p1, P2 p2, P3 p3) { + return F(p1, p2, p3); +} + +/* Function wrapper that munges the return value from bool to void*. */ +template <class P1, class P2, bool F(P1, P2)> +void *ReturnClosureOrBreak2(P1 p1, P2 p2) { + return F(p1, p2) ? p1 : UPB_BREAK; +} + +template <class P1, class P2, class P3, bool F(P1, P2, P3)> +void *ReturnClosureOrBreak3(P1 p1, P2 p2, P3 p3) { + return F(p1, p2, p3) ? p1 : UPB_BREAK; +} + +/* For the string callback, which takes five params, returns the size param. */ +template <class P1, class P2, + void F(P1, P2, const char *, size_t, const BufferHandle *)> +size_t ReturnStringLen(P1 p1, P2 p2, const char *p3, size_t p4, + const BufferHandle *p5) { + F(p1, p2, p3, p4, p5); + return p4; +} + +/* For the string callback, which takes five params, returns the size param or + * zero. */ +template <class P1, class P2, + bool F(P1, P2, const char *, size_t, const BufferHandle *)> +size_t ReturnNOr0(P1 p1, P2 p2, const char *p3, size_t p4, + const BufferHandle *p5) { + return F(p1, p2, p3, p4, p5) ? p4 : 0; +} + +/* If we have a function returning void but want a function returning bool, wrap + * it in a function that returns true. */ +template <class P1, class P2, void F(P1, P2), class I> +struct MaybeWrapReturn<Func2<void, P1, P2, F, I>, bool> { + typedef Func2<bool, P1, P2, ReturnTrue2<P1, P2, F>, I> Func; +}; + +template <class P1, class P2, class P3, void F(P1, P2, P3), class I> +struct MaybeWrapReturn<Func3<void, P1, P2, P3, F, I>, bool> { + typedef Func3<bool, P1, P2, P3, ReturnTrue3<P1, P2, P3, F>, I> Func; +}; + +/* If our function returns void but we want one returning void*, wrap it in a + * function that returns the first argument. */ +template <class P1, class P2, void F(P1, P2), class I> +struct MaybeWrapReturn<Func2<void, P1, P2, F, I>, void *> { + typedef Func2<void *, P1, P2, ReturnClosure2<P1, P2, F>, I> Func; +}; + +template <class P1, class P2, class P3, void F(P1, P2, P3), class I> +struct MaybeWrapReturn<Func3<void, P1, P2, P3, F, I>, void *> { + typedef Func3<void *, P1, P2, P3, ReturnClosure3<P1, P2, P3, F>, I> Func; +}; + +/* If our function returns R* but we want one returning void*, wrap it in a + * function that casts to void*. */ +template <class R, class P1, class P2, R *F(P1, P2), class I> +struct MaybeWrapReturn<Func2<R *, P1, P2, F, I>, void *, + typename disable_if_same<R *, void *>::Type> { + typedef Func2<void *, P1, P2, CastReturnToVoidPtr2<R *, P1, P2, F>, I> Func; +}; + +template <class R, class P1, class P2, class P3, R *F(P1, P2, P3), class I> +struct MaybeWrapReturn<Func3<R *, P1, P2, P3, F, I>, void *, + typename disable_if_same<R *, void *>::Type> { + typedef Func3<void *, P1, P2, P3, CastReturnToVoidPtr3<R *, P1, P2, P3, F>, I> + Func; +}; + +/* If our function returns bool but we want one returning void*, wrap it in a + * function that returns either the first param or UPB_BREAK. */ +template <class P1, class P2, bool F(P1, P2), class I> +struct MaybeWrapReturn<Func2<bool, P1, P2, F, I>, void *> { + typedef Func2<void *, P1, P2, ReturnClosureOrBreak2<P1, P2, F>, I> Func; +}; + +template <class P1, class P2, class P3, bool F(P1, P2, P3), class I> +struct MaybeWrapReturn<Func3<bool, P1, P2, P3, F, I>, void *> { + typedef Func3<void *, P1, P2, P3, ReturnClosureOrBreak3<P1, P2, P3, F>, I> + Func; +}; + +/* If our function returns void but we want one returning size_t, wrap it in a + * function that returns the size argument. */ +template <class P1, class P2, + void F(P1, P2, const char *, size_t, const BufferHandle *), class I> +struct MaybeWrapReturn< + Func5<void, P1, P2, const char *, size_t, const BufferHandle *, F, I>, + size_t> { + typedef Func5<size_t, P1, P2, const char *, size_t, const BufferHandle *, + ReturnStringLen<P1, P2, F>, I> Func; +}; + +/* If our function returns bool but we want one returning size_t, wrap it in a + * function that returns either 0 or the buf size. */ +template <class P1, class P2, + bool F(P1, P2, const char *, size_t, const BufferHandle *), class I> +struct MaybeWrapReturn< + Func5<bool, P1, P2, const char *, size_t, const BufferHandle *, F, I>, + size_t> { + typedef Func5<size_t, P1, P2, const char *, size_t, const BufferHandle *, + ReturnNOr0<P1, P2, F>, I> Func; +}; + +/* ConvertParams **************************************************************/ + +/* Template class that converts the function parameters if necessary, and + * ignores the HandlerData parameter if appropriate. + * + * Template parameter is the are FuncN function type. */ +template <class F, class T> +struct ConvertParams; + +/* Function that discards the handler data parameter. */ +template <class R, class P1, R F(P1)> +R IgnoreHandlerData2(void *p1, const void *hd) { + UPB_UNUSED(hd); + return F(static_cast<P1>(p1)); +} + +template <class R, class P1, class P2Wrapper, class P2Wrapped, + R F(P1, P2Wrapped)> +R IgnoreHandlerData3(void *p1, const void *hd, P2Wrapper p2) { + UPB_UNUSED(hd); + return F(static_cast<P1>(p1), p2); +} + +template <class R, class P1, class P2, class P3, R F(P1, P2, P3)> +R IgnoreHandlerData4(void *p1, const void *hd, P2 p2, P3 p3) { + UPB_UNUSED(hd); + return F(static_cast<P1>(p1), p2, p3); +} + +template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4)> +R IgnoreHandlerData5(void *p1, const void *hd, P2 p2, P3 p3, P4 p4) { + UPB_UNUSED(hd); + return F(static_cast<P1>(p1), p2, p3, p4); +} + +template <class R, class P1, R F(P1, const char*, size_t)> +R IgnoreHandlerDataIgnoreHandle(void *p1, const void *hd, const char *p2, + size_t p3, const BufferHandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + return F(static_cast<P1>(p1), p2, p3); +} + +/* Function that casts the handler data parameter. */ +template <class R, class P1, class P2, R F(P1, P2)> +R CastHandlerData2(void *c, const void *hd) { + return F(static_cast<P1>(c), static_cast<P2>(hd)); +} + +template <class R, class P1, class P2, class P3Wrapper, class P3Wrapped, + R F(P1, P2, P3Wrapped)> +R CastHandlerData3(void *c, const void *hd, P3Wrapper p3) { + return F(static_cast<P1>(c), static_cast<P2>(hd), p3); +} + +template <class R, class P1, class P2, class P3, class P4, class P5, + R F(P1, P2, P3, P4, P5)> +R CastHandlerData5(void *c, const void *hd, P3 p3, P4 p4, P5 p5) { + return F(static_cast<P1>(c), static_cast<P2>(hd), p3, p4, p5); +} + +template <class R, class P1, class P2, R F(P1, P2, const char *, size_t)> +R CastHandlerDataIgnoreHandle(void *c, const void *hd, const char *p3, + size_t p4, const BufferHandle *handle) { + UPB_UNUSED(handle); + return F(static_cast<P1>(c), static_cast<P2>(hd), p3, p4); +} + +/* For unbound functions, ignore the handler data. */ +template <class R, class P1, R F(P1), class I, class T> +struct ConvertParams<Func1<R, P1, F, I>, T> { + typedef Func2<R, void *, const void *, IgnoreHandlerData2<R, P1, F>, I> Func; +}; + +template <class R, class P1, class P2, R F(P1, P2), class I, + class R2, class P1_2, class P2_2, class P3_2> +struct ConvertParams<Func2<R, P1, P2, F, I>, + R2 (*)(P1_2, P2_2, P3_2)> { + typedef Func3<R, void *, const void *, P3_2, + IgnoreHandlerData3<R, P1, P3_2, P2, F>, I> Func; +}; + +/* For StringBuffer only; this ignores both the handler data and the + * BufferHandle. */ +template <class R, class P1, R F(P1, const char *, size_t), class I, class T> +struct ConvertParams<Func3<R, P1, const char *, size_t, F, I>, T> { + typedef Func5<R, void *, const void *, const char *, size_t, + const BufferHandle *, IgnoreHandlerDataIgnoreHandle<R, P1, F>, + I> Func; +}; + +template <class R, class P1, class P2, class P3, class P4, R F(P1, P2, P3, P4), + class I, class T> +struct ConvertParams<Func4<R, P1, P2, P3, P4, F, I>, T> { + typedef Func5<R, void *, const void *, P2, P3, P4, + IgnoreHandlerData5<R, P1, P2, P3, P4, F>, I> Func; +}; + +/* For bound functions, cast the handler data. */ +template <class R, class P1, class P2, R F(P1, P2), class I, class T> +struct ConvertParams<BoundFunc2<R, P1, P2, F, I>, T> { + typedef Func2<R, void *, const void *, CastHandlerData2<R, P1, P2, F>, I> + Func; +}; + +template <class R, class P1, class P2, class P3, R F(P1, P2, P3), class I, + class R2, class P1_2, class P2_2, class P3_2> +struct ConvertParams<BoundFunc3<R, P1, P2, P3, F, I>, + R2 (*)(P1_2, P2_2, P3_2)> { + typedef Func3<R, void *, const void *, P3_2, + CastHandlerData3<R, P1, P2, P3_2, P3, F>, I> Func; +}; + +/* For StringBuffer only; this ignores the BufferHandle. */ +template <class R, class P1, class P2, R F(P1, P2, const char *, size_t), + class I, class T> +struct ConvertParams<BoundFunc4<R, P1, P2, const char *, size_t, F, I>, T> { + typedef Func5<R, void *, const void *, const char *, size_t, + const BufferHandle *, CastHandlerDataIgnoreHandle<R, P1, P2, F>, + I> Func; +}; + +template <class R, class P1, class P2, class P3, class P4, class P5, + R F(P1, P2, P3, P4, P5), class I, class T> +struct ConvertParams<BoundFunc5<R, P1, P2, P3, P4, P5, F, I>, T> { + typedef Func5<R, void *, const void *, P3, P4, P5, + CastHandlerData5<R, P1, P2, P3, P4, P5, F>, I> Func; +}; + +/* utype/ltype are upper/lower-case, ctype is canonical C type, vtype is + * variant C type. */ +#define TYPE_METHODS(utype, ltype, ctype, vtype) \ + template <> struct CanonicalType<vtype> { \ + typedef ctype Type; \ + }; \ + template <> \ + inline bool Handlers::SetValueHandler<vtype>( \ + const FieldDef *f, \ + const Handlers::utype ## Handler& handler) { \ + assert(!handler.registered_); \ + handler.AddCleanup(this); \ + handler.registered_ = true; \ + return upb_handlers_set##ltype(this, f, handler.handler_, &handler.attr_); \ + } \ + +TYPE_METHODS(Double, double, double, double) +TYPE_METHODS(Float, float, float, float) +TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64_T) +TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32_T) +TYPE_METHODS(Int64, int64, int64_t, UPB_INT64_T) +TYPE_METHODS(Int32, int32, int32_t, UPB_INT32_T) +TYPE_METHODS(Bool, bool, bool, bool) + +#ifdef UPB_TWO_32BIT_TYPES +TYPE_METHODS(Int32, int32, int32_t, UPB_INT32ALT_T) +TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32ALT_T) +#endif + +#ifdef UPB_TWO_64BIT_TYPES +TYPE_METHODS(Int64, int64, int64_t, UPB_INT64ALT_T) +TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64ALT_T) +#endif +#undef TYPE_METHODS + +template <> struct CanonicalType<Status*> { + typedef Status* Type; +}; + +/* Type methods that are only one-per-canonical-type and not + * one-per-cvariant. */ + +#define TYPE_METHODS(utype, ctype) \ + inline bool Handlers::Set##utype##Handler(const FieldDef *f, \ + const utype##Handler &h) { \ + return SetValueHandler<ctype>(f, h); \ + } \ + +TYPE_METHODS(Double, double) +TYPE_METHODS(Float, float) +TYPE_METHODS(UInt64, uint64_t) +TYPE_METHODS(UInt32, uint32_t) +TYPE_METHODS(Int64, int64_t) +TYPE_METHODS(Int32, int32_t) +TYPE_METHODS(Bool, bool) +#undef TYPE_METHODS + +template <class F> struct ReturnOf; + +template <class R, class P1, class P2> +struct ReturnOf<R (*)(P1, P2)> { + typedef R Return; +}; + +template <class R, class P1, class P2, class P3> +struct ReturnOf<R (*)(P1, P2, P3)> { + typedef R Return; +}; + +template <class R, class P1, class P2, class P3, class P4> +struct ReturnOf<R (*)(P1, P2, P3, P4)> { + typedef R Return; +}; + +template <class R, class P1, class P2, class P3, class P4, class P5> +struct ReturnOf<R (*)(P1, P2, P3, P4, P5)> { + typedef R Return; +}; + +template<class T> const void *UniquePtrForType() { + static const char ch = 0; + return &ch; +} + +template <class T> +template <class F> +inline Handler<T>::Handler(F func) + : registered_(false), + cleanup_data_(func.GetData()), + cleanup_func_(func.GetCleanup()) { + upb_handlerattr_sethandlerdata(&attr_, func.GetData()); + typedef typename ReturnOf<T>::Return Return; + typedef typename ConvertParams<F, T>::Func ConvertedParamsFunc; + typedef typename MaybeWrapReturn<ConvertedParamsFunc, Return>::Func + ReturnWrappedFunc; + handler_ = ReturnWrappedFunc().Call; + + /* Set attributes based on what templates can statically tell us about the + * user's function. */ + + /* If the original function returns void, then we know that we wrapped it to + * always return ok. */ + bool always_ok = is_same<typename F::FuncInfo::Return, void>::value; + attr_.SetAlwaysOk(always_ok); + + /* Closure parameter and return type. */ + attr_.SetClosureType(UniquePtrForType<typename F::FuncInfo::Closure>()); + + /* We use the closure type (from the first parameter) if the return type is + * void or bool, since these are the two cases we wrap to return the closure's + * type anyway. + * + * This is all nonsense for non START* handlers, but it doesn't matter because + * in that case the value will be ignored. */ + typedef typename FirstUnlessVoidOrBool<typename F::FuncInfo::Return, + typename F::FuncInfo::Closure>::value + EffectiveReturn; + attr_.SetReturnClosureType(UniquePtrForType<EffectiveReturn>()); +} + +template <class T> +inline Handler<T>::~Handler() { + assert(registered_); +} + +inline HandlerAttributes::HandlerAttributes() { upb_handlerattr_init(this); } +inline HandlerAttributes::~HandlerAttributes() { upb_handlerattr_uninit(this); } +inline bool HandlerAttributes::SetHandlerData(const void *hd) { + return upb_handlerattr_sethandlerdata(this, hd); +} +inline const void* HandlerAttributes::handler_data() const { + return upb_handlerattr_handlerdata(this); +} +inline bool HandlerAttributes::SetClosureType(const void *type) { + return upb_handlerattr_setclosuretype(this, type); +} +inline const void* HandlerAttributes::closure_type() const { + return upb_handlerattr_closuretype(this); +} +inline bool HandlerAttributes::SetReturnClosureType(const void *type) { + return upb_handlerattr_setreturnclosuretype(this, type); +} +inline const void* HandlerAttributes::return_closure_type() const { + return upb_handlerattr_returnclosuretype(this); +} +inline bool HandlerAttributes::SetAlwaysOk(bool always_ok) { + return upb_handlerattr_setalwaysok(this, always_ok); +} +inline bool HandlerAttributes::always_ok() const { + return upb_handlerattr_alwaysok(this); +} + +inline BufferHandle::BufferHandle() { upb_bufhandle_init(this); } +inline BufferHandle::~BufferHandle() { upb_bufhandle_uninit(this); } +inline const char* BufferHandle::buffer() const { + return upb_bufhandle_buf(this); +} +inline size_t BufferHandle::object_offset() const { + return upb_bufhandle_objofs(this); +} +inline void BufferHandle::SetBuffer(const char* buf, size_t ofs) { + upb_bufhandle_setbuf(this, buf, ofs); +} +template <class T> +void BufferHandle::SetAttachedObject(const T* obj) { + upb_bufhandle_setobj(this, obj, UniquePtrForType<T>()); +} +template <class T> +const T* BufferHandle::GetAttachedObject() const { + return upb_bufhandle_objtype(this) == UniquePtrForType<T>() + ? static_cast<const T *>(upb_bufhandle_obj(this)) + : NULL; +} + +inline reffed_ptr<Handlers> Handlers::New(const MessageDef *m) { + upb_handlers *h = upb_handlers_new(m, &h); + return reffed_ptr<Handlers>(h, &h); +} +inline reffed_ptr<const Handlers> Handlers::NewFrozen( + const MessageDef *m, upb_handlers_callback *callback, + const void *closure) { + const upb_handlers *h = upb_handlers_newfrozen(m, &h, callback, closure); + return reffed_ptr<const Handlers>(h, &h); +} +inline const Status* Handlers::status() { + return upb_handlers_status(this); +} +inline void Handlers::ClearError() { + return upb_handlers_clearerr(this); +} +inline bool Handlers::Freeze(Status *s) { + upb::Handlers* h = this; + return upb_handlers_freeze(&h, 1, s); +} +inline bool Handlers::Freeze(Handlers *const *handlers, int n, Status *s) { + return upb_handlers_freeze(handlers, n, s); +} +inline bool Handlers::Freeze(const std::vector<Handlers*>& h, Status* status) { + return upb_handlers_freeze((Handlers* const*)&h[0], h.size(), status); +} +inline const MessageDef *Handlers::message_def() const { + return upb_handlers_msgdef(this); +} +inline bool Handlers::AddCleanup(void *p, upb_handlerfree *func) { + return upb_handlers_addcleanup(this, p, func); +} +inline bool Handlers::SetStartMessageHandler( + const Handlers::StartMessageHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartmsg(this, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndMessageHandler( + const Handlers::EndMessageHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendmsg(this, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartStringHandler(const FieldDef *f, + const StartStringHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartstr(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndStringHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendstr(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStringHandler(const FieldDef *f, + const StringHandler& handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstring(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartSequenceHandler( + const FieldDef *f, const StartFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartseq(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartSubMessageHandler( + const FieldDef *f, const StartFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartsubmsg(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndSubMessageHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendsubmsg(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndSequenceHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendseq(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetSubHandlers(const FieldDef *f, const Handlers *sub) { + return upb_handlers_setsubhandlers(this, f, sub); +} +inline const Handlers *Handlers::GetSubHandlers(const FieldDef *f) const { + return upb_handlers_getsubhandlers(this, f); +} +inline const Handlers *Handlers::GetSubHandlers(Handlers::Selector sel) const { + return upb_handlers_getsubhandlers_sel(this, sel); +} +inline bool Handlers::GetSelector(const FieldDef *f, Handlers::Type type, + Handlers::Selector *s) { + return upb_handlers_getselector(f, type, s); +} +inline Handlers::Selector Handlers::GetEndSelector(Handlers::Selector start) { + return upb_handlers_getendselector(start); +} +inline Handlers::GenericFunction *Handlers::GetHandler( + Handlers::Selector selector) { + return upb_handlers_gethandler(this, selector); +} +inline const void *Handlers::GetHandlerData(Handlers::Selector selector) { + return upb_handlers_gethandlerdata(this, selector); +} + +inline BytesHandler::BytesHandler() { + upb_byteshandler_init(this); +} + +inline BytesHandler::~BytesHandler() {} + +} /* namespace upb */ + +#endif /* __cplusplus */ + + +#undef UPB_TWO_32BIT_TYPES +#undef UPB_TWO_64BIT_TYPES +#undef UPB_INT32_T +#undef UPB_UINT32_T +#undef UPB_INT32ALT_T +#undef UPB_UINT32ALT_T +#undef UPB_INT64_T +#undef UPB_UINT64_T +#undef UPB_INT64ALT_T +#undef UPB_UINT64ALT_T + +#endif /* UPB_HANDLERS_INL_H_ */ + +#endif /* UPB_HANDLERS_H */ +/* +** upb::Sink (upb_sink) +** upb::BytesSink (upb_bytessink) +** +** A upb_sink is an object that binds a upb_handlers object to some runtime +** state. It is the object that can actually receive data via the upb_handlers +** interface. +** +** Unlike upb_def and upb_handlers, upb_sink is never frozen, immutable, or +** thread-safe. You can create as many of them as you want, but each one may +** only be used in a single thread at a time. +** +** If we compare with class-based OOP, a you can think of a upb_def as an +** abstract base class, a upb_handlers as a concrete derived class, and a +** upb_sink as an object (class instance). +*/ + +#ifndef UPB_SINK_H +#define UPB_SINK_H + + +#ifdef __cplusplus +namespace upb { +class BufferSource; +class BytesSink; +class Sink; +} +#endif + +UPB_DECLARE_TYPE(upb::BufferSource, upb_bufsrc) +UPB_DECLARE_TYPE(upb::BytesSink, upb_bytessink) +UPB_DECLARE_TYPE(upb::Sink, upb_sink) + +#ifdef __cplusplus + +/* A upb::Sink is an object that binds a upb::Handlers object to some runtime + * state. It represents an endpoint to which data can be sent. + * + * TODO(haberman): right now all of these functions take selectors. Should they + * take selectorbase instead? + * + * ie. instead of calling: + * sink->StartString(FOO_FIELD_START_STRING, ...) + * a selector base would let you say: + * sink->StartString(FOO_FIELD, ...) + * + * This would make call sites a little nicer and require emitting fewer selector + * definitions in .h files. + * + * But the current scheme has the benefit that you can retrieve a function + * pointer for any handler with handlers->GetHandler(selector), without having + * to have a separate GetHandler() function for each handler type. The JIT + * compiler uses this. To accommodate we'd have to expose a separate + * GetHandler() for every handler type. + * + * Also to ponder: selectors right now are independent of a specific Handlers + * instance. In other words, they allocate a number to every possible handler + * that *could* be registered, without knowing anything about what handlers + * *are* registered. That means that using selectors as table offsets prohibits + * us from compacting the handler table at Freeze() time. If the table is very + * sparse, this could be wasteful. + * + * Having another selector-like thing that is specific to a Handlers instance + * would allow this compacting, but then it would be impossible to write code + * ahead-of-time that can be bound to any Handlers instance at runtime. For + * example, a .proto file parser written as straight C will not know what + * Handlers it will be bound to, so when it calls sink->StartString() what + * selector will it pass? It needs a selector like we have today, that is + * independent of any particular upb::Handlers. + * + * Is there a way then to allow Handlers table compaction? */ +class upb::Sink { + public: + /* Constructor with no initialization; must be Reset() before use. */ + Sink() {} + + /* Constructs a new sink for the given frozen handlers and closure. + * + * TODO: once the Handlers know the expected closure type, verify that T + * matches it. */ + template <class T> Sink(const Handlers* handlers, T* closure); + + /* Resets the value of the sink. */ + template <class T> void Reset(const Handlers* handlers, T* closure); + + /* Returns the top-level object that is bound to this sink. + * + * TODO: once the Handlers know the expected closure type, verify that T + * matches it. */ + template <class T> T* GetObject() const; + + /* Functions for pushing data into the sink. + * + * These return false if processing should stop (either due to error or just + * to suspend). + * + * These may not be called from within one of the same sink's handlers (in + * other words, handlers are not re-entrant). */ + + /* Should be called at the start and end of every message; both the top-level + * message and submessages. This means that submessages should use the + * following sequence: + * sink->StartSubMessage(startsubmsg_selector); + * sink->StartMessage(); + * // ... + * sink->EndMessage(&status); + * sink->EndSubMessage(endsubmsg_selector); */ + bool StartMessage(); + bool EndMessage(Status* status); + + /* Putting of individual values. These work for both repeated and + * non-repeated fields, but for repeated fields you must wrap them in + * calls to StartSequence()/EndSequence(). */ + bool PutInt32(Handlers::Selector s, int32_t val); + bool PutInt64(Handlers::Selector s, int64_t val); + bool PutUInt32(Handlers::Selector s, uint32_t val); + bool PutUInt64(Handlers::Selector s, uint64_t val); + bool PutFloat(Handlers::Selector s, float val); + bool PutDouble(Handlers::Selector s, double val); + bool PutBool(Handlers::Selector s, bool val); + + /* Putting of string/bytes values. Each string can consist of zero or more + * non-contiguous buffers of data. + * + * For StartString(), the function will write a sink for the string to "sub." + * The sub-sink must be used for any/all PutStringBuffer() calls. */ + bool StartString(Handlers::Selector s, size_t size_hint, Sink* sub); + size_t PutStringBuffer(Handlers::Selector s, const char *buf, size_t len, + const BufferHandle *handle); + bool EndString(Handlers::Selector s); + + /* For submessage fields. + * + * For StartSubMessage(), the function will write a sink for the string to + * "sub." The sub-sink must be used for any/all handlers called within the + * submessage. */ + bool StartSubMessage(Handlers::Selector s, Sink* sub); + bool EndSubMessage(Handlers::Selector s); + + /* For repeated fields of any type, the sequence of values must be wrapped in + * these calls. + * + * For StartSequence(), the function will write a sink for the string to + * "sub." The sub-sink must be used for any/all handlers called within the + * sequence. */ + bool StartSequence(Handlers::Selector s, Sink* sub); + bool EndSequence(Handlers::Selector s); + + /* Copy and assign specifically allowed. + * We don't even bother making these members private because so many + * functions need them and this is mainly just a dumb data container anyway. + */ +#else +struct upb_sink { +#endif + const upb_handlers *handlers; + void *closure; +}; + +#ifdef __cplusplus +class upb::BytesSink { + public: + BytesSink() {} + + /* Constructs a new sink for the given frozen handlers and closure. + * + * TODO(haberman): once the Handlers know the expected closure type, verify + * that T matches it. */ + template <class T> BytesSink(const BytesHandler* handler, T* closure); + + /* Resets the value of the sink. */ + template <class T> void Reset(const BytesHandler* handler, T* closure); + + bool Start(size_t size_hint, void **subc); + size_t PutBuffer(void *subc, const char *buf, size_t len, + const BufferHandle *handle); + bool End(); +#else +struct upb_bytessink { +#endif + const upb_byteshandler *handler; + void *closure; +}; + +#ifdef __cplusplus + +/* A class for pushing a flat buffer of data to a BytesSink. + * You can construct an instance of this to get a resumable source, + * or just call the static PutBuffer() to do a non-resumable push all in one + * go. */ +class upb::BufferSource { + public: + BufferSource(); + BufferSource(const char* buf, size_t len, BytesSink* sink); + + /* Returns true if the entire buffer was pushed successfully. Otherwise the + * next call to PutNext() will resume where the previous one left off. + * TODO(haberman): implement this. */ + bool PutNext(); + + /* A static version; with this version is it not possible to resume in the + * case of failure or a partially-consumed buffer. */ + static bool PutBuffer(const char* buf, size_t len, BytesSink* sink); + + template <class T> static bool PutBuffer(const T& str, BytesSink* sink) { + return PutBuffer(str.c_str(), str.size(), sink); + } +#else +struct upb_bufsrc { + char dummy; +#endif +}; + +UPB_BEGIN_EXTERN_C + +/* Inline definitions. */ + +UPB_INLINE void upb_bytessink_reset(upb_bytessink *s, const upb_byteshandler *h, + void *closure) { + s->handler = h; + s->closure = closure; +} + +UPB_INLINE bool upb_bytessink_start(upb_bytessink *s, size_t size_hint, + void **subc) { + typedef upb_startstr_handlerfunc func; + func *start; + *subc = s->closure; + if (!s->handler) return true; + start = (func *)s->handler->table[UPB_STARTSTR_SELECTOR].func; + + if (!start) return true; + *subc = start(s->closure, upb_handlerattr_handlerdata( + &s->handler->table[UPB_STARTSTR_SELECTOR].attr), + size_hint); + return *subc != NULL; +} + +UPB_INLINE size_t upb_bytessink_putbuf(upb_bytessink *s, void *subc, + const char *buf, size_t size, + const upb_bufhandle* handle) { + typedef upb_string_handlerfunc func; + func *putbuf; + if (!s->handler) return true; + putbuf = (func *)s->handler->table[UPB_STRING_SELECTOR].func; + + if (!putbuf) return true; + return putbuf(subc, upb_handlerattr_handlerdata( + &s->handler->table[UPB_STRING_SELECTOR].attr), + buf, size, handle); +} + +UPB_INLINE bool upb_bytessink_end(upb_bytessink *s) { + typedef upb_endfield_handlerfunc func; + func *end; + if (!s->handler) return true; + end = (func *)s->handler->table[UPB_ENDSTR_SELECTOR].func; + + if (!end) return true; + return end(s->closure, + upb_handlerattr_handlerdata( + &s->handler->table[UPB_ENDSTR_SELECTOR].attr)); +} + +UPB_INLINE bool upb_bufsrc_putbuf(const char *buf, size_t len, + upb_bytessink *sink) { + void *subc; + bool ret; + upb_bufhandle handle; + upb_bufhandle_init(&handle); + upb_bufhandle_setbuf(&handle, buf, 0); + ret = upb_bytessink_start(sink, len, &subc); + if (ret && len != 0) { + ret = (upb_bytessink_putbuf(sink, subc, buf, len, &handle) >= len); + } + if (ret) { + ret = upb_bytessink_end(sink); + } + upb_bufhandle_uninit(&handle); + return ret; +} + +#define PUTVAL(type, ctype) \ + UPB_INLINE bool upb_sink_put##type(upb_sink *s, upb_selector_t sel, \ + ctype val) { \ + typedef upb_##type##_handlerfunc functype; \ + functype *func; \ + const void *hd; \ + if (!s->handlers) return true; \ + func = (functype *)upb_handlers_gethandler(s->handlers, sel); \ + if (!func) return true; \ + hd = upb_handlers_gethandlerdata(s->handlers, sel); \ + return func(s->closure, hd, val); \ + } + +PUTVAL(int32, int32_t) +PUTVAL(int64, int64_t) +PUTVAL(uint32, uint32_t) +PUTVAL(uint64, uint64_t) +PUTVAL(float, float) +PUTVAL(double, double) +PUTVAL(bool, bool) +#undef PUTVAL + +UPB_INLINE void upb_sink_reset(upb_sink *s, const upb_handlers *h, void *c) { + s->handlers = h; + s->closure = c; +} + +UPB_INLINE size_t upb_sink_putstring(upb_sink *s, upb_selector_t sel, + const char *buf, size_t n, + const upb_bufhandle *handle) { + typedef upb_string_handlerfunc func; + func *handler; + const void *hd; + if (!s->handlers) return n; + handler = (func *)upb_handlers_gethandler(s->handlers, sel); + + if (!handler) return n; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + return handler(s->closure, hd, buf, n, handle); +} + +UPB_INLINE bool upb_sink_startmsg(upb_sink *s) { + typedef upb_startmsg_handlerfunc func; + func *startmsg; + const void *hd; + if (!s->handlers) return true; + startmsg = (func*)upb_handlers_gethandler(s->handlers, UPB_STARTMSG_SELECTOR); + + if (!startmsg) return true; + hd = upb_handlers_gethandlerdata(s->handlers, UPB_STARTMSG_SELECTOR); + return startmsg(s->closure, hd); +} + +UPB_INLINE bool upb_sink_endmsg(upb_sink *s, upb_status *status) { + typedef upb_endmsg_handlerfunc func; + func *endmsg; + const void *hd; + if (!s->handlers) return true; + endmsg = (func *)upb_handlers_gethandler(s->handlers, UPB_ENDMSG_SELECTOR); + + if (!endmsg) return true; + hd = upb_handlers_gethandlerdata(s->handlers, UPB_ENDMSG_SELECTOR); + return endmsg(s->closure, hd, status); +} + +UPB_INLINE bool upb_sink_startseq(upb_sink *s, upb_selector_t sel, + upb_sink *sub) { + typedef upb_startfield_handlerfunc func; + func *startseq; + const void *hd; + sub->closure = s->closure; + sub->handlers = s->handlers; + if (!s->handlers) return true; + startseq = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!startseq) return true; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startseq(s->closure, hd); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endseq(upb_sink *s, upb_selector_t sel) { + typedef upb_endfield_handlerfunc func; + func *endseq; + const void *hd; + if (!s->handlers) return true; + endseq = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!endseq) return true; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endseq(s->closure, hd); +} + +UPB_INLINE bool upb_sink_startstr(upb_sink *s, upb_selector_t sel, + size_t size_hint, upb_sink *sub) { + typedef upb_startstr_handlerfunc func; + func *startstr; + const void *hd; + sub->closure = s->closure; + sub->handlers = s->handlers; + if (!s->handlers) return true; + startstr = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!startstr) return true; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startstr(s->closure, hd, size_hint); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endstr(upb_sink *s, upb_selector_t sel) { + typedef upb_endfield_handlerfunc func; + func *endstr; + const void *hd; + if (!s->handlers) return true; + endstr = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!endstr) return true; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endstr(s->closure, hd); +} + +UPB_INLINE bool upb_sink_startsubmsg(upb_sink *s, upb_selector_t sel, + upb_sink *sub) { + typedef upb_startfield_handlerfunc func; + func *startsubmsg; + const void *hd; + sub->closure = s->closure; + if (!s->handlers) { + sub->handlers = NULL; + return true; + } + sub->handlers = upb_handlers_getsubhandlers_sel(s->handlers, sel); + startsubmsg = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!startsubmsg) return true; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startsubmsg(s->closure, hd); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endsubmsg(upb_sink *s, upb_selector_t sel) { + typedef upb_endfield_handlerfunc func; + func *endsubmsg; + const void *hd; + if (!s->handlers) return true; + endsubmsg = (func*)upb_handlers_gethandler(s->handlers, sel); + + if (!endsubmsg) return s->closure; + hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endsubmsg(s->closure, hd); +} + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { + +template <class T> Sink::Sink(const Handlers* handlers, T* closure) { + upb_sink_reset(this, handlers, closure); +} +template <class T> +inline void Sink::Reset(const Handlers* handlers, T* closure) { + upb_sink_reset(this, handlers, closure); +} +inline bool Sink::StartMessage() { + return upb_sink_startmsg(this); +} +inline bool Sink::EndMessage(Status* status) { + return upb_sink_endmsg(this, status); +} +inline bool Sink::PutInt32(Handlers::Selector sel, int32_t val) { + return upb_sink_putint32(this, sel, val); +} +inline bool Sink::PutInt64(Handlers::Selector sel, int64_t val) { + return upb_sink_putint64(this, sel, val); +} +inline bool Sink::PutUInt32(Handlers::Selector sel, uint32_t val) { + return upb_sink_putuint32(this, sel, val); +} +inline bool Sink::PutUInt64(Handlers::Selector sel, uint64_t val) { + return upb_sink_putuint64(this, sel, val); +} +inline bool Sink::PutFloat(Handlers::Selector sel, float val) { + return upb_sink_putfloat(this, sel, val); +} +inline bool Sink::PutDouble(Handlers::Selector sel, double val) { + return upb_sink_putdouble(this, sel, val); +} +inline bool Sink::PutBool(Handlers::Selector sel, bool val) { + return upb_sink_putbool(this, sel, val); +} +inline bool Sink::StartString(Handlers::Selector sel, size_t size_hint, + Sink *sub) { + return upb_sink_startstr(this, sel, size_hint, sub); +} +inline size_t Sink::PutStringBuffer(Handlers::Selector sel, const char *buf, + size_t len, const BufferHandle* handle) { + return upb_sink_putstring(this, sel, buf, len, handle); +} +inline bool Sink::EndString(Handlers::Selector sel) { + return upb_sink_endstr(this, sel); +} +inline bool Sink::StartSubMessage(Handlers::Selector sel, Sink* sub) { + return upb_sink_startsubmsg(this, sel, sub); +} +inline bool Sink::EndSubMessage(Handlers::Selector sel) { + return upb_sink_endsubmsg(this, sel); +} +inline bool Sink::StartSequence(Handlers::Selector sel, Sink* sub) { + return upb_sink_startseq(this, sel, sub); +} +inline bool Sink::EndSequence(Handlers::Selector sel) { + return upb_sink_endseq(this, sel); +} + +template <class T> +BytesSink::BytesSink(const BytesHandler* handler, T* closure) { + Reset(handler, closure); +} + +template <class T> +void BytesSink::Reset(const BytesHandler *handler, T *closure) { + upb_bytessink_reset(this, handler, closure); +} +inline bool BytesSink::Start(size_t size_hint, void **subc) { + return upb_bytessink_start(this, size_hint, subc); +} +inline size_t BytesSink::PutBuffer(void *subc, const char *buf, size_t len, + const BufferHandle *handle) { + return upb_bytessink_putbuf(this, subc, buf, len, handle); +} +inline bool BytesSink::End() { + return upb_bytessink_end(this); +} + +inline bool BufferSource::PutBuffer(const char *buf, size_t len, + BytesSink *sink) { + return upb_bufsrc_putbuf(buf, len, sink); +} + +} /* namespace upb */ +#endif + +#endif +/* +** For handlers that do very tiny, very simple operations, the function call +** overhead of calling a handler can be significant. This file allows the +** user to define handlers that do something very simple like store the value +** to memory and/or set a hasbit. JIT compilers can then special-case these +** handlers and emit specialized code for them instead of actually calling the +** handler. +** +** The functionality is very simple/limited right now but may expand to be able +** to call another function. +*/ + +#ifndef UPB_SHIM_H +#define UPB_SHIM_H + + +typedef struct { + size_t offset; + int32_t hasbit; +} upb_shim_data; + +#ifdef __cplusplus + +namespace upb { + +struct Shim { + typedef upb_shim_data Data; + + /* Sets a handler for the given field that writes the value to the given + * offset and, if hasbit >= 0, sets a bit at the given bit offset. Returns + * true if the handler was set successfully. */ + static bool Set(Handlers *h, const FieldDef *f, size_t ofs, int32_t hasbit); + + /* If this handler is a shim, returns the corresponding upb::Shim::Data and + * stores the type in "type". Otherwise returns NULL. */ + static const Data* GetData(const Handlers* h, Handlers::Selector s, + FieldDef::Type* type); +}; + +} /* namespace upb */ + +#endif + +UPB_BEGIN_EXTERN_C + +/* C API. */ +bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset, + int32_t hasbit); +const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s, + upb_fieldtype_t *type); + +UPB_END_EXTERN_C + +#ifdef __cplusplus +/* C++ Wrappers. */ +namespace upb { +inline bool Shim::Set(Handlers* h, const FieldDef* f, size_t ofs, + int32_t hasbit) { + return upb_shim_set(h, f, ofs, hasbit); +} +inline const Shim::Data* Shim::GetData(const Handlers* h, Handlers::Selector s, + FieldDef::Type* type) { + return upb_shim_getdata(h, s, type); +} +} /* namespace upb */ +#endif + +#endif /* UPB_SHIM_H */ +/* +** upb::SymbolTable (upb_symtab) +** +** A symtab (symbol table) stores a name->def map of upb_defs. Clients could +** always create such tables themselves, but upb_symtab has logic for resolving +** symbolic references, and in particular, for keeping a whole set of consistent +** defs when replacing some subset of those defs. This logic is nontrivial. +** +** This is a mixed C/C++ interface that offers a full API to both languages. +** See the top-level README for more information. +*/ + +#ifndef UPB_SYMTAB_H_ +#define UPB_SYMTAB_H_ + + +#ifdef __cplusplus +#include <vector> +namespace upb { class SymbolTable; } +#endif + +UPB_DECLARE_DERIVED_TYPE(upb::SymbolTable, upb::RefCounted, + upb_symtab, upb_refcounted) + +typedef struct { + UPB_PRIVATE_FOR_CPP + upb_strtable_iter iter; + upb_deftype_t type; +} upb_symtab_iter; + +#ifdef __cplusplus + +/* Non-const methods in upb::SymbolTable are NOT thread-safe. */ +class upb::SymbolTable { + public: + /* Returns a new symbol table with a single ref owned by "owner." + * Returns NULL if memory allocation failed. */ + static reffed_ptr<SymbolTable> New(); + + /* Include RefCounted base methods. */ + UPB_REFCOUNTED_CPPMETHODS + + /* For all lookup functions, the returned pointer is not owned by the + * caller; it may be invalidated by any non-const call or unref of the + * SymbolTable! To protect against this, take a ref if desired. */ + + /* Freezes the symbol table: prevents further modification of it. + * After the Freeze() operation is successful, the SymbolTable must only be + * accessed via a const pointer. + * + * Unlike with upb::MessageDef/upb::EnumDef/etc, freezing a SymbolTable is not + * a necessary step in using a SymbolTable. If you have no need for it to be + * immutable, there is no need to freeze it ever. However sometimes it is + * useful, and SymbolTables that are statically compiled into the binary are + * always frozen by nature. */ + void Freeze(); + + /* Resolves the given symbol using the rules described in descriptor.proto, + * namely: + * + * If the name starts with a '.', it is fully-qualified. Otherwise, + * C++-like scoping rules are used to find the type (i.e. first the nested + * types within this message are searched, then within the parent, on up + * to the root namespace). + * + * If not found, returns NULL. */ + const Def* Resolve(const char* base, const char* sym) const; + + /* Finds an entry in the symbol table with this exact name. If not found, + * returns NULL. */ + const Def* Lookup(const char *sym) const; + const MessageDef* LookupMessage(const char *sym) const; + const EnumDef* LookupEnum(const char *sym) const; + + /* TODO: introduce a C++ iterator, but make it nice and templated so that if + * you ask for an iterator of MessageDef the iterated elements are strongly + * typed as MessageDef*. */ + + /* Adds the given mutable defs to the symtab, resolving all symbols + * (including enum default values) and finalizing the defs. Only one def per + * name may be in the list, but defs can replace existing defs in the symtab. + * All defs must have a name -- anonymous defs are not allowed. Anonymous + * defs can still be frozen by calling upb_def_freeze() directly. + * + * Any existing defs that can reach defs that are being replaced will + * themselves be replaced also, so that the resulting set of defs is fully + * consistent. + * + * This logic implemented in this method is a convenience; ultimately it + * calls some combination of upb_fielddef_setsubdef(), upb_def_dup(), and + * upb_freeze(), any of which the client could call themself. However, since + * the logic for doing so is nontrivial, we provide it here. + * + * The entire operation either succeeds or fails. If the operation fails, + * the symtab is unchanged, false is returned, and status indicates the + * error. The caller passes a ref on all defs to the symtab (even if the + * operation fails). + * + * TODO(haberman): currently failure will leave the symtab unchanged, but may + * leave the defs themselves partially resolved. Does this matter? If so we + * could do a prepass that ensures that all symbols are resolvable and bail + * if not, so we don't mutate anything until we know the operation will + * succeed. + * + * TODO(haberman): since the defs must be mutable, refining a frozen def + * requires making mutable copies of the entire tree. This is wasteful if + * only a few messages are changing. We may want to add a way of adding a + * tree of frozen defs to the symtab (perhaps an alternate constructor where + * you pass the root of the tree?) */ + bool Add(Def*const* defs, size_t n, void* ref_donor, Status* status); + + bool Add(const std::vector<Def*>& defs, void *owner, Status* status) { + return Add((Def*const*)&defs[0], defs.size(), owner, status); + } + + /* Resolves all subdefs for messages in this file and attempts to freeze the + * file. If this succeeds, adds all the symbols to this SymbolTable + * (replacing any existing ones with the same names). */ + bool AddFile(FileDef* file, Status* s); + + private: + UPB_DISALLOW_POD_OPS(SymbolTable, upb::SymbolTable) +}; + +#endif /* __cplusplus */ + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ + +/* Include refcounted methods like upb_symtab_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_symtab, upb_symtab_upcast) + +upb_symtab *upb_symtab_new(const void *owner); +void upb_symtab_freeze(upb_symtab *s); +const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base, + const char *sym); +const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym); +const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym); +const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym); +bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, size_t n, + void *ref_donor, upb_status *status); +bool upb_symtab_addfile(upb_symtab *s, upb_filedef *file, upb_status* status); + +/* upb_symtab_iter i; + * for(upb_symtab_begin(&i, s, type); !upb_symtab_done(&i); + * upb_symtab_next(&i)) { + * const upb_def *def = upb_symtab_iter_def(&i); + * // ... + * } + * + * For C we don't have separate iterators for const and non-const. + * It is the caller's responsibility to cast the upb_fielddef* to + * const if the upb_msgdef* is const. */ +void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s, + upb_deftype_t type); +void upb_symtab_next(upb_symtab_iter *iter); +bool upb_symtab_done(const upb_symtab_iter *iter); +const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter); + +UPB_END_EXTERN_C + +#ifdef __cplusplus +/* C++ inline wrappers. */ +namespace upb { +inline reffed_ptr<SymbolTable> SymbolTable::New() { + upb_symtab *s = upb_symtab_new(&s); + return reffed_ptr<SymbolTable>(s, &s); +} + +inline void SymbolTable::Freeze() { + return upb_symtab_freeze(this); +} +inline const Def *SymbolTable::Resolve(const char *base, + const char *sym) const { + return upb_symtab_resolve(this, base, sym); +} +inline const Def* SymbolTable::Lookup(const char *sym) const { + return upb_symtab_lookup(this, sym); +} +inline const MessageDef *SymbolTable::LookupMessage(const char *sym) const { + return upb_symtab_lookupmsg(this, sym); +} +inline bool SymbolTable::Add( + Def*const* defs, size_t n, void* ref_donor, Status* status) { + return upb_symtab_add(this, (upb_def*const*)defs, n, ref_donor, status); +} +inline bool SymbolTable::AddFile(FileDef* file, Status* s) { + return upb_symtab_addfile(this, file, s); +} +} /* namespace upb */ +#endif + +#endif /* UPB_SYMTAB_H_ */ +/* +** upb::descriptor::Reader (upb_descreader) +** +** Provides a way of building upb::Defs from data in descriptor.proto format. +*/ + +#ifndef UPB_DESCRIPTOR_H +#define UPB_DESCRIPTOR_H + + +#ifdef __cplusplus +namespace upb { +namespace descriptor { +class Reader; +} /* namespace descriptor */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::descriptor::Reader, upb_descreader) + +#ifdef __cplusplus + +/* Class that receives descriptor data according to the descriptor.proto schema + * and use it to build upb::Defs corresponding to that schema. */ +class upb::descriptor::Reader { + public: + /* These handlers must have come from NewHandlers() and must outlive the + * Reader. + * + * TODO: generate the handlers statically (like we do with the + * descriptor.proto defs) so that there is no need to pass this parameter (or + * to build/memory-manage the handlers at runtime at all). Unfortunately this + * is a bit tricky to implement for Handlers, but necessary to simplify this + * interface. */ + static Reader* Create(Environment* env, const Handlers* handlers); + + /* The reader's input; this is where descriptor.proto data should be sent. */ + Sink* input(); + + /* Use to get the FileDefs that have been parsed. */ + size_t file_count() const; + FileDef* file(size_t i) const; + + /* Builds and returns handlers for the reader, owned by "owner." */ + static Handlers* NewHandlers(const void* owner); + + private: + UPB_DISALLOW_POD_OPS(Reader, upb::descriptor::Reader) +}; + +#endif + +UPB_BEGIN_EXTERN_C + +/* C API. */ +upb_descreader *upb_descreader_create(upb_env *e, const upb_handlers *h); +upb_sink *upb_descreader_input(upb_descreader *r); +size_t upb_descreader_filecount(const upb_descreader *r); +upb_filedef *upb_descreader_file(const upb_descreader *r, size_t i); +const upb_handlers *upb_descreader_newhandlers(const void *owner); + +UPB_END_EXTERN_C + +#ifdef __cplusplus +/* C++ implementation details. ************************************************/ +namespace upb { +namespace descriptor { +inline Reader* Reader::Create(Environment* e, const Handlers *h) { + return upb_descreader_create(e, h); +} +inline Sink* Reader::input() { return upb_descreader_input(this); } +inline size_t Reader::file_count() const { + return upb_descreader_filecount(this); +} +inline FileDef* Reader::file(size_t i) const { + return upb_descreader_file(this, i); +} +} /* namespace descriptor */ +} /* namespace upb */ +#endif + +#endif /* UPB_DESCRIPTOR_H */ +/* This file contains accessors for a set of compiled-in defs. + * Note that unlike Google's protobuf, it does *not* define + * generated classes or any other kind of data structure for + * actually storing protobufs. It only contains *defs* which + * let you reflect over a protobuf *schema*. + */ +/* This file was generated by upbc (the upb compiler) from the input + * file: + * + * upb/descriptor/descriptor.proto + * + * Do not edit -- your changes will be discarded when the file is + * regenerated. */ + +#ifndef UPB_DESCRIPTOR_DESCRIPTOR_PROTO_UPB_H_ +#define UPB_DESCRIPTOR_DESCRIPTOR_PROTO_UPB_H_ + + +UPB_BEGIN_EXTERN_C + +/* Enums */ + +typedef enum { + google_protobuf_FieldDescriptorProto_LABEL_OPTIONAL = 1, + google_protobuf_FieldDescriptorProto_LABEL_REQUIRED = 2, + google_protobuf_FieldDescriptorProto_LABEL_REPEATED = 3 +} google_protobuf_FieldDescriptorProto_Label; + +typedef enum { + google_protobuf_FieldDescriptorProto_TYPE_DOUBLE = 1, + google_protobuf_FieldDescriptorProto_TYPE_FLOAT = 2, + google_protobuf_FieldDescriptorProto_TYPE_INT64 = 3, + google_protobuf_FieldDescriptorProto_TYPE_UINT64 = 4, + google_protobuf_FieldDescriptorProto_TYPE_INT32 = 5, + google_protobuf_FieldDescriptorProto_TYPE_FIXED64 = 6, + google_protobuf_FieldDescriptorProto_TYPE_FIXED32 = 7, + google_protobuf_FieldDescriptorProto_TYPE_BOOL = 8, + google_protobuf_FieldDescriptorProto_TYPE_STRING = 9, + google_protobuf_FieldDescriptorProto_TYPE_GROUP = 10, + google_protobuf_FieldDescriptorProto_TYPE_MESSAGE = 11, + google_protobuf_FieldDescriptorProto_TYPE_BYTES = 12, + google_protobuf_FieldDescriptorProto_TYPE_UINT32 = 13, + google_protobuf_FieldDescriptorProto_TYPE_ENUM = 14, + google_protobuf_FieldDescriptorProto_TYPE_SFIXED32 = 15, + google_protobuf_FieldDescriptorProto_TYPE_SFIXED64 = 16, + google_protobuf_FieldDescriptorProto_TYPE_SINT32 = 17, + google_protobuf_FieldDescriptorProto_TYPE_SINT64 = 18 +} google_protobuf_FieldDescriptorProto_Type; + +typedef enum { + google_protobuf_FieldOptions_STRING = 0, + google_protobuf_FieldOptions_CORD = 1, + google_protobuf_FieldOptions_STRING_PIECE = 2 +} google_protobuf_FieldOptions_CType; + +typedef enum { + google_protobuf_FieldOptions_JS_NORMAL = 0, + google_protobuf_FieldOptions_JS_STRING = 1, + google_protobuf_FieldOptions_JS_NUMBER = 2 +} google_protobuf_FieldOptions_JSType; + +typedef enum { + google_protobuf_FileOptions_SPEED = 1, + google_protobuf_FileOptions_CODE_SIZE = 2, + google_protobuf_FileOptions_LITE_RUNTIME = 3 +} google_protobuf_FileOptions_OptimizeMode; + +/* MessageDefs: call these functions to get a ref to a msgdef. */ +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ReservedRange_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_EnumDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_EnumOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_EnumValueDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_EnumValueOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_FieldDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_FieldOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_FileDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_FileDescriptorSet_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_FileOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_MessageOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_MethodDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_MethodOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_OneofDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_ServiceDescriptorProto_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_ServiceOptions_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_Location_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_get(const void *owner); +const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_NamePart_get(const void *owner); + +/* EnumDefs: call these functions to get a ref to an enumdef. */ +const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Label_get(const void *owner); +const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Type_get(const void *owner); +const upb_enumdef *upbdefs_google_protobuf_FieldOptions_CType_get(const void *owner); +const upb_enumdef *upbdefs_google_protobuf_FieldOptions_JSType_get(const void *owner); +const upb_enumdef *upbdefs_google_protobuf_FileOptions_OptimizeMode_get(const void *owner); + +/* Functions to test whether this message is of a certain type. */ +UPB_INLINE bool upbdefs_google_protobuf_DescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.DescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_DescriptorProto_ExtensionRange_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.DescriptorProto.ExtensionRange") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_DescriptorProto_ReservedRange_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.DescriptorProto.ReservedRange") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_EnumDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.EnumDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_EnumOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.EnumOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_EnumValueDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.EnumValueDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_EnumValueOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.EnumValueOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FieldDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.FieldDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FieldOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.FieldOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FileDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.FileDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FileDescriptorSet_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.FileDescriptorSet") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FileOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.FileOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_MessageOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.MessageOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_MethodDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.MethodDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_MethodOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.MethodOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_OneofDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.OneofDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_ServiceDescriptorProto_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.ServiceDescriptorProto") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_ServiceOptions_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.ServiceOptions") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_SourceCodeInfo_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.SourceCodeInfo") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_SourceCodeInfo_Location_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.SourceCodeInfo.Location") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_UninterpretedOption_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.UninterpretedOption") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_UninterpretedOption_NamePart_is(const upb_msgdef *m) { + return strcmp(upb_msgdef_fullname(m), "google.protobuf.UninterpretedOption.NamePart") == 0; +} + +/* Functions to test whether this enum is of a certain type. */ +UPB_INLINE bool upbdefs_google_protobuf_FieldDescriptorProto_Label_is(const upb_enumdef *e) { + return strcmp(upb_enumdef_fullname(e), "google.protobuf.FieldDescriptorProto.Label") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FieldDescriptorProto_Type_is(const upb_enumdef *e) { + return strcmp(upb_enumdef_fullname(e), "google.protobuf.FieldDescriptorProto.Type") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FieldOptions_CType_is(const upb_enumdef *e) { + return strcmp(upb_enumdef_fullname(e), "google.protobuf.FieldOptions.CType") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FieldOptions_JSType_is(const upb_enumdef *e) { + return strcmp(upb_enumdef_fullname(e), "google.protobuf.FieldOptions.JSType") == 0; +} +UPB_INLINE bool upbdefs_google_protobuf_FileOptions_OptimizeMode_is(const upb_enumdef *e) { + return strcmp(upb_enumdef_fullname(e), "google.protobuf.FileOptions.OptimizeMode") == 0; +} + + +/* Functions to get a fielddef from a msgdef reference. */ +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_f_end(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_ExtensionRange_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_f_start(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_ExtensionRange_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ReservedRange_f_end(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_ReservedRange_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ReservedRange_f_start(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_ReservedRange_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_enum_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_extension(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_extension_range(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_field(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_nested_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_oneof_decl(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_reserved_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_f_reserved_range(const upb_msgdef *m) { assert(upbdefs_google_protobuf_DescriptorProto_is(m)); return upb_msgdef_itof(m, 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_f_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_f_allow_alias(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumOptions_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumOptions_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumValueDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_f_number(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumValueDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumValueDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumValueOptions_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_EnumValueOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_default_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_extendee(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_json_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_label(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_number(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_oneof_index(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_f_type_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_ctype(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_jstype(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_lazy(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_packed(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_f_weak(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FieldOptions_is(m)); return upb_msgdef_itof(m, 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_dependency(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_enum_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_extension(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_message_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_package(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_public_dependency(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_service(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_source_code_info(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_syntax(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 12); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_f_weak_dependency(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); return upb_msgdef_itof(m, 11); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorSet_f_file(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileDescriptorSet_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_cc_enable_arenas(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 31); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_cc_generic_services(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 16); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_csharp_namespace(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 37); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 23); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_go_package(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 11); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_generate_equals_and_hash(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 20); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_generic_services(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 17); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_multiple_files(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_outer_classname(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_package(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_java_string_check_utf8(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 27); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_javanano_use_deprecated_package(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 38); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_objc_class_prefix(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 36); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_optimize_for(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_py_generic_services(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 18); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_FileOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MessageOptions_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_f_map_entry(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MessageOptions_is(m)); return upb_msgdef_itof(m, 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_f_message_set_wire_format(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MessageOptions_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_f_no_standard_descriptor_accessor(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MessageOptions_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MessageOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_client_streaming(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_input_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_output_type(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_f_server_streaming(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodOptions_is(m)); return upb_msgdef_itof(m, 33); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_MethodOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_OneofDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_OneofDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_f_method(const upb_msgdef *m) { assert(upbdefs_google_protobuf_ServiceDescriptorProto_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_ServiceDescriptorProto_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_f_options(const upb_msgdef *m) { assert(upbdefs_google_protobuf_ServiceDescriptorProto_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceOptions_f_deprecated(const upb_msgdef *m) { assert(upbdefs_google_protobuf_ServiceOptions_is(m)); return upb_msgdef_itof(m, 33); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceOptions_f_uninterpreted_option(const upb_msgdef *m) { assert(upbdefs_google_protobuf_ServiceOptions_is(m)); return upb_msgdef_itof(m, 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_f_leading_comments(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_f_leading_detached_comments(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_f_path(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_f_span(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_f_trailing_comments(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_f_location(const upb_msgdef *m) { assert(upbdefs_google_protobuf_SourceCodeInfo_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_f_is_extension(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_NamePart_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_f_name_part(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_NamePart_is(m)); return upb_msgdef_itof(m, 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_aggregate_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_double_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_identifier_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_name(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_negative_int_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_positive_int_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_f_string_value(const upb_msgdef *m) { assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); return upb_msgdef_itof(m, 7); } + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upbdefs { +namespace google { +namespace protobuf { + +class DescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + DescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_DescriptorProto_is(m)); + } + + static DescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_DescriptorProto_get(&m); + return DescriptorProto(m, &m); + } + + class ExtensionRange : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + ExtensionRange(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_DescriptorProto_ExtensionRange_is(m)); + } + + static ExtensionRange get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_DescriptorProto_ExtensionRange_get(&m); + return ExtensionRange(m, &m); + } + }; + + class ReservedRange : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + ReservedRange(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_DescriptorProto_ReservedRange_is(m)); + } + + static ReservedRange get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_DescriptorProto_ReservedRange_get(&m); + return ReservedRange(m, &m); + } + }; +}; + +class EnumDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + EnumDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_EnumDescriptorProto_is(m)); + } + + static EnumDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_EnumDescriptorProto_get(&m); + return EnumDescriptorProto(m, &m); + } +}; + +class EnumOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + EnumOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_EnumOptions_is(m)); + } + + static EnumOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_EnumOptions_get(&m); + return EnumOptions(m, &m); + } +}; + +class EnumValueDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + EnumValueDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_EnumValueDescriptorProto_is(m)); + } + + static EnumValueDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_EnumValueDescriptorProto_get(&m); + return EnumValueDescriptorProto(m, &m); + } +}; + +class EnumValueOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + EnumValueOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_EnumValueOptions_is(m)); + } + + static EnumValueOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_EnumValueOptions_get(&m); + return EnumValueOptions(m, &m); + } +}; + +class FieldDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + FieldDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_FieldDescriptorProto_is(m)); + } + + static FieldDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_FieldDescriptorProto_get(&m); + return FieldDescriptorProto(m, &m); + } + + class Label : public ::upb::reffed_ptr<const ::upb::EnumDef> { + public: + Label(const ::upb::EnumDef* e, const void *ref_donor = NULL) + : reffed_ptr(e, ref_donor) { + assert(upbdefs_google_protobuf_FieldDescriptorProto_Label_is(e)); + } + static Label get() { + const ::upb::EnumDef* e = upbdefs_google_protobuf_FieldDescriptorProto_Label_get(&e); + return Label(e, &e); + } + }; + + class Type : public ::upb::reffed_ptr<const ::upb::EnumDef> { + public: + Type(const ::upb::EnumDef* e, const void *ref_donor = NULL) + : reffed_ptr(e, ref_donor) { + assert(upbdefs_google_protobuf_FieldDescriptorProto_Type_is(e)); + } + static Type get() { + const ::upb::EnumDef* e = upbdefs_google_protobuf_FieldDescriptorProto_Type_get(&e); + return Type(e, &e); + } + }; +}; + +class FieldOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + FieldOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_FieldOptions_is(m)); + } + + static FieldOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_FieldOptions_get(&m); + return FieldOptions(m, &m); + } + + class CType : public ::upb::reffed_ptr<const ::upb::EnumDef> { + public: + CType(const ::upb::EnumDef* e, const void *ref_donor = NULL) + : reffed_ptr(e, ref_donor) { + assert(upbdefs_google_protobuf_FieldOptions_CType_is(e)); + } + static CType get() { + const ::upb::EnumDef* e = upbdefs_google_protobuf_FieldOptions_CType_get(&e); + return CType(e, &e); + } + }; + + class JSType : public ::upb::reffed_ptr<const ::upb::EnumDef> { + public: + JSType(const ::upb::EnumDef* e, const void *ref_donor = NULL) + : reffed_ptr(e, ref_donor) { + assert(upbdefs_google_protobuf_FieldOptions_JSType_is(e)); + } + static JSType get() { + const ::upb::EnumDef* e = upbdefs_google_protobuf_FieldOptions_JSType_get(&e); + return JSType(e, &e); + } + }; +}; + +class FileDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + FileDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_FileDescriptorProto_is(m)); + } + + static FileDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_FileDescriptorProto_get(&m); + return FileDescriptorProto(m, &m); + } +}; + +class FileDescriptorSet : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + FileDescriptorSet(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_FileDescriptorSet_is(m)); + } + + static FileDescriptorSet get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_FileDescriptorSet_get(&m); + return FileDescriptorSet(m, &m); + } +}; + +class FileOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + FileOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_FileOptions_is(m)); + } + + static FileOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_FileOptions_get(&m); + return FileOptions(m, &m); + } + + class OptimizeMode : public ::upb::reffed_ptr<const ::upb::EnumDef> { + public: + OptimizeMode(const ::upb::EnumDef* e, const void *ref_donor = NULL) + : reffed_ptr(e, ref_donor) { + assert(upbdefs_google_protobuf_FileOptions_OptimizeMode_is(e)); + } + static OptimizeMode get() { + const ::upb::EnumDef* e = upbdefs_google_protobuf_FileOptions_OptimizeMode_get(&e); + return OptimizeMode(e, &e); + } + }; +}; + +class MessageOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + MessageOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_MessageOptions_is(m)); + } + + static MessageOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_MessageOptions_get(&m); + return MessageOptions(m, &m); + } +}; + +class MethodDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + MethodDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_MethodDescriptorProto_is(m)); + } + + static MethodDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_MethodDescriptorProto_get(&m); + return MethodDescriptorProto(m, &m); + } +}; + +class MethodOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + MethodOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_MethodOptions_is(m)); + } + + static MethodOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_MethodOptions_get(&m); + return MethodOptions(m, &m); + } +}; + +class OneofDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + OneofDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_OneofDescriptorProto_is(m)); + } + + static OneofDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_OneofDescriptorProto_get(&m); + return OneofDescriptorProto(m, &m); + } +}; + +class ServiceDescriptorProto : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + ServiceDescriptorProto(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_ServiceDescriptorProto_is(m)); + } + + static ServiceDescriptorProto get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_ServiceDescriptorProto_get(&m); + return ServiceDescriptorProto(m, &m); + } +}; + +class ServiceOptions : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + ServiceOptions(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_ServiceOptions_is(m)); + } + + static ServiceOptions get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_ServiceOptions_get(&m); + return ServiceOptions(m, &m); + } +}; + +class SourceCodeInfo : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + SourceCodeInfo(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_SourceCodeInfo_is(m)); + } + + static SourceCodeInfo get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_SourceCodeInfo_get(&m); + return SourceCodeInfo(m, &m); + } + + class Location : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + Location(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_SourceCodeInfo_Location_is(m)); + } + + static Location get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_SourceCodeInfo_Location_get(&m); + return Location(m, &m); + } + }; +}; + +class UninterpretedOption : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + UninterpretedOption(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_UninterpretedOption_is(m)); + } + + static UninterpretedOption get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_UninterpretedOption_get(&m); + return UninterpretedOption(m, &m); + } + + class NamePart : public ::upb::reffed_ptr<const ::upb::MessageDef> { + public: + NamePart(const ::upb::MessageDef* m, const void *ref_donor = NULL) + : reffed_ptr(m, ref_donor) { + assert(upbdefs_google_protobuf_UninterpretedOption_NamePart_is(m)); + } + + static NamePart get() { + const ::upb::MessageDef* m = upbdefs_google_protobuf_UninterpretedOption_NamePart_get(&m); + return NamePart(m, &m); + } + }; +}; + +} /* namespace protobuf */ +} /* namespace google */ +} /* namespace upbdefs */ + +#endif /* __cplusplus */ + +#endif /* UPB_DESCRIPTOR_DESCRIPTOR_PROTO_UPB_H_ */ +/* +** Internal-only definitions for the decoder. +*/ + +#ifndef UPB_DECODER_INT_H_ +#define UPB_DECODER_INT_H_ + +/* +** upb::pb::Decoder +** +** A high performance, streaming, resumable decoder for the binary protobuf +** format. +** +** This interface works the same regardless of what decoder backend is being +** used. A client of this class does not need to know whether decoding is using +** a JITted decoder (DynASM, LLVM, etc) or an interpreted decoder. By default, +** it will always use the fastest available decoder. However, you can call +** set_allow_jit(false) to disable any JIT decoder that might be available. +** This is primarily useful for testing purposes. +*/ + +#ifndef UPB_DECODER_H_ +#define UPB_DECODER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class CodeCache; +class Decoder; +class DecoderMethod; +class DecoderMethodOptions; +} /* namespace pb */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::pb::CodeCache, upb_pbcodecache) +UPB_DECLARE_TYPE(upb::pb::Decoder, upb_pbdecoder) +UPB_DECLARE_TYPE(upb::pb::DecoderMethodOptions, upb_pbdecodermethodopts) + +UPB_DECLARE_DERIVED_TYPE(upb::pb::DecoderMethod, upb::RefCounted, + upb_pbdecodermethod, upb_refcounted) + +/* The maximum number of bytes we are required to buffer internally between + * calls to the decoder. The value is 14: a 5 byte unknown tag plus ten-byte + * varint, less one because we are buffering an incomplete value. + * + * Should only be used by unit tests. */ +#define UPB_DECODER_MAX_RESIDUAL_BYTES 14 + +#ifdef __cplusplus + +/* The parameters one uses to construct a DecoderMethod. + * TODO(haberman): move allowjit here? Seems more convenient for users. + * TODO(haberman): move this to be heap allocated for ABI stability. */ +class upb::pb::DecoderMethodOptions { + public: + /* Parameter represents the destination handlers that this method will push + * to. */ + explicit DecoderMethodOptions(const Handlers* dest_handlers); + + /* Should the decoder push submessages to lazy handlers for fields that have + * them? The caller should set this iff the lazy handlers expect data that is + * in protobuf binary format and the caller wishes to lazy parse it. */ + void set_lazy(bool lazy); +#else +struct upb_pbdecodermethodopts { +#endif + const upb_handlers *handlers; + bool lazy; +}; + +#ifdef __cplusplus + +/* Represents the code to parse a protobuf according to a destination + * Handlers. */ +class upb::pb::DecoderMethod { + public: + /* Include base methods from upb::ReferenceCounted. */ + UPB_REFCOUNTED_CPPMETHODS + + /* The destination handlers that are statically bound to this method. + * This method is only capable of outputting to a sink that uses these + * handlers. */ + const Handlers* dest_handlers() const; + + /* The input handlers for this decoder method. */ + const BytesHandler* input_handler() const; + + /* Whether this method is native. */ + bool is_native() const; + + /* Convenience method for generating a DecoderMethod without explicitly + * creating a CodeCache. */ + static reffed_ptr<const DecoderMethod> New(const DecoderMethodOptions& opts); + + private: + UPB_DISALLOW_POD_OPS(DecoderMethod, upb::pb::DecoderMethod) +}; + +#endif + +/* Preallocation hint: decoder won't allocate more bytes than this when first + * constructed. This hint may be an overestimate for some build configurations. + * But if the decoder library is upgraded without recompiling the application, + * it may be an underestimate. */ +#define UPB_PB_DECODER_SIZE 4416 + +#ifdef __cplusplus + +/* A Decoder receives binary protobuf data on its input sink and pushes the + * decoded data to its output sink. */ +class upb::pb::Decoder { + public: + /* Constructs a decoder instance for the given method, which must outlive this + * decoder. Any errors during parsing will be set on the given status, which + * must also outlive this decoder. + * + * The sink must match the given method. */ + static Decoder* Create(Environment* env, const DecoderMethod* method, + Sink* output); + + /* Returns the DecoderMethod this decoder is parsing from. */ + const DecoderMethod* method() const; + + /* The sink on which this decoder receives input. */ + BytesSink* input(); + + /* Returns number of bytes successfully parsed. + * + * This can be useful for determining the stream position where an error + * occurred. + * + * This value may not be up-to-date when called from inside a parsing + * callback. */ + uint64_t BytesParsed() const; + + /* Gets/sets the parsing nexting limit. If the total number of nested + * submessages and repeated fields hits this limit, parsing will fail. This + * is a resource limit that controls the amount of memory used by the parsing + * stack. + * + * Setting the limit will fail if the parser is currently suspended at a depth + * greater than this, or if memory allocation of the stack fails. */ + size_t max_nesting() const; + bool set_max_nesting(size_t max); + + void Reset(); + + static const size_t kSize = UPB_PB_DECODER_SIZE; + + private: + UPB_DISALLOW_POD_OPS(Decoder, upb::pb::Decoder) +}; + +#endif /* __cplusplus */ + +#ifdef __cplusplus + +/* A class for caching protobuf processing code, whether bytecode for the + * interpreted decoder or machine code for the JIT. + * + * This class is not thread-safe. + * + * TODO(haberman): move this to be heap allocated for ABI stability. */ +class upb::pb::CodeCache { + public: + CodeCache(); + ~CodeCache(); + + /* Whether the cache is allowed to generate machine code. Defaults to true. + * There is no real reason to turn it off except for testing or if you are + * having a specific problem with the JIT. + * + * Note that allow_jit = true does not *guarantee* that the code will be JIT + * compiled. If this platform is not supported or the JIT was not compiled + * in, the code may still be interpreted. */ + bool allow_jit() const; + + /* This may only be called when the object is first constructed, and prior to + * any code generation, otherwise returns false and does nothing. */ + bool set_allow_jit(bool allow); + + /* Returns a DecoderMethod that can push data to the given handlers. + * If a suitable method already exists, it will be returned from the cache. + * + * Specifying the destination handlers here allows the DecoderMethod to be + * statically bound to the destination handlers if possible, which can allow + * more efficient decoding. However the returned method may or may not + * actually be statically bound. But in all cases, the returned method can + * push data to the given handlers. */ + const DecoderMethod *GetDecoderMethod(const DecoderMethodOptions& opts); + + /* If/when someone needs to explicitly create a dynamically-bound + * DecoderMethod*, we can add a method to get it here. */ + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(CodeCache) +#else +struct upb_pbcodecache { +#endif + bool allow_jit_; + + /* Array of mgroups. */ + upb_inttable groups; +}; + +UPB_BEGIN_EXTERN_C + +upb_pbdecoder *upb_pbdecoder_create(upb_env *e, + const upb_pbdecodermethod *method, + upb_sink *output); +const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d); +upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d); +uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d); +size_t upb_pbdecoder_maxnesting(const upb_pbdecoder *d); +bool upb_pbdecoder_setmaxnesting(upb_pbdecoder *d, size_t max); +void upb_pbdecoder_reset(upb_pbdecoder *d); + +void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts, + const upb_handlers *h); +void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy); + + +/* Include refcounted methods like upb_pbdecodermethod_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_pbdecodermethod, upb_pbdecodermethod_upcast) + +const upb_handlers *upb_pbdecodermethod_desthandlers( + const upb_pbdecodermethod *m); +const upb_byteshandler *upb_pbdecodermethod_inputhandler( + const upb_pbdecodermethod *m); +bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m); +const upb_pbdecodermethod *upb_pbdecodermethod_new( + const upb_pbdecodermethodopts *opts, const void *owner); + +void upb_pbcodecache_init(upb_pbcodecache *c); +void upb_pbcodecache_uninit(upb_pbcodecache *c); +bool upb_pbcodecache_allowjit(const upb_pbcodecache *c); +bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow); +const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod( + upb_pbcodecache *c, const upb_pbdecodermethodopts *opts); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { + +namespace pb { + +/* static */ +inline Decoder* Decoder::Create(Environment* env, const DecoderMethod* m, + Sink* sink) { + return upb_pbdecoder_create(env, m, sink); +} +inline const DecoderMethod* Decoder::method() const { + return upb_pbdecoder_method(this); +} +inline BytesSink* Decoder::input() { + return upb_pbdecoder_input(this); +} +inline uint64_t Decoder::BytesParsed() const { + return upb_pbdecoder_bytesparsed(this); +} +inline size_t Decoder::max_nesting() const { + return upb_pbdecoder_maxnesting(this); +} +inline bool Decoder::set_max_nesting(size_t max) { + return upb_pbdecoder_setmaxnesting(this, max); +} +inline void Decoder::Reset() { upb_pbdecoder_reset(this); } + +inline DecoderMethodOptions::DecoderMethodOptions(const Handlers* h) { + upb_pbdecodermethodopts_init(this, h); +} +inline void DecoderMethodOptions::set_lazy(bool lazy) { + upb_pbdecodermethodopts_setlazy(this, lazy); +} + +inline const Handlers* DecoderMethod::dest_handlers() const { + return upb_pbdecodermethod_desthandlers(this); +} +inline const BytesHandler* DecoderMethod::input_handler() const { + return upb_pbdecodermethod_inputhandler(this); +} +inline bool DecoderMethod::is_native() const { + return upb_pbdecodermethod_isnative(this); +} +/* static */ +inline reffed_ptr<const DecoderMethod> DecoderMethod::New( + const DecoderMethodOptions &opts) { + const upb_pbdecodermethod *m = upb_pbdecodermethod_new(&opts, &m); + return reffed_ptr<const DecoderMethod>(m, &m); +} + +inline CodeCache::CodeCache() { + upb_pbcodecache_init(this); +} +inline CodeCache::~CodeCache() { + upb_pbcodecache_uninit(this); +} +inline bool CodeCache::allow_jit() const { + return upb_pbcodecache_allowjit(this); +} +inline bool CodeCache::set_allow_jit(bool allow) { + return upb_pbcodecache_setallowjit(this, allow); +} +inline const DecoderMethod *CodeCache::GetDecoderMethod( + const DecoderMethodOptions& opts) { + return upb_pbcodecache_getdecodermethod(this, &opts); +} + +} /* namespace pb */ +} /* namespace upb */ + +#endif /* __cplusplus */ + +#endif /* UPB_DECODER_H_ */ + +/* C++ names are not actually used since this type isn't exposed to users. */ +#ifdef __cplusplus +namespace upb { +namespace pb { +class MessageGroup; +} /* namespace pb */ +} /* namespace upb */ +#endif +UPB_DECLARE_DERIVED_TYPE(upb::pb::MessageGroup, upb::RefCounted, + mgroup, upb_refcounted) + +/* Opcode definitions. The canonical meaning of each opcode is its + * implementation in the interpreter (the JIT is written to match this). + * + * All instructions have the opcode in the low byte. + * Instruction format for most instructions is: + * + * +-------------------+--------+ + * | arg (24) | op (8) | + * +-------------------+--------+ + * + * Exceptions are indicated below. A few opcodes are multi-word. */ +typedef enum { + /* Opcodes 1-8, 13, 15-18 parse their respective descriptor types. + * Arg for all of these is the upb selector for this field. */ +#define T(type) OP_PARSE_ ## type = UPB_DESCRIPTOR_TYPE_ ## type + T(DOUBLE), T(FLOAT), T(INT64), T(UINT64), T(INT32), T(FIXED64), T(FIXED32), + T(BOOL), T(UINT32), T(SFIXED32), T(SFIXED64), T(SINT32), T(SINT64), +#undef T + OP_STARTMSG = 9, /* No arg. */ + OP_ENDMSG = 10, /* No arg. */ + OP_STARTSEQ = 11, + OP_ENDSEQ = 12, + OP_STARTSUBMSG = 14, + OP_ENDSUBMSG = 19, + OP_STARTSTR = 20, + OP_STRING = 21, + OP_ENDSTR = 22, + + OP_PUSHTAGDELIM = 23, /* No arg. */ + OP_PUSHLENDELIM = 24, /* No arg. */ + OP_POP = 25, /* No arg. */ + OP_SETDELIM = 26, /* No arg. */ + OP_SETBIGGROUPNUM = 27, /* two words: + * | unused (24) | opc (8) | + * | groupnum (32) | */ + OP_CHECKDELIM = 28, + OP_CALL = 29, + OP_RET = 30, + OP_BRANCH = 31, + + /* Different opcodes depending on how many bytes expected. */ + OP_TAG1 = 32, /* | match tag (16) | jump target (8) | opc (8) | */ + OP_TAG2 = 33, /* | match tag (16) | jump target (8) | opc (8) | */ + OP_TAGN = 34, /* three words: */ + /* | unused (16) | jump target(8) | opc (8) | */ + /* | match tag 1 (32) | */ + /* | match tag 2 (32) | */ + + OP_SETDISPATCH = 35, /* N words: */ + /* | unused (24) | opc | */ + /* | upb_inttable* (32 or 64) | */ + + OP_DISPATCH = 36, /* No arg. */ + + OP_HALT = 37 /* No arg. */ +} opcode; + +#define OP_MAX OP_HALT + +UPB_INLINE opcode getop(uint32_t instr) { return instr & 0xff; } + +/* Method group; represents a set of decoder methods that had their code + * emitted together, and must therefore be freed together. Immutable once + * created. It is possible we may want to expose this to users at some point. + * + * Overall ownership of Decoder objects looks like this: + * + * +----------+ + * | | <---> DecoderMethod + * | method | + * CodeCache ---> | group | <---> DecoderMethod + * | | + * | (mgroup) | <---> DecoderMethod + * +----------+ + */ +struct mgroup { + upb_refcounted base; + + /* Maps upb_msgdef/upb_handlers -> upb_pbdecodermethod. We own refs on the + * methods. */ + upb_inttable methods; + + /* When we add the ability to link to previously existing mgroups, we'll + * need an array of mgroups we reference here, and own refs on them. */ + + /* The bytecode for our methods, if any exists. Owned by us. */ + uint32_t *bytecode; + uint32_t *bytecode_end; + +#ifdef UPB_USE_JIT_X64 + /* JIT-generated machine code, if any. */ + upb_string_handlerfunc *jit_code; + /* The size of the jit_code (required to munmap()). */ + size_t jit_size; + char *debug_info; + void *dl; +#endif +}; + +/* The maximum that any submessages can be nested. Matches proto2's limit. + * This specifies the size of the decoder's statically-sized array and therefore + * setting it high will cause the upb::pb::Decoder object to be larger. + * + * If necessary we can add a runtime-settable property to Decoder that allow + * this to be larger than the compile-time setting, but this would add + * complexity, particularly since we would have to decide how/if to give users + * the ability to set a custom memory allocation function. */ +#define UPB_DECODER_MAX_NESTING 64 + +/* Internal-only struct used by the decoder. */ +typedef struct { + /* Space optimization note: we store two pointers here that the JIT + * doesn't need at all; the upb_handlers* inside the sink and + * the dispatch table pointer. We can optimze so that the JIT uses + * smaller stack frames than the interpreter. The only thing we need + * to guarantee is that the fallback routines can find end_ofs. */ + upb_sink sink; + + /* The absolute stream offset of the end-of-frame delimiter. + * Non-delimited frames (groups and non-packed repeated fields) reuse the + * delimiter of their parent, even though the frame may not end there. + * + * NOTE: the JIT stores a slightly different value here for non-top frames. + * It stores the value relative to the end of the enclosed message. But the + * top frame is still stored the same way, which is important for ensuring + * that calls from the JIT into C work correctly. */ + uint64_t end_ofs; + const uint32_t *base; + + /* 0 indicates a length-delimited field. + * A positive number indicates a known group. + * A negative number indicates an unknown group. */ + int32_t groupnum; + upb_inttable *dispatch; /* Not used by the JIT. */ +} upb_pbdecoder_frame; + +struct upb_pbdecodermethod { + upb_refcounted base; + + /* While compiling, the base is relative in "ofs", after compiling it is + * absolute in "ptr". */ + union { + uint32_t ofs; /* PC offset of method. */ + void *ptr; /* Pointer to bytecode or machine code for this method. */ + } code_base; + + /* The decoder method group to which this method belongs. We own a ref. + * Owning a ref on the entire group is more coarse-grained than is strictly + * necessary; all we truly require is that methods we directly reference + * outlive us, while the group could contain many other messages we don't + * require. But the group represents the messages that were + * allocated+compiled together, so it makes the most sense to free them + * together also. */ + const upb_refcounted *group; + + /* Whether this method is native code or bytecode. */ + bool is_native_; + + /* The handler one calls to invoke this method. */ + upb_byteshandler input_handler_; + + /* The destination handlers this method is bound to. We own a ref. */ + const upb_handlers *dest_handlers_; + + /* Dispatch table -- used by both bytecode decoder and JIT when encountering a + * field number that wasn't the one we were expecting to see. See + * decoder.int.h for the layout of this table. */ + upb_inttable dispatch; +}; + +struct upb_pbdecoder { + upb_env *env; + + /* Our input sink. */ + upb_bytessink input_; + + /* The decoder method we are parsing with (owned). */ + const upb_pbdecodermethod *method_; + + size_t call_len; + const uint32_t *pc, *last; + + /* Current input buffer and its stream offset. */ + const char *buf, *ptr, *end, *checkpoint; + + /* End of the delimited region, relative to ptr, NULL if not in this buf. */ + const char *delim_end; + + /* End of the delimited region, relative to ptr, end if not in this buf. */ + const char *data_end; + + /* Overall stream offset of "buf." */ + uint64_t bufstart_ofs; + + /* Buffer for residual bytes not parsed from the previous buffer. */ + char residual[UPB_DECODER_MAX_RESIDUAL_BYTES]; + char *residual_end; + + /* Bytes of data that should be discarded from the input beore we start + * parsing again. We set this when we internally determine that we can + * safely skip the next N bytes, but this region extends past the current + * user buffer. */ + size_t skip; + + /* Stores the user buffer passed to our decode function. */ + const char *buf_param; + size_t size_param; + const upb_bufhandle *handle; + + /* Our internal stack. */ + upb_pbdecoder_frame *stack, *top, *limit; + const uint32_t **callstack; + size_t stack_size; + + upb_status *status; + +#ifdef UPB_USE_JIT_X64 + /* Used momentarily by the generated code to store a value while a user + * function is called. */ + uint32_t tmp_len; + + const void *saved_rsp; +#endif +}; + +/* Decoder entry points; used as handlers. */ +void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint); +void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint); +size_t upb_pbdecoder_decode(void *closure, const void *hd, const char *buf, + size_t size, const upb_bufhandle *handle); +bool upb_pbdecoder_end(void *closure, const void *handler_data); + +/* Decoder-internal functions that the JIT calls to handle fallback paths. */ +int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf, + size_t size, const upb_bufhandle *handle); +size_t upb_pbdecoder_suspend(upb_pbdecoder *d); +int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum, + uint8_t wire_type); +int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d, uint64_t expected); +int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d, uint64_t *u64); +int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32); +int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64); +void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg); + +/* Error messages that are shared between the bytecode and JIT decoders. */ +extern const char *kPbDecoderStackOverflow; +extern const char *kPbDecoderSubmessageTooLong; + +/* Access to decoderplan members needed by the decoder. */ +const char *upb_pbdecoder_getopname(unsigned int op); + +/* JIT codegen entry point. */ +void upb_pbdecoder_jit(mgroup *group); +void upb_pbdecoder_freejit(mgroup *group); +UPB_REFCOUNTED_CMETHODS(mgroup, mgroup_upcast) + +/* A special label that means "do field dispatch for this message and branch to + * wherever that takes you." */ +#define LABEL_DISPATCH 0 + +/* A special slot in the dispatch table that stores the epilogue (ENDMSG and/or + * RET) for branching to when we find an appropriate ENDGROUP tag. */ +#define DISPATCH_ENDMSG 0 + +/* It's important to use this invalid wire type instead of 0 (which is a valid + * wire type). */ +#define NO_WIRE_TYPE 0xff + +/* The dispatch table layout is: + * [field number] -> [ 48-bit offset ][ 8-bit wt2 ][ 8-bit wt1 ] + * + * If wt1 matches, jump to the 48-bit offset. If wt2 matches, lookup + * (UPB_MAX_FIELDNUMBER + fieldnum) and jump there. + * + * We need two wire types because of packed/non-packed compatibility. A + * primitive repeated field can use either wire type and be valid. While we + * could key the table on fieldnum+wiretype, the table would be 8x sparser. + * + * Storing two wire types in the primary value allows us to quickly rule out + * the second wire type without needing to do a separate lookup (this case is + * less common than an unknown field). */ +UPB_INLINE uint64_t upb_pbdecoder_packdispatch(uint64_t ofs, uint8_t wt1, + uint8_t wt2) { + return (ofs << 16) | (wt2 << 8) | wt1; +} + +UPB_INLINE void upb_pbdecoder_unpackdispatch(uint64_t dispatch, uint64_t *ofs, + uint8_t *wt1, uint8_t *wt2) { + *wt1 = (uint8_t)dispatch; + *wt2 = (uint8_t)(dispatch >> 8); + *ofs = dispatch >> 16; +} + +/* All of the functions in decoder.c that return int32_t return values according + * to the following scheme: + * 1. negative values indicate a return code from the following list. + * 2. positive values indicate that error or end of buffer was hit, and + * that the decode function should immediately return the given value + * (the decoder state has already been suspended and is ready to be + * resumed). */ +#define DECODE_OK -1 +#define DECODE_MISMATCH -2 /* Used only from checktag_slow(). */ +#define DECODE_ENDGROUP -3 /* Used only from checkunknown(). */ + +#define CHECK_RETURN(x) { int32_t ret = x; if (ret >= 0) return ret; } + +#endif /* UPB_DECODER_INT_H_ */ +/* +** A number of routines for varint manipulation (we keep them all around to +** have multiple approaches available for benchmarking). +*/ + +#ifndef UPB_VARINT_DECODER_H_ +#define UPB_VARINT_DECODER_H_ + +#include <assert.h> +#include <stdint.h> +#include <string.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* A list of types as they are encoded on-the-wire. */ +typedef enum { + UPB_WIRE_TYPE_VARINT = 0, + UPB_WIRE_TYPE_64BIT = 1, + UPB_WIRE_TYPE_DELIMITED = 2, + UPB_WIRE_TYPE_START_GROUP = 3, + UPB_WIRE_TYPE_END_GROUP = 4, + UPB_WIRE_TYPE_32BIT = 5 +} upb_wiretype_t; + +#define UPB_MAX_WIRE_TYPE 5 + +/* The maximum number of bytes that it takes to encode a 64-bit varint. + * Note that with a better encoding this could be 9 (TODO: write up a + * wiki document about this). */ +#define UPB_PB_VARINT_MAX_LEN 10 + +/* Array of the "native" (ie. non-packed-repeated) wire type for the given a + * descriptor type (upb_descriptortype_t). */ +extern const uint8_t upb_pb_native_wire_types[]; + +/* Zig-zag encoding/decoding **************************************************/ + +UPB_INLINE int32_t upb_zzdec_32(uint32_t n) { + return (n >> 1) ^ -(int32_t)(n & 1); +} +UPB_INLINE int64_t upb_zzdec_64(uint64_t n) { + return (n >> 1) ^ -(int64_t)(n & 1); +} +UPB_INLINE uint32_t upb_zzenc_32(int32_t n) { return (n << 1) ^ (n >> 31); } +UPB_INLINE uint64_t upb_zzenc_64(int64_t n) { return (n << 1) ^ (n >> 63); } + +/* Decoding *******************************************************************/ + +/* All decoding functions return this struct by value. */ +typedef struct { + const char *p; /* NULL if the varint was unterminated. */ + uint64_t val; +} upb_decoderet; + +UPB_INLINE upb_decoderet upb_decoderet_make(const char *p, uint64_t val) { + upb_decoderet ret; + ret.p = p; + ret.val = val; + return ret; +} + +/* Four functions for decoding a varint of at most eight bytes. They are all + * functionally identical, but are implemented in different ways and likely have + * different performance profiles. We keep them around for performance testing. + * + * Note that these functions may not read byte-by-byte, so they must not be used + * unless there are at least eight bytes left in the buffer! */ +upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r); +upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r); +upb_decoderet upb_vdecode_max8_wright(upb_decoderet r); +upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r); + +/* Template for a function that checks the first two bytes with branching + * and dispatches 2-10 bytes with a separate function. Note that this may read + * up to 10 bytes, so it must not be used unless there are at least ten bytes + * left in the buffer! */ +#define UPB_VARINT_DECODER_CHECK2(name, decode_max8_function) \ +UPB_INLINE upb_decoderet upb_vdecode_check2_ ## name(const char *_p) { \ + uint8_t *p = (uint8_t*)_p; \ + upb_decoderet r; \ + if ((*p & 0x80) == 0) { \ + /* Common case: one-byte varint. */ \ + return upb_decoderet_make(_p + 1, *p & 0x7fU); \ + } \ + r = upb_decoderet_make(_p + 2, (*p & 0x7fU) | ((*(p + 1) & 0x7fU) << 7)); \ + if ((*(p + 1) & 0x80) == 0) { \ + /* Two-byte varint. */ \ + return r; \ + } \ + /* Longer varint, fallback to out-of-line function. */ \ + return decode_max8_function(r); \ +} + +UPB_VARINT_DECODER_CHECK2(branch32, upb_vdecode_max8_branch32) +UPB_VARINT_DECODER_CHECK2(branch64, upb_vdecode_max8_branch64) +UPB_VARINT_DECODER_CHECK2(wright, upb_vdecode_max8_wright) +UPB_VARINT_DECODER_CHECK2(massimino, upb_vdecode_max8_massimino) +#undef UPB_VARINT_DECODER_CHECK2 + +/* Our canonical functions for decoding varints, based on the currently + * favored best-performing implementations. */ +UPB_INLINE upb_decoderet upb_vdecode_fast(const char *p) { + if (sizeof(long) == 8) + return upb_vdecode_check2_branch64(p); + else + return upb_vdecode_check2_branch32(p); +} + +UPB_INLINE upb_decoderet upb_vdecode_max8_fast(upb_decoderet r) { + return upb_vdecode_max8_massimino(r); +} + + +/* Encoding *******************************************************************/ + +UPB_INLINE int upb_value_size(uint64_t val) { +#ifdef __GNUC__ + int high_bit = 63 - __builtin_clzll(val); /* 0-based, undef if val == 0. */ +#else + int high_bit = 0; + uint64_t tmp = val; + while(tmp >>= 1) high_bit++; +#endif + return val == 0 ? 1 : high_bit / 8 + 1; +} + +/* Encodes a 64-bit varint into buf (which must be >=UPB_PB_VARINT_MAX_LEN + * bytes long), returning how many bytes were used. + * + * TODO: benchmark and optimize if necessary. */ +UPB_INLINE size_t upb_vencode64(uint64_t val, char *buf) { + size_t i; + if (val == 0) { buf[0] = 0; return 1; } + i = 0; + while (val) { + uint8_t byte = val & 0x7fU; + val >>= 7; + if (val) byte |= 0x80U; + buf[i++] = byte; + } + return i; +} + +UPB_INLINE size_t upb_varint_size(uint64_t val) { + char buf[UPB_PB_VARINT_MAX_LEN]; + return upb_vencode64(val, buf); +} + +/* Encodes a 32-bit varint, *not* sign-extended. */ +UPB_INLINE uint64_t upb_vencode32(uint32_t val) { + char buf[UPB_PB_VARINT_MAX_LEN]; + size_t bytes = upb_vencode64(val, buf); + uint64_t ret = 0; + assert(bytes <= 5); + memcpy(&ret, buf, bytes); + assert(ret <= 0xffffffffffU); + return ret; +} + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_VARINT_DECODER_H_ */ +/* +** upb::pb::Encoder (upb_pb_encoder) +** +** Implements a set of upb_handlers that write protobuf data to the binary wire +** format. +** +** This encoder implementation does not have any access to any out-of-band or +** precomputed lengths for submessages, so it must buffer submessages internally +** before it can emit the first byte. +*/ + +#ifndef UPB_ENCODER_H_ +#define UPB_ENCODER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class Encoder; +} /* namespace pb */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::pb::Encoder, upb_pb_encoder) + +#define UPB_PBENCODER_MAX_NESTING 100 + +/* upb::pb::Encoder ***********************************************************/ + +/* Preallocation hint: decoder won't allocate more bytes than this when first + * constructed. This hint may be an overestimate for some build configurations. + * But if the decoder library is upgraded without recompiling the application, + * it may be an underestimate. */ +#define UPB_PB_ENCODER_SIZE 768 + +#ifdef __cplusplus + +class upb::pb::Encoder { + public: + /* Creates a new encoder in the given environment. The Handlers must have + * come from NewHandlers() below. */ + static Encoder* Create(Environment* env, const Handlers* handlers, + BytesSink* output); + + /* The input to the encoder. */ + Sink* input(); + + /* Creates a new set of handlers for this MessageDef. */ + static reffed_ptr<const Handlers> NewHandlers(const MessageDef* msg); + + static const size_t kSize = UPB_PB_ENCODER_SIZE; + + private: + UPB_DISALLOW_POD_OPS(Encoder, upb::pb::Encoder) +}; + +#endif + +UPB_BEGIN_EXTERN_C + +const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m, + const void *owner); +upb_sink *upb_pb_encoder_input(upb_pb_encoder *p); +upb_pb_encoder* upb_pb_encoder_create(upb_env* e, const upb_handlers* h, + upb_bytessink* output); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace pb { +inline Encoder* Encoder::Create(Environment* env, const Handlers* handlers, + BytesSink* output) { + return upb_pb_encoder_create(env, handlers, output); +} +inline Sink* Encoder::input() { + return upb_pb_encoder_input(this); +} +inline reffed_ptr<const Handlers> Encoder::NewHandlers( + const upb::MessageDef *md) { + const Handlers* h = upb_pb_encoder_newhandlers(md, &h); + return reffed_ptr<const Handlers>(h, &h); +} +} /* namespace pb */ +} /* namespace upb */ + +#endif + +#endif /* UPB_ENCODER_H_ */ +/* +** upb's core components like upb_decoder and upb_msg are carefully designed to +** avoid depending on each other for maximum orthogonality. In other words, +** you can use a upb_decoder to decode into *any* kind of structure; upb_msg is +** just one such structure. A upb_msg can be serialized/deserialized into any +** format, protobuf binary format is just one such format. +** +** However, for convenience we provide functions here for doing common +** operations like deserializing protobuf binary format into a upb_msg. The +** compromise is that this file drags in almost all of upb as a dependency, +** which could be undesirable if you're trying to use a trimmed-down build of +** upb. +** +** While these routines are convenient, they do not reuse any encoding/decoding +** state. For example, if a decoder is JIT-based, it will be re-JITted every +** time these functions are called. For this reason, if you are parsing lots +** of data and efficiency is an issue, these may not be the best functions to +** use (though they are useful for prototyping, before optimizing). +*/ + +#ifndef UPB_GLUE_H +#define UPB_GLUE_H + +#include <stdbool.h> + +#ifdef __cplusplus +#include <vector> + +extern "C" { +#endif + +/* Loads a binary descriptor and returns a NULL-terminated array of unfrozen + * filedefs. The caller owns the returned array, which must be freed with + * upb_gfree(). */ +upb_filedef **upb_loaddescriptor(const char *buf, size_t n, const void *owner, + upb_status *status); + +#ifdef __cplusplus +} /* extern "C" */ + +namespace upb { + +inline bool LoadDescriptor(const char* buf, size_t n, Status* status, + std::vector<reffed_ptr<FileDef> >* files) { + FileDef** parsed_files = upb_loaddescriptor(buf, n, &parsed_files, status); + + if (parsed_files) { + FileDef** p = parsed_files; + while (*p) { + files->push_back(reffed_ptr<FileDef>(*p, &parsed_files)); + ++p; + } + free(parsed_files); + return true; + } else { + return false; + } +} + +/* Templated so it can accept both string and std::string. */ +template <typename T> +bool LoadDescriptor(const T& desc, Status* status, + std::vector<reffed_ptr<FileDef> >* files) { + return LoadDescriptor(desc.c_str(), desc.size(), status, files); +} + +} /* namespace upb */ + +#endif + +#endif /* UPB_GLUE_H */ +/* +** upb::pb::TextPrinter (upb_textprinter) +** +** Handlers for writing to protobuf text format. +*/ + +#ifndef UPB_TEXT_H_ +#define UPB_TEXT_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class TextPrinter; +} /* namespace pb */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::pb::TextPrinter, upb_textprinter) + +#ifdef __cplusplus + +class upb::pb::TextPrinter { + public: + /* The given handlers must have come from NewHandlers(). It must outlive the + * TextPrinter. */ + static TextPrinter *Create(Environment *env, const upb::Handlers *handlers, + BytesSink *output); + + void SetSingleLineMode(bool single_line); + + Sink* input(); + + /* If handler caching becomes a requirement we can add a code cache as in + * decoder.h */ + static reffed_ptr<const Handlers> NewHandlers(const MessageDef* md); +}; + +#endif + +UPB_BEGIN_EXTERN_C + +/* C API. */ +upb_textprinter *upb_textprinter_create(upb_env *env, const upb_handlers *h, + upb_bytessink *output); +void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line); +upb_sink *upb_textprinter_input(upb_textprinter *p); + +const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m, + const void *owner); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace pb { +inline TextPrinter *TextPrinter::Create(Environment *env, + const upb::Handlers *handlers, + BytesSink *output) { + return upb_textprinter_create(env, handlers, output); +} +inline void TextPrinter::SetSingleLineMode(bool single_line) { + upb_textprinter_setsingleline(this, single_line); +} +inline Sink* TextPrinter::input() { + return upb_textprinter_input(this); +} +inline reffed_ptr<const Handlers> TextPrinter::NewHandlers( + const MessageDef *md) { + const Handlers* h = upb_textprinter_newhandlers(md, &h); + return reffed_ptr<const Handlers>(h, &h); +} +} /* namespace pb */ +} /* namespace upb */ + +#endif + +#endif /* UPB_TEXT_H_ */ +/* +** upb::json::Parser (upb_json_parser) +** +** Parses JSON according to a specific schema. +** Support for parsing arbitrary JSON (schema-less) will be added later. +*/ + +#ifndef UPB_JSON_PARSER_H_ +#define UPB_JSON_PARSER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace json { +class Parser; +class ParserMethod; +} /* namespace json */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::json::Parser, upb_json_parser) +UPB_DECLARE_DERIVED_TYPE(upb::json::ParserMethod, upb::RefCounted, + upb_json_parsermethod, upb_refcounted) + +/* upb::json::Parser **********************************************************/ + +/* Preallocation hint: parser won't allocate more bytes than this when first + * constructed. This hint may be an overestimate for some build configurations. + * But if the parser library is upgraded without recompiling the application, + * it may be an underestimate. */ +#define UPB_JSON_PARSER_SIZE 4112 + +#ifdef __cplusplus + +/* Parses an incoming BytesStream, pushing the results to the destination + * sink. */ +class upb::json::Parser { + public: + static Parser* Create(Environment* env, const ParserMethod* method, + Sink* output); + + BytesSink* input(); + + private: + UPB_DISALLOW_POD_OPS(Parser, upb::json::Parser) +}; + +class upb::json::ParserMethod { + public: + /* Include base methods from upb::ReferenceCounted. */ + UPB_REFCOUNTED_CPPMETHODS + + /* Returns handlers for parsing according to the specified schema. */ + static reffed_ptr<const ParserMethod> New(const upb::MessageDef* md); + + /* The destination handlers that are statically bound to this method. + * This method is only capable of outputting to a sink that uses these + * handlers. */ + const Handlers* dest_handlers() const; + + /* The input handlers for this decoder method. */ + const BytesHandler* input_handler() const; + + private: + UPB_DISALLOW_POD_OPS(ParserMethod, upb::json::ParserMethod) +}; + +#endif + +UPB_BEGIN_EXTERN_C + +upb_json_parser* upb_json_parser_create(upb_env* e, + const upb_json_parsermethod* m, + upb_sink* output); +upb_bytessink *upb_json_parser_input(upb_json_parser *p); + +upb_json_parsermethod* upb_json_parsermethod_new(const upb_msgdef* md, + const void* owner); +const upb_handlers *upb_json_parsermethod_desthandlers( + const upb_json_parsermethod *m); +const upb_byteshandler *upb_json_parsermethod_inputhandler( + const upb_json_parsermethod *m); + +/* Include refcounted methods like upb_json_parsermethod_ref(). */ +UPB_REFCOUNTED_CMETHODS(upb_json_parsermethod, upb_json_parsermethod_upcast) + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace json { +inline Parser* Parser::Create(Environment* env, const ParserMethod* method, + Sink* output) { + return upb_json_parser_create(env, method, output); +} +inline BytesSink* Parser::input() { + return upb_json_parser_input(this); +} + +inline const Handlers* ParserMethod::dest_handlers() const { + return upb_json_parsermethod_desthandlers(this); +} +inline const BytesHandler* ParserMethod::input_handler() const { + return upb_json_parsermethod_inputhandler(this); +} +/* static */ +inline reffed_ptr<const ParserMethod> ParserMethod::New( + const MessageDef* md) { + const upb_json_parsermethod *m = upb_json_parsermethod_new(md, &m); + return reffed_ptr<const ParserMethod>(m, &m); +} + +} /* namespace json */ +} /* namespace upb */ + +#endif + + +#endif /* UPB_JSON_PARSER_H_ */ +/* +** upb::json::Printer +** +** Handlers that emit JSON according to a specific protobuf schema. +*/ + +#ifndef UPB_JSON_TYPED_PRINTER_H_ +#define UPB_JSON_TYPED_PRINTER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace json { +class Printer; +} /* namespace json */ +} /* namespace upb */ +#endif + +UPB_DECLARE_TYPE(upb::json::Printer, upb_json_printer) + + +/* upb::json::Printer *********************************************************/ + +#define UPB_JSON_PRINTER_SIZE 176 + +#ifdef __cplusplus + +/* Prints an incoming stream of data to a BytesSink in JSON format. */ +class upb::json::Printer { + public: + static Printer* Create(Environment* env, const upb::Handlers* handlers, + BytesSink* output); + + /* The input to the printer. */ + Sink* input(); + + /* Returns handlers for printing according to the specified schema. + * If preserve_proto_fieldnames is true, the output JSON will use the + * original .proto field names (ie. {"my_field":3}) instead of using + * camelCased names, which is the default: (eg. {"myField":3}). */ + static reffed_ptr<const Handlers> NewHandlers(const upb::MessageDef* md, + bool preserve_proto_fieldnames); + + static const size_t kSize = UPB_JSON_PRINTER_SIZE; + + private: + UPB_DISALLOW_POD_OPS(Printer, upb::json::Printer) +}; + +#endif + +UPB_BEGIN_EXTERN_C + +/* Native C API. */ +upb_json_printer *upb_json_printer_create(upb_env *e, const upb_handlers *h, + upb_bytessink *output); +upb_sink *upb_json_printer_input(upb_json_printer *p); +const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md, + bool preserve_fieldnames, + const void *owner); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace json { +inline Printer* Printer::Create(Environment* env, const upb::Handlers* handlers, + BytesSink* output) { + return upb_json_printer_create(env, handlers, output); +} +inline Sink* Printer::input() { return upb_json_printer_input(this); } +inline reffed_ptr<const Handlers> Printer::NewHandlers( + const upb::MessageDef *md, bool preserve_proto_fieldnames) { + const Handlers* h = upb_json_printer_newhandlers( + md, preserve_proto_fieldnames, &h); + return reffed_ptr<const Handlers>(h, &h); +} +} /* namespace json */ +} /* namespace upb */ + +#endif + +#endif /* UPB_JSON_TYPED_PRINTER_H_ */ |