aboutsummaryrefslogtreecommitdiff
path: root/third_party/abseil-cpp/absl/strings/cord.cc
diff options
context:
space:
mode:
Diffstat (limited to 'third_party/abseil-cpp/absl/strings/cord.cc')
-rw-r--r--third_party/abseil-cpp/absl/strings/cord.cc1310
1 files changed, 669 insertions, 641 deletions
diff --git a/third_party/abseil-cpp/absl/strings/cord.cc b/third_party/abseil-cpp/absl/strings/cord.cc
index d9503ae332..854047ca98 100644
--- a/third_party/abseil-cpp/absl/strings/cord.cc
+++ b/third_party/abseil-cpp/absl/strings/cord.cc
@@ -15,10 +15,12 @@
#include "absl/strings/cord.h"
#include <algorithm>
+#include <atomic>
#include <cstddef>
#include <cstdio>
#include <cstdlib>
#include <iomanip>
+#include <iostream>
#include <limits>
#include <ostream>
#include <sstream>
@@ -28,11 +30,17 @@
#include "absl/base/casts.h"
#include "absl/base/internal/raw_logging.h"
+#include "absl/base/macros.h"
#include "absl/base/port.h"
#include "absl/container/fixed_array.h"
#include "absl/container/inlined_vector.h"
#include "absl/strings/escaping.h"
#include "absl/strings/internal/cord_internal.h"
+#include "absl/strings/internal/cord_rep_btree.h"
+#include "absl/strings/internal/cord_rep_flat.h"
+#include "absl/strings/internal/cordz_statistics.h"
+#include "absl/strings/internal/cordz_update_scope.h"
+#include "absl/strings/internal/cordz_update_tracker.h"
#include "absl/strings/internal/resize_uninitialized.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_format.h"
@@ -43,144 +51,18 @@ namespace absl {
ABSL_NAMESPACE_BEGIN
using ::absl::cord_internal::CordRep;
+using ::absl::cord_internal::CordRepBtree;
using ::absl::cord_internal::CordRepConcat;
using ::absl::cord_internal::CordRepExternal;
+using ::absl::cord_internal::CordRepFlat;
using ::absl::cord_internal::CordRepSubstring;
+using ::absl::cord_internal::CordzUpdateTracker;
+using ::absl::cord_internal::InlineData;
+using ::absl::cord_internal::kMaxFlatLength;
+using ::absl::cord_internal::kMinFlatLength;
-// Various representations that we allow
-enum CordRepKind {
- CONCAT = 0,
- EXTERNAL = 1,
- SUBSTRING = 2,
-
- // We have different tags for different sized flat arrays,
- // starting with FLAT
- FLAT = 3,
-};
-
-namespace {
-
-// Type used with std::allocator for allocating and deallocating
-// `CordRepExternal`. std::allocator is used because it opaquely handles the
-// different new / delete overloads available on a given platform.
-struct alignas(absl::cord_internal::ExternalRepAlignment()) ExternalAllocType {
- unsigned char value[absl::cord_internal::ExternalRepAlignment()];
-};
-
-// Returns the number of objects to pass in to std::allocator<ExternalAllocType>
-// allocate() and deallocate() to create enough room for `CordRepExternal` with
-// `releaser_size` bytes on the end.
-constexpr size_t GetExternalAllocNumObjects(size_t releaser_size) {
- // Be sure to round up since `releaser_size` could be smaller than
- // `sizeof(ExternalAllocType)`.
- return (sizeof(CordRepExternal) + releaser_size + sizeof(ExternalAllocType) -
- 1) /
- sizeof(ExternalAllocType);
-}
-
-// Allocates enough memory for `CordRepExternal` and a releaser with size
-// `releaser_size` bytes.
-void* AllocateExternal(size_t releaser_size) {
- return std::allocator<ExternalAllocType>().allocate(
- GetExternalAllocNumObjects(releaser_size));
-}
-
-// Deallocates the memory for a `CordRepExternal` assuming it was allocated with
-// a releaser of given size and alignment.
-void DeallocateExternal(CordRepExternal* p, size_t releaser_size) {
- std::allocator<ExternalAllocType>().deallocate(
- reinterpret_cast<ExternalAllocType*>(p),
- GetExternalAllocNumObjects(releaser_size));
-}
-
-// Returns a pointer to the type erased releaser for the given CordRepExternal.
-void* GetExternalReleaser(CordRepExternal* rep) {
- return rep + 1;
-}
-
-} // namespace
-
-namespace cord_internal {
-
-inline CordRepConcat* CordRep::concat() {
- assert(tag == CONCAT);
- return static_cast<CordRepConcat*>(this);
-}
-
-inline const CordRepConcat* CordRep::concat() const {
- assert(tag == CONCAT);
- return static_cast<const CordRepConcat*>(this);
-}
-
-inline CordRepSubstring* CordRep::substring() {
- assert(tag == SUBSTRING);
- return static_cast<CordRepSubstring*>(this);
-}
-
-inline const CordRepSubstring* CordRep::substring() const {
- assert(tag == SUBSTRING);
- return static_cast<const CordRepSubstring*>(this);
-}
-
-inline CordRepExternal* CordRep::external() {
- assert(tag == EXTERNAL);
- return static_cast<CordRepExternal*>(this);
-}
-
-inline const CordRepExternal* CordRep::external() const {
- assert(tag == EXTERNAL);
- return static_cast<const CordRepExternal*>(this);
-}
-
-} // namespace cord_internal
-
-static const size_t kFlatOverhead = offsetof(CordRep, data);
-
-static_assert(kFlatOverhead == 13, "Unittests assume kFlatOverhead == 13");
-
-// Largest and smallest flat node lengths we are willing to allocate
-// Flat allocation size is stored in tag, which currently can encode sizes up
-// to 4K, encoded as multiple of either 8 or 32 bytes.
-// If we allow for larger sizes, we need to change this to 8/64, 16/128, etc.
-static constexpr size_t kMaxFlatSize = 4096;
-static constexpr size_t kMaxFlatLength = kMaxFlatSize - kFlatOverhead;
-static constexpr size_t kMinFlatLength = 32 - kFlatOverhead;
-
-// Prefer copying blocks of at most this size, otherwise reference count.
-static const size_t kMaxBytesToCopy = 511;
-
-// Helper functions for rounded div, and rounding to exact sizes.
-static size_t DivUp(size_t n, size_t m) { return (n + m - 1) / m; }
-static size_t RoundUp(size_t n, size_t m) { return DivUp(n, m) * m; }
-
-// Returns the size to the nearest equal or larger value that can be
-// expressed exactly as a tag value.
-static size_t RoundUpForTag(size_t size) {
- return RoundUp(size, (size <= 1024) ? 8 : 32);
-}
-
-// Converts the allocated size to a tag, rounding down if the size
-// does not exactly match a 'tag expressible' size value. The result is
-// undefined if the size exceeds the maximum size that can be encoded in
-// a tag, i.e., if size is larger than TagToAllocatedSize(<max tag>).
-static uint8_t AllocatedSizeToTag(size_t size) {
- const size_t tag = (size <= 1024) ? size / 8 : 128 + size / 32 - 1024 / 32;
- assert(tag <= std::numeric_limits<uint8_t>::max());
- return tag;
-}
-
-// Converts the provided tag to the corresponding allocated size
-static constexpr size_t TagToAllocatedSize(uint8_t tag) {
- return (tag <= 128) ? (tag * 8) : (1024 + (tag - 128) * 32);
-}
-
-// Converts the provided tag to the corresponding available data length
-static constexpr size_t TagToLength(uint8_t tag) {
- return TagToAllocatedSize(tag) - kFlatOverhead;
-}
-
-// Enforce that kMaxFlatSize maps to a well-known exact tag value.
-static_assert(TagToAllocatedSize(224) == kMaxFlatSize, "Bad tag logic");
+using ::absl::cord_internal::kInlinedVectorSize;
+using ::absl::cord_internal::kMaxBytesToCopy;
constexpr uint64_t Fibonacci(unsigned char n, uint64_t a = 0, uint64_t b = 1) {
return n == 0 ? a : Fibonacci(n - 1, b, a + b);
@@ -195,70 +77,30 @@ static_assert(Fibonacci(63) == 6557470319842,
// The root node depth is allowed to become twice as large to reduce rebalancing
// for larger strings (see IsRootBalanced).
static constexpr uint64_t min_length[] = {
- Fibonacci(2),
- Fibonacci(3),
- Fibonacci(4),
- Fibonacci(5),
- Fibonacci(6),
- Fibonacci(7),
- Fibonacci(8),
- Fibonacci(9),
- Fibonacci(10),
- Fibonacci(11),
- Fibonacci(12),
- Fibonacci(13),
- Fibonacci(14),
- Fibonacci(15),
- Fibonacci(16),
- Fibonacci(17),
- Fibonacci(18),
- Fibonacci(19),
- Fibonacci(20),
- Fibonacci(21),
- Fibonacci(22),
- Fibonacci(23),
- Fibonacci(24),
- Fibonacci(25),
- Fibonacci(26),
- Fibonacci(27),
- Fibonacci(28),
- Fibonacci(29),
- Fibonacci(30),
- Fibonacci(31),
- Fibonacci(32),
- Fibonacci(33),
- Fibonacci(34),
- Fibonacci(35),
- Fibonacci(36),
- Fibonacci(37),
- Fibonacci(38),
- Fibonacci(39),
- Fibonacci(40),
- Fibonacci(41),
- Fibonacci(42),
- Fibonacci(43),
- Fibonacci(44),
- Fibonacci(45),
- Fibonacci(46),
- Fibonacci(47),
+ Fibonacci(2), Fibonacci(3), Fibonacci(4), Fibonacci(5),
+ Fibonacci(6), Fibonacci(7), Fibonacci(8), Fibonacci(9),
+ Fibonacci(10), Fibonacci(11), Fibonacci(12), Fibonacci(13),
+ Fibonacci(14), Fibonacci(15), Fibonacci(16), Fibonacci(17),
+ Fibonacci(18), Fibonacci(19), Fibonacci(20), Fibonacci(21),
+ Fibonacci(22), Fibonacci(23), Fibonacci(24), Fibonacci(25),
+ Fibonacci(26), Fibonacci(27), Fibonacci(28), Fibonacci(29),
+ Fibonacci(30), Fibonacci(31), Fibonacci(32), Fibonacci(33),
+ Fibonacci(34), Fibonacci(35), Fibonacci(36), Fibonacci(37),
+ Fibonacci(38), Fibonacci(39), Fibonacci(40), Fibonacci(41),
+ Fibonacci(42), Fibonacci(43), Fibonacci(44), Fibonacci(45),
+ Fibonacci(46), Fibonacci(47),
0xffffffffffffffffull, // Avoid overflow
};
static const int kMinLengthSize = ABSL_ARRAYSIZE(min_length);
-// The inlined size to use with absl::InlinedVector.
-//
-// Note: The InlinedVectors in this file (and in cord.h) do not need to use
-// the same value for their inlined size. The fact that they do is historical.
-// It may be desirable for each to use a different inlined size optimized for
-// that InlinedVector's usage.
-//
-// TODO(jgm): Benchmark to see if there's a more optimal value than 47 for
-// the inlined vector size (47 exists for backward compatibility).
-static const int kInlinedVectorSize = 47;
+static inline bool btree_enabled() {
+ return cord_internal::cord_btree_enabled.load(
+ std::memory_order_relaxed);
+}
static inline bool IsRootBalanced(CordRep* node) {
- if (node->tag != CONCAT) {
+ if (!node->IsConcat()) {
return true;
} else if (node->concat()->depth() <= 15) {
return true;
@@ -272,7 +114,8 @@ static inline bool IsRootBalanced(CordRep* node) {
}
static CordRep* Rebalance(CordRep* node);
-static void DumpNode(CordRep* rep, bool include_data, std::ostream* os);
+static void DumpNode(CordRep* rep, bool include_data, std::ostream* os,
+ int indent = 0);
static bool VerifyNode(CordRep* root, CordRep* start_node,
bool full_validation);
@@ -292,103 +135,9 @@ static inline CordRep* VerifyTree(CordRep* node) {
return node;
}
-// --------------------------------------------------------------------
-// Memory management
-
-inline CordRep* Ref(CordRep* rep) {
- if (rep != nullptr) {
- rep->refcount.Increment();
- }
- return rep;
-}
-
-// This internal routine is called from the cold path of Unref below. Keeping it
-// in a separate routine allows good inlining of Unref into many profitable call
-// sites. However, the call to this function can be highly disruptive to the
-// register pressure in those callers. To minimize the cost to callers, we use
-// a special LLVM calling convention that preserves most registers. This allows
-// the call to this routine in cold paths to not disrupt the caller's register
-// pressure. This calling convention is not available on all platforms; we
-// intentionally allow LLVM to ignore the attribute rather than attempting to
-// hardcode the list of supported platforms.
-#if defined(__clang__) && !defined(__i386__)
-#pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wattributes"
-__attribute__((preserve_most))
-#pragma clang diagnostic pop
-#endif
-static void UnrefInternal(CordRep* rep) {
- assert(rep != nullptr);
-
- absl::InlinedVector<CordRep*, kInlinedVectorSize> pending;
- while (true) {
- if (rep->tag == CONCAT) {
- CordRepConcat* rep_concat = rep->concat();
- CordRep* right = rep_concat->right;
- if (!right->refcount.Decrement()) {
- pending.push_back(right);
- }
- CordRep* left = rep_concat->left;
- delete rep_concat;
- rep = nullptr;
- if (!left->refcount.Decrement()) {
- rep = left;
- continue;
- }
- } else if (rep->tag == EXTERNAL) {
- CordRepExternal* rep_external = rep->external();
- absl::string_view data(rep_external->base, rep->length);
- void* releaser = GetExternalReleaser(rep_external);
- size_t releaser_size = rep_external->releaser_invoker(releaser, data);
- rep_external->~CordRepExternal();
- DeallocateExternal(rep_external, releaser_size);
- rep = nullptr;
- } else if (rep->tag == SUBSTRING) {
- CordRepSubstring* rep_substring = rep->substring();
- CordRep* child = rep_substring->child;
- delete rep_substring;
- rep = nullptr;
- if (!child->refcount.Decrement()) {
- rep = child;
- continue;
- }
- } else {
- // Flat CordReps are allocated and constructed with raw ::operator new
- // and placement new, and must be destructed and deallocated
- // accordingly.
-#if defined(__cpp_sized_deallocation)
- size_t size = TagToAllocatedSize(rep->tag);
- rep->~CordRep();
- ::operator delete(rep, size);
-#else
- rep->~CordRep();
- ::operator delete(rep);
-#endif
- rep = nullptr;
- }
-
- if (!pending.empty()) {
- rep = pending.back();
- pending.pop_back();
- } else {
- break;
- }
- }
-}
-
-inline void Unref(CordRep* rep) {
- // Fast-path for two common, hot cases: a null rep and a shared root.
- if (ABSL_PREDICT_TRUE(rep == nullptr ||
- rep->refcount.DecrementExpectHighRefcount())) {
- return;
- }
-
- UnrefInternal(rep);
-}
-
// Return the depth of a node
static int Depth(const CordRep* rep) {
- if (rep->tag == CONCAT) {
+ if (rep->IsConcat()) {
return rep->concat()->depth();
} else {
return 0;
@@ -409,17 +158,19 @@ static void SetConcatChildren(CordRepConcat* concat, CordRep* left,
// The returned node has a refcount of 1.
static CordRep* RawConcat(CordRep* left, CordRep* right) {
// Avoid making degenerate concat nodes (one child is empty)
- if (left == nullptr || left->length == 0) {
- Unref(left);
+ if (left == nullptr) return right;
+ if (right == nullptr) return left;
+ if (left->length == 0) {
+ CordRep::Unref(left);
return right;
}
- if (right == nullptr || right->length == 0) {
- Unref(right);
+ if (right->length == 0) {
+ CordRep::Unref(right);
return left;
}
CordRepConcat* rep = new CordRepConcat();
- rep->tag = CONCAT;
+ rep->tag = cord_internal::CONCAT;
SetConcatChildren(rep, left, right);
return rep;
@@ -453,35 +204,41 @@ static CordRep* MakeBalancedTree(CordRep** reps, size_t n) {
return reps[0];
}
-// Create a new flat node.
-static CordRep* NewFlat(size_t length_hint) {
- if (length_hint <= kMinFlatLength) {
- length_hint = kMinFlatLength;
- } else if (length_hint > kMaxFlatLength) {
- length_hint = kMaxFlatLength;
- }
+static CordRepFlat* CreateFlat(const char* data, size_t length,
+ size_t alloc_hint) {
+ CordRepFlat* flat = CordRepFlat::New(length + alloc_hint);
+ flat->length = length;
+ memcpy(flat->Data(), data, length);
+ return flat;
+}
- // Round size up so it matches a size we can exactly express in a tag.
- const size_t size = RoundUpForTag(length_hint + kFlatOverhead);
- void* const raw_rep = ::operator new(size);
- CordRep* rep = new (raw_rep) CordRep();
- rep->tag = AllocatedSizeToTag(size);
- return VerifyTree(rep);
+// Creates a new flat or Btree out of the specified array.
+// The returned node has a refcount of 1.
+static CordRep* NewBtree(const char* data, size_t length, size_t alloc_hint) {
+ if (length <= kMaxFlatLength) {
+ return CreateFlat(data, length, alloc_hint);
+ }
+ CordRepFlat* flat = CreateFlat(data, kMaxFlatLength, 0);
+ data += kMaxFlatLength;
+ length -= kMaxFlatLength;
+ auto* root = CordRepBtree::Create(flat);
+ return CordRepBtree::Append(root, {data, length}, alloc_hint);
}
// Create a new tree out of the specified array.
// The returned node has a refcount of 1.
-static CordRep* NewTree(const char* data,
- size_t length,
- size_t alloc_hint) {
+static CordRep* NewTree(const char* data, size_t length, size_t alloc_hint) {
if (length == 0) return nullptr;
+ if (btree_enabled()) {
+ return NewBtree(data, length, alloc_hint);
+ }
absl::FixedArray<CordRep*> reps((length - 1) / kMaxFlatLength + 1);
size_t n = 0;
do {
const size_t len = std::min(length, kMaxFlatLength);
- CordRep* rep = NewFlat(len + alloc_hint);
+ CordRepFlat* rep = CordRepFlat::New(len + alloc_hint);
rep->length = len;
- memcpy(rep->data, data, len);
+ memcpy(rep->Data(), data, len);
reps[n++] = VerifyTree(rep);
data += len;
length -= len;
@@ -491,18 +248,12 @@ static CordRep* NewTree(const char* data,
namespace cord_internal {
-ExternalRepReleaserPair NewExternalWithUninitializedReleaser(
- absl::string_view data, ExternalReleaserInvoker invoker,
- size_t releaser_size) {
+void InitializeCordRepExternal(absl::string_view data, CordRepExternal* rep) {
assert(!data.empty());
-
- void* raw_rep = AllocateExternal(releaser_size);
- auto* rep = new (raw_rep) CordRepExternal();
rep->length = data.size();
rep->tag = EXTERNAL;
rep->base = data.data();
- rep->releaser_invoker = invoker;
- return {VerifyTree(rep), GetExternalReleaser(rep)};
+ VerifyTree(rep);
}
} // namespace cord_internal
@@ -510,87 +261,155 @@ ExternalRepReleaserPair NewExternalWithUninitializedReleaser(
static CordRep* NewSubstring(CordRep* child, size_t offset, size_t length) {
// Never create empty substring nodes
if (length == 0) {
- Unref(child);
+ CordRep::Unref(child);
return nullptr;
} else {
CordRepSubstring* rep = new CordRepSubstring();
assert((offset + length) <= child->length);
rep->length = length;
- rep->tag = SUBSTRING;
+ rep->tag = cord_internal::SUBSTRING;
rep->start = offset;
rep->child = child;
return VerifyTree(rep);
}
}
+// Creates a CordRep from the provided string. If the string is large enough,
+// and not wasteful, we move the string into an external cord rep, preserving
+// the already allocated string contents.
+// Requires the provided string length to be larger than `kMaxInline`.
+static CordRep* CordRepFromString(std::string&& src) {
+ assert(src.length() > cord_internal::kMaxInline);
+ if (
+ // String is short: copy data to avoid external block overhead.
+ src.size() <= kMaxBytesToCopy ||
+ // String is wasteful: copy data to avoid pinning too much unused memory.
+ src.size() < src.capacity() / 2
+ ) {
+ return NewTree(src.data(), src.size(), 0);
+ }
+
+ struct StringReleaser {
+ void operator()(absl::string_view /* data */) {}
+ std::string data;
+ };
+ const absl::string_view original_data = src;
+ auto* rep =
+ static_cast<::absl::cord_internal::CordRepExternalImpl<StringReleaser>*>(
+ absl::cord_internal::NewExternalRep(original_data,
+ StringReleaser{std::move(src)}));
+ // Moving src may have invalidated its data pointer, so adjust it.
+ rep->base = rep->template get<0>().data.data();
+ return rep;
+}
+
// --------------------------------------------------------------------
// Cord::InlineRep functions
-// This will trigger LNK2005 in MSVC.
-#ifndef COMPILER_MSVC
-const unsigned char Cord::InlineRep::kMaxInline;
-#endif // COMPILER_MSVC
+constexpr unsigned char Cord::InlineRep::kMaxInline;
inline void Cord::InlineRep::set_data(const char* data, size_t n,
bool nullify_tail) {
static_assert(kMaxInline == 15, "set_data is hard-coded for a length of 15");
- cord_internal::SmallMemmove(data_, data, n, nullify_tail);
- data_[kMaxInline] = static_cast<char>(n);
+ cord_internal::SmallMemmove(data_.as_chars(), data, n, nullify_tail);
+ set_inline_size(n);
}
inline char* Cord::InlineRep::set_data(size_t n) {
assert(n <= kMaxInline);
- memset(data_, 0, sizeof(data_));
- data_[kMaxInline] = static_cast<char>(n);
- return data_;
-}
-
-inline CordRep* Cord::InlineRep::force_tree(size_t extra_hint) {
- size_t len = data_[kMaxInline];
- CordRep* result;
- if (len > kMaxInline) {
- memcpy(&result, data_, sizeof(result));
- } else {
- result = NewFlat(len + extra_hint);
- result->length = len;
- memcpy(result->data, data_, len);
- set_tree(result);
- }
- return result;
+ ResetToEmpty();
+ set_inline_size(n);
+ return data_.as_chars();
}
inline void Cord::InlineRep::reduce_size(size_t n) {
- size_t tag = data_[kMaxInline];
+ size_t tag = inline_size();
assert(tag <= kMaxInline);
assert(tag >= n);
tag -= n;
- memset(data_ + tag, 0, n);
- data_[kMaxInline] = static_cast<char>(tag);
+ memset(data_.as_chars() + tag, 0, n);
+ set_inline_size(static_cast<char>(tag));
}
inline void Cord::InlineRep::remove_prefix(size_t n) {
- cord_internal::SmallMemmove(data_, data_ + n, data_[kMaxInline] - n);
+ cord_internal::SmallMemmove(data_.as_chars(), data_.as_chars() + n,
+ inline_size() - n);
reduce_size(n);
}
-void Cord::InlineRep::AppendTree(CordRep* tree) {
- if (tree == nullptr) return;
- size_t len = data_[kMaxInline];
- if (len == 0) {
- set_tree(tree);
+// Returns `rep` converted into a CordRepBtree.
+// Directly returns `rep` if `rep` is already a CordRepBtree.
+static CordRepBtree* ForceBtree(CordRep* rep) {
+ return rep->IsBtree() ? rep->btree() : CordRepBtree::Create(rep);
+}
+
+void Cord::InlineRep::AppendTreeToInlined(CordRep* tree,
+ MethodIdentifier method) {
+ assert(!is_tree());
+ if (!data_.is_empty()) {
+ CordRepFlat* flat = MakeFlatWithExtraCapacity(0);
+ if (btree_enabled()) {
+ tree = CordRepBtree::Append(CordRepBtree::Create(flat), tree);
+ } else {
+ tree = Concat(flat, tree);
+ }
+ }
+ EmplaceTree(tree, method);
+}
+
+void Cord::InlineRep::AppendTreeToTree(CordRep* tree, MethodIdentifier method) {
+ assert(is_tree());
+ const CordzUpdateScope scope(data_.cordz_info(), method);
+ if (btree_enabled()) {
+ tree = CordRepBtree::Append(ForceBtree(data_.as_tree()), tree);
} else {
- set_tree(Concat(force_tree(0), tree));
+ tree = Concat(data_.as_tree(), tree);
}
+ SetTree(tree, scope);
}
-void Cord::InlineRep::PrependTree(CordRep* tree) {
+void Cord::InlineRep::AppendTree(CordRep* tree, MethodIdentifier method) {
if (tree == nullptr) return;
- size_t len = data_[kMaxInline];
- if (len == 0) {
- set_tree(tree);
+ if (data_.is_tree()) {
+ AppendTreeToTree(tree, method);
} else {
- set_tree(Concat(tree, force_tree(0)));
+ AppendTreeToInlined(tree, method);
+ }
+}
+
+void Cord::InlineRep::PrependTreeToInlined(CordRep* tree,
+ MethodIdentifier method) {
+ assert(!is_tree());
+ if (!data_.is_empty()) {
+ CordRepFlat* flat = MakeFlatWithExtraCapacity(0);
+ if (btree_enabled()) {
+ tree = CordRepBtree::Prepend(CordRepBtree::Create(flat), tree);
+ } else {
+ tree = Concat(tree, flat);
+ }
+ }
+ EmplaceTree(tree, method);
+}
+
+void Cord::InlineRep::PrependTreeToTree(CordRep* tree,
+ MethodIdentifier method) {
+ assert(is_tree());
+ const CordzUpdateScope scope(data_.cordz_info(), method);
+ if (btree_enabled()) {
+ tree = CordRepBtree::Prepend(ForceBtree(data_.as_tree()), tree);
+ } else {
+ tree = Concat(tree, data_.as_tree());
+ }
+ SetTree(tree, scope);
+}
+
+void Cord::InlineRep::PrependTree(CordRep* tree, MethodIdentifier method) {
+ assert(tree != nullptr);
+ if (data_.is_tree()) {
+ PrependTreeToTree(tree, method);
+ } else {
+ PrependTreeToInlined(tree, method);
}
}
@@ -600,20 +419,29 @@ void Cord::InlineRep::PrependTree(CordRep* tree) {
// written to region and the actual size increase will be written to size.
static inline bool PrepareAppendRegion(CordRep* root, char** region,
size_t* size, size_t max_length) {
+ if (root->IsBtree() && root->refcount.IsMutable()) {
+ Span<char> span = root->btree()->GetAppendBuffer(max_length);
+ if (!span.empty()) {
+ *region = span.data();
+ *size = span.size();
+ return true;
+ }
+ }
+
// Search down the right-hand path for a non-full FLAT node.
CordRep* dst = root;
- while (dst->tag == CONCAT && dst->refcount.IsOne()) {
+ while (dst->IsConcat() && dst->refcount.IsMutable()) {
dst = dst->concat()->right;
}
- if (dst->tag < FLAT || !dst->refcount.IsOne()) {
+ if (!dst->IsFlat() || !dst->refcount.IsMutable()) {
*region = nullptr;
*size = 0;
return false;
}
const size_t in_use = dst->length;
- const size_t capacity = TagToLength(dst->tag);
+ const size_t capacity = dst->flat()->Capacity();
if (in_use == capacity) {
*region = nullptr;
*size = 0;
@@ -628,214 +456,281 @@ static inline bool PrepareAppendRegion(CordRep* root, char** region,
}
dst->length += size_increase;
- *region = dst->data + in_use;
+ *region = dst->flat()->Data() + in_use;
*size = size_increase;
return true;
}
+template <bool has_length>
void Cord::InlineRep::GetAppendRegion(char** region, size_t* size,
- size_t max_length) {
- if (max_length == 0) {
- *region = nullptr;
- *size = 0;
- return;
- }
+ size_t length) {
+ auto constexpr method = CordzUpdateTracker::kGetAppendRegion;
- // Try to fit in the inline buffer if possible.
- size_t inline_length = data_[kMaxInline];
- if (inline_length < kMaxInline && max_length <= kMaxInline - inline_length) {
- *region = data_ + inline_length;
- *size = max_length;
- data_[kMaxInline] = static_cast<char>(inline_length + max_length);
- return;
+ CordRep* root = tree();
+ size_t sz = root ? root->length : inline_size();
+ if (root == nullptr) {
+ size_t available = kMaxInline - sz;
+ if (available >= (has_length ? length : 1)) {
+ *region = data_.as_chars() + sz;
+ *size = has_length ? length : available;
+ set_inline_size(has_length ? sz + length : kMaxInline);
+ return;
+ }
}
- CordRep* root = force_tree(max_length);
-
- if (PrepareAppendRegion(root, region, size, max_length)) {
+ size_t extra = has_length ? length : (std::max)(sz, kMinFlatLength);
+ CordRep* rep = root ? root : MakeFlatWithExtraCapacity(extra);
+ CordzUpdateScope scope(root ? data_.cordz_info() : nullptr, method);
+ if (PrepareAppendRegion(rep, region, size, length)) {
+ CommitTree(root, rep, scope, method);
return;
}
// Allocate new node.
- CordRep* new_node =
- NewFlat(std::max(static_cast<size_t>(root->length), max_length));
- new_node->length =
- std::min(static_cast<size_t>(TagToLength(new_node->tag)), max_length);
- *region = new_node->data;
+ CordRepFlat* new_node = CordRepFlat::New(extra);
+ new_node->length = std::min(new_node->Capacity(), length);
+ *region = new_node->Data();
*size = new_node->length;
- replace_tree(Concat(root, new_node));
-}
-
-void Cord::InlineRep::GetAppendRegion(char** region, size_t* size) {
- const size_t max_length = std::numeric_limits<size_t>::max();
- // Try to fit in the inline buffer if possible.
- size_t inline_length = data_[kMaxInline];
- if (inline_length < kMaxInline) {
- *region = data_ + inline_length;
- *size = kMaxInline - inline_length;
- data_[kMaxInline] = kMaxInline;
- return;
+ if (btree_enabled()) {
+ rep = CordRepBtree::Append(ForceBtree(rep), new_node);
+ } else {
+ rep = Concat(rep, new_node);
}
+ CommitTree(root, rep, scope, method);
+}
- CordRep* root = force_tree(max_length);
-
- if (PrepareAppendRegion(root, region, size, max_length)) {
- return;
+// Computes the memory side of the provided edge which must be a valid data edge
+// for a btrtee, i.e., a FLAT, EXTERNAL or SUBSTRING of a FLAT or EXTERNAL node.
+static bool RepMemoryUsageDataEdge(const CordRep* rep,
+ size_t* total_mem_usage) {
+ size_t maybe_sub_size = 0;
+ if (ABSL_PREDICT_FALSE(rep->IsSubstring())) {
+ maybe_sub_size = sizeof(cord_internal::CordRepSubstring);
+ rep = rep->substring()->child;
}
-
- // Allocate new node.
- CordRep* new_node = NewFlat(root->length);
- new_node->length = TagToLength(new_node->tag);
- *region = new_node->data;
- *size = new_node->length;
- replace_tree(Concat(root, new_node));
+ if (rep->IsFlat()) {
+ *total_mem_usage += maybe_sub_size + rep->flat()->AllocatedSize();
+ return true;
+ }
+ if (rep->IsExternal()) {
+ // We don't know anything about the embedded / bound data, but we can safely
+ // assume it is 'at least' a word / pointer to data. In the future we may
+ // choose to use the 'data' byte as a tag to identify the types of some
+ // well-known externals, such as a std::string instance.
+ *total_mem_usage += maybe_sub_size +
+ sizeof(cord_internal::CordRepExternalImpl<intptr_t>) +
+ rep->length;
+ return true;
+ }
+ return false;
}
// If the rep is a leaf, this will increment the value at total_mem_usage and
// will return true.
static bool RepMemoryUsageLeaf(const CordRep* rep, size_t* total_mem_usage) {
- if (rep->tag >= FLAT) {
- *total_mem_usage += TagToAllocatedSize(rep->tag);
+ if (rep->IsFlat()) {
+ *total_mem_usage += rep->flat()->AllocatedSize();
return true;
}
- if (rep->tag == EXTERNAL) {
- *total_mem_usage += sizeof(CordRepConcat) + rep->length;
+ if (rep->IsExternal()) {
+ // We don't know anything about the embedded / bound data, but we can safely
+ // assume it is 'at least' a word / pointer to data. In the future we may
+ // choose to use the 'data' byte as a tag to identify the types of some
+ // well-known externals, such as a std::string instance.
+ *total_mem_usage +=
+ sizeof(cord_internal::CordRepExternalImpl<intptr_t>) + rep->length;
return true;
}
return false;
}
void Cord::InlineRep::AssignSlow(const Cord::InlineRep& src) {
- ClearSlow();
+ assert(&src != this);
+ assert(is_tree() || src.is_tree());
+ auto constexpr method = CordzUpdateTracker::kAssignCord;
+ if (ABSL_PREDICT_TRUE(!is_tree())) {
+ EmplaceTree(CordRep::Ref(src.as_tree()), src.data_, method);
+ return;
+ }
- memcpy(data_, src.data_, sizeof(data_));
- if (is_tree()) {
- Ref(tree());
+ CordRep* tree = as_tree();
+ if (CordRep* src_tree = src.tree()) {
+ // Leave any existing `cordz_info` in place, and let MaybeTrackCord()
+ // decide if this cord should be (or remains to be) sampled or not.
+ data_.set_tree(CordRep::Ref(src_tree));
+ CordzInfo::MaybeTrackCord(data_, src.data_, method);
+ } else {
+ CordzInfo::MaybeUntrackCord(data_.cordz_info());
+ data_ = src.data_;
}
+ CordRep::Unref(tree);
}
-void Cord::InlineRep::ClearSlow() {
+void Cord::InlineRep::UnrefTree() {
if (is_tree()) {
- Unref(tree());
+ CordzInfo::MaybeUntrackCord(data_.cordz_info());
+ CordRep::Unref(tree());
}
- memset(data_, 0, sizeof(data_));
}
// --------------------------------------------------------------------
// Constructors and destructors
-Cord::Cord(const Cord& src) : contents_(src.contents_) {
- Ref(contents_.tree()); // Does nothing if contents_ has embedded data
-}
-
-Cord::Cord(absl::string_view src) {
+Cord::Cord(absl::string_view src, MethodIdentifier method)
+ : contents_(InlineData::kDefaultInit) {
const size_t n = src.size();
if (n <= InlineRep::kMaxInline) {
- contents_.set_data(src.data(), n, false);
+ contents_.set_data(src.data(), n, true);
+ } else {
+ CordRep* rep = NewTree(src.data(), n, 0);
+ contents_.EmplaceTree(rep, method);
+ }
+}
+
+template <typename T, Cord::EnableIfString<T>>
+Cord::Cord(T&& src) : contents_(InlineData::kDefaultInit) {
+ if (src.size() <= InlineRep::kMaxInline) {
+ contents_.set_data(src.data(), src.size(), true);
} else {
- contents_.set_tree(NewTree(src.data(), n, 0));
+ CordRep* rep = CordRepFromString(std::forward<T>(src));
+ contents_.EmplaceTree(rep, CordzUpdateTracker::kConstructorString);
}
}
+template Cord::Cord(std::string&& src);
+
// The destruction code is separate so that the compiler can determine
// that it does not need to call the destructor on a moved-from Cord.
void Cord::DestroyCordSlow() {
- Unref(VerifyTree(contents_.tree()));
+ assert(contents_.is_tree());
+ CordzInfo::MaybeUntrackCord(contents_.cordz_info());
+ CordRep::Unref(VerifyTree(contents_.as_tree()));
}
// --------------------------------------------------------------------
// Mutators
void Cord::Clear() {
- Unref(contents_.clear());
+ if (CordRep* tree = contents_.clear()) {
+ CordRep::Unref(tree);
+ }
}
-Cord& Cord::operator=(absl::string_view src) {
+Cord& Cord::AssignLargeString(std::string&& src) {
+ auto constexpr method = CordzUpdateTracker::kAssignString;
+ assert(src.size() > kMaxBytesToCopy);
+ CordRep* rep = CordRepFromString(std::move(src));
+ if (CordRep* tree = contents_.tree()) {
+ CordzUpdateScope scope(contents_.cordz_info(), method);
+ contents_.SetTree(rep, scope);
+ CordRep::Unref(tree);
+ } else {
+ contents_.EmplaceTree(rep, method);
+ }
+ return *this;
+}
+Cord& Cord::operator=(absl::string_view src) {
+ auto constexpr method = CordzUpdateTracker::kAssignString;
const char* data = src.data();
size_t length = src.size();
CordRep* tree = contents_.tree();
if (length <= InlineRep::kMaxInline) {
- // Embed into this->contents_
+ // Embed into this->contents_, which is somewhat subtle:
+ // - MaybeUntrackCord must be called before Unref(tree).
+ // - MaybeUntrackCord must be called before set_data() clobbers cordz_info.
+ // - set_data() must be called before Unref(tree) as it may reference tree.
+ if (tree != nullptr) CordzInfo::MaybeUntrackCord(contents_.cordz_info());
contents_.set_data(data, length, true);
- Unref(tree);
+ if (tree != nullptr) CordRep::Unref(tree);
return *this;
}
- if (tree != nullptr && tree->tag >= FLAT &&
- TagToLength(tree->tag) >= length && tree->refcount.IsOne()) {
- // Copy in place if the existing FLAT node is reusable.
- memmove(tree->data, data, length);
- tree->length = length;
- VerifyTree(tree);
- return *this;
+ if (tree != nullptr) {
+ CordzUpdateScope scope(contents_.cordz_info(), method);
+ if (tree->IsFlat() && tree->flat()->Capacity() >= length &&
+ tree->refcount.IsMutable()) {
+ // Copy in place if the existing FLAT node is reusable.
+ memmove(tree->flat()->Data(), data, length);
+ tree->length = length;
+ VerifyTree(tree);
+ return *this;
+ }
+ contents_.SetTree(NewTree(data, length, 0), scope);
+ CordRep::Unref(tree);
+ } else {
+ contents_.EmplaceTree(NewTree(data, length, 0), method);
}
- contents_.set_tree(NewTree(data, length, 0));
- Unref(tree);
return *this;
}
// TODO(sanjay): Move to Cord::InlineRep section of file. For now,
// we keep it here to make diffs easier.
-void Cord::InlineRep::AppendArray(const char* src_data, size_t src_size) {
- if (src_size == 0) return; // memcpy(_, nullptr, 0) is undefined.
- // Try to fit in the inline buffer if possible.
- size_t inline_length = data_[kMaxInline];
- if (inline_length < kMaxInline && src_size <= kMaxInline - inline_length) {
- // Append new data to embedded array
- data_[kMaxInline] = static_cast<char>(inline_length + src_size);
- memcpy(data_ + inline_length, src_data, src_size);
- return;
- }
-
- CordRep* root = tree();
+void Cord::InlineRep::AppendArray(absl::string_view src,
+ MethodIdentifier method) {
+ if (src.empty()) return; // memcpy(_, nullptr, 0) is undefined.
size_t appended = 0;
- if (root) {
+ CordRep* rep = tree();
+ const CordRep* const root = rep;
+ CordzUpdateScope scope(root ? cordz_info() : nullptr, method);
+ if (root != nullptr) {
char* region;
- if (PrepareAppendRegion(root, &region, &appended, src_size)) {
- memcpy(region, src_data, appended);
+ if (PrepareAppendRegion(rep, &region, &appended, src.size())) {
+ memcpy(region, src.data(), appended);
}
} else {
- // It is possible that src_data == data_, but when we transition from an
- // InlineRep to a tree we need to assign data_ = root via set_tree. To
- // avoid corrupting the source data before we copy it, delay calling
- // set_tree until after we've copied data.
- // We are going from an inline size to beyond inline size. Make the new size
- // either double the inlined size, or the added size + 10%.
- const size_t size1 = inline_length * 2 + src_size;
- const size_t size2 = inline_length + src_size / 10;
- root = NewFlat(std::max<size_t>(size1, size2));
- appended = std::min(src_size, TagToLength(root->tag) - inline_length);
- memcpy(root->data, data_, inline_length);
- memcpy(root->data + inline_length, src_data, appended);
- root->length = inline_length + appended;
- set_tree(root);
- }
-
- src_data += appended;
- src_size -= appended;
- if (src_size == 0) {
+ // Try to fit in the inline buffer if possible.
+ size_t inline_length = inline_size();
+ if (src.size() <= kMaxInline - inline_length) {
+ // Append new data to embedded array
+ memcpy(data_.as_chars() + inline_length, src.data(), src.size());
+ set_inline_size(inline_length + src.size());
+ return;
+ }
+
+ // Allocate flat to be a perfect fit on first append exceeding inlined size.
+ // Subsequent growth will use amortized growth until we reach maximum flat
+ // size.
+ rep = CordRepFlat::New(inline_length + src.size());
+ appended = std::min(src.size(), rep->flat()->Capacity() - inline_length);
+ memcpy(rep->flat()->Data(), data_.as_chars(), inline_length);
+ memcpy(rep->flat()->Data() + inline_length, src.data(), appended);
+ rep->length = inline_length + appended;
+ }
+
+ src.remove_prefix(appended);
+ if (src.empty()) {
+ CommitTree(root, rep, scope, method);
return;
}
- // Use new block(s) for any remaining bytes that were not handled above.
- // Alloc extra memory only if the right child of the root of the new tree is
- // going to be a FLAT node, which will permit further inplace appends.
- size_t length = src_size;
- if (src_size < kMaxFlatLength) {
- // The new length is either
- // - old size + 10%
- // - old_size + src_size
- // This will cause a reasonable conservative step-up in size that is still
- // large enough to avoid excessive amounts of small fragments being added.
- length = std::max<size_t>(root->length / 10, src_size);
+ if (btree_enabled()) {
+ // TODO(b/192061034): keep legacy 10% growth rate: consider other rates.
+ rep = ForceBtree(rep);
+ const size_t min_growth = std::max<size_t>(rep->length / 10, src.size());
+ rep = CordRepBtree::Append(rep->btree(), src, min_growth - src.size());
+ } else {
+ // Use new block(s) for any remaining bytes that were not handled above.
+ // Alloc extra memory only if the right child of the root of the new tree
+ // is going to be a FLAT node, which will permit further inplace appends.
+ size_t length = src.size();
+ if (src.size() < kMaxFlatLength) {
+ // The new length is either
+ // - old size + 10%
+ // - old_size + src.size()
+ // This will cause a reasonable conservative step-up in size that is
+ // still large enough to avoid excessive amounts of small fragments
+ // being added.
+ length = std::max<size_t>(rep->length / 10, src.size());
+ }
+ rep = Concat(rep, NewTree(src.data(), src.size(), length - src.size()));
}
- set_tree(Concat(root, NewTree(src_data, src_size, length - src_size)));
+ CommitTree(root, rep, scope, method);
}
inline CordRep* Cord::TakeRep() const& {
- return Ref(contents_.tree());
+ return CordRep::Ref(contents_.tree());
}
inline CordRep* Cord::TakeRep() && {
@@ -846,10 +741,17 @@ inline CordRep* Cord::TakeRep() && {
template <typename C>
inline void Cord::AppendImpl(C&& src) {
+ auto constexpr method = CordzUpdateTracker::kAppendCord;
if (empty()) {
- // In case of an empty destination avoid allocating a new node, do not copy
- // data.
- *this = std::forward<C>(src);
+ // Since destination is empty, we can avoid allocating a node,
+ if (src.contents_.is_tree()) {
+ // by taking the tree directly
+ CordRep* rep = std::forward<C>(src).TakeRep();
+ contents_.EmplaceTree(rep, method);
+ } else {
+ // or copying over inline data
+ contents_.data_ = src.contents_.data_;
+ }
return;
}
@@ -859,12 +761,12 @@ inline void Cord::AppendImpl(C&& src) {
CordRep* src_tree = src.contents_.tree();
if (src_tree == nullptr) {
// src has embedded data.
- contents_.AppendArray(src.contents_.data(), src_size);
+ contents_.AppendArray({src.contents_.data(), src_size}, method);
return;
}
- if (src_tree->tag >= FLAT) {
+ if (src_tree->IsFlat()) {
// src tree just has one flat node.
- contents_.AppendArray(src_tree->data, src_size);
+ contents_.AppendArray({src_tree->flat()->Data(), src_size}, method);
return;
}
if (&src == this) {
@@ -879,18 +781,36 @@ inline void Cord::AppendImpl(C&& src) {
return;
}
- contents_.AppendTree(std::forward<C>(src).TakeRep());
+ // Guaranteed to be a tree (kMaxBytesToCopy > kInlinedSize)
+ CordRep* rep = std::forward<C>(src).TakeRep();
+ contents_.AppendTree(rep, CordzUpdateTracker::kAppendCord);
+}
+
+void Cord::Append(const Cord& src) {
+ AppendImpl(src);
}
-void Cord::Append(const Cord& src) { AppendImpl(src); }
+void Cord::Append(Cord&& src) {
+ AppendImpl(std::move(src));
+}
-void Cord::Append(Cord&& src) { AppendImpl(std::move(src)); }
+template <typename T, Cord::EnableIfString<T>>
+void Cord::Append(T&& src) {
+ if (src.size() <= kMaxBytesToCopy) {
+ Append(absl::string_view(src));
+ } else {
+ CordRep* rep = CordRepFromString(std::forward<T>(src));
+ contents_.AppendTree(rep, CordzUpdateTracker::kAppendString);
+ }
+}
+
+template void Cord::Append(std::string&& src);
void Cord::Prepend(const Cord& src) {
CordRep* src_tree = src.contents_.tree();
if (src_tree != nullptr) {
- Ref(src_tree);
- contents_.PrependTree(src_tree);
+ CordRep::Ref(src_tree);
+ contents_.PrependTree(src_tree, CordzUpdateTracker::kPrependCord);
return;
}
@@ -899,28 +819,42 @@ void Cord::Prepend(const Cord& src) {
return Prepend(src_contents);
}
-void Cord::Prepend(absl::string_view src) {
+void Cord::PrependArray(absl::string_view src, MethodIdentifier method) {
if (src.empty()) return; // memcpy(_, nullptr, 0) is undefined.
- size_t cur_size = contents_.size();
- if (!contents_.is_tree() && cur_size + src.size() <= InlineRep::kMaxInline) {
- // Use embedded storage.
- char data[InlineRep::kMaxInline + 1] = {0};
- data[InlineRep::kMaxInline] = cur_size + src.size(); // set size
- memcpy(data, src.data(), src.size());
- memcpy(data + src.size(), contents_.data(), cur_size);
- memcpy(reinterpret_cast<void*>(&contents_), data,
- InlineRep::kMaxInline + 1);
+ if (!contents_.is_tree()) {
+ size_t cur_size = contents_.inline_size();
+ if (cur_size + src.size() <= InlineRep::kMaxInline) {
+ // Use embedded storage.
+ char data[InlineRep::kMaxInline + 1] = {0};
+ memcpy(data, src.data(), src.size());
+ memcpy(data + src.size(), contents_.data(), cur_size);
+ memcpy(contents_.data_.as_chars(), data, InlineRep::kMaxInline + 1);
+ contents_.set_inline_size(cur_size + src.size());
+ return;
+ }
+ }
+ CordRep* rep = NewTree(src.data(), src.size(), 0);
+ contents_.PrependTree(rep, method);
+}
+
+template <typename T, Cord::EnableIfString<T>>
+inline void Cord::Prepend(T&& src) {
+ if (src.size() <= kMaxBytesToCopy) {
+ Prepend(absl::string_view(src));
} else {
- contents_.PrependTree(NewTree(src.data(), src.size(), 0));
+ CordRep* rep = CordRepFromString(std::forward<T>(src));
+ contents_.PrependTree(rep, CordzUpdateTracker::kPrependString);
}
}
+template void Cord::Prepend(std::string&& src);
+
static CordRep* RemovePrefixFrom(CordRep* node, size_t n) {
if (n >= node->length) return nullptr;
- if (n == 0) return Ref(node);
+ if (n == 0) return CordRep::Ref(node);
absl::InlinedVector<CordRep*, kInlinedVectorSize> rhs_stack;
- while (node->tag == CONCAT) {
+ while (node->IsConcat()) {
assert(n <= node->length);
if (n < node->concat()->left->length) {
// Push right to stack, descend left.
@@ -935,19 +869,19 @@ static CordRep* RemovePrefixFrom(CordRep* node, size_t n) {
assert(n <= node->length);
if (n == 0) {
- Ref(node);
+ CordRep::Ref(node);
} else {
size_t start = n;
size_t len = node->length - n;
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
// Consider in-place update of node, similar to in RemoveSuffixFrom().
start += node->substring()->start;
node = node->substring()->child;
}
- node = NewSubstring(Ref(node), start, len);
+ node = NewSubstring(CordRep::Ref(node), start, len);
}
while (!rhs_stack.empty()) {
- node = Concat(node, Ref(rhs_stack.back()));
+ node = Concat(node, CordRep::Ref(rhs_stack.back()));
rhs_stack.pop_back();
}
return node;
@@ -958,11 +892,11 @@ static CordRep* RemovePrefixFrom(CordRep* node, size_t n) {
// edited in place iff that node and all its ancestors have a refcount of 1.
static CordRep* RemoveSuffixFrom(CordRep* node, size_t n) {
if (n >= node->length) return nullptr;
- if (n == 0) return Ref(node);
+ if (n == 0) return CordRep::Ref(node);
absl::InlinedVector<CordRep*, kInlinedVectorSize> lhs_stack;
- bool inplace_ok = node->refcount.IsOne();
+ bool inplace_ok = node->refcount.IsMutable();
- while (node->tag == CONCAT) {
+ while (node->IsConcat()) {
assert(n <= node->length);
if (n < node->concat()->right->length) {
// Push left to stack, descend right.
@@ -973,28 +907,28 @@ static CordRep* RemoveSuffixFrom(CordRep* node, size_t n) {
n -= node->concat()->right->length;
node = node->concat()->left;
}
- inplace_ok = inplace_ok && node->refcount.IsOne();
+ inplace_ok = inplace_ok && node->refcount.IsMutable();
}
assert(n <= node->length);
if (n == 0) {
- Ref(node);
- } else if (inplace_ok && node->tag != EXTERNAL) {
+ CordRep::Ref(node);
+ } else if (inplace_ok && !node->IsExternal()) {
// Consider making a new buffer if the current node capacity is much
// larger than the new length.
- Ref(node);
+ CordRep::Ref(node);
node->length -= n;
} else {
size_t start = 0;
size_t len = node->length - n;
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
start = node->substring()->start;
node = node->substring()->child;
}
- node = NewSubstring(Ref(node), start, len);
+ node = NewSubstring(CordRep::Ref(node), start, len);
}
while (!lhs_stack.empty()) {
- node = Concat(Ref(lhs_stack.back()), node);
+ node = Concat(CordRep::Ref(lhs_stack.back()), node);
lhs_stack.pop_back();
}
return node;
@@ -1008,9 +942,18 @@ void Cord::RemovePrefix(size_t n) {
if (tree == nullptr) {
contents_.remove_prefix(n);
} else {
- CordRep* newrep = RemovePrefixFrom(tree, n);
- Unref(tree);
- contents_.replace_tree(VerifyTree(newrep));
+ auto constexpr method = CordzUpdateTracker::kRemovePrefix;
+ CordzUpdateScope scope(contents_.cordz_info(), method);
+ if (tree->IsBtree()) {
+ CordRep* old = tree;
+ tree = tree->btree()->SubTree(n, tree->length - n);
+ CordRep::Unref(old);
+ } else {
+ CordRep* newrep = RemovePrefixFrom(tree, n);
+ CordRep::Unref(tree);
+ tree = VerifyTree(newrep);
+ }
+ contents_.SetTreeOrEmpty(tree, scope);
}
}
@@ -1022,9 +965,16 @@ void Cord::RemoveSuffix(size_t n) {
if (tree == nullptr) {
contents_.reduce_size(n);
} else {
- CordRep* newrep = RemoveSuffixFrom(tree, n);
- Unref(tree);
- contents_.replace_tree(VerifyTree(newrep));
+ auto constexpr method = CordzUpdateTracker::kRemoveSuffix;
+ CordzUpdateScope scope(contents_.cordz_info(), method);
+ if (tree->IsBtree()) {
+ tree = CordRepBtree::RemoveSuffix(tree->btree(), n);
+ } else {
+ CordRep* newrep = RemoveSuffixFrom(tree, n);
+ CordRep::Unref(tree);
+ tree = VerifyTree(newrep);
+ }
+ contents_.SetTreeOrEmpty(tree, scope);
}
}
@@ -1056,13 +1006,13 @@ static CordRep* NewSubRange(CordRep* node, size_t pos, size_t n) {
results.pop_back();
results.push_back(Concat(left, right));
} else if (pos == 0 && n == node->length) {
- results.push_back(Ref(node));
- } else if (node->tag != CONCAT) {
- if (node->tag == SUBSTRING) {
+ results.push_back(CordRep::Ref(node));
+ } else if (!node->IsConcat()) {
+ if (node->IsSubstring()) {
pos += node->substring()->start;
node = node->substring()->child;
}
- results.push_back(NewSubstring(Ref(node), pos, n));
+ results.push_back(NewSubstring(CordRep::Ref(node), pos, n));
} else if (pos + n <= node->concat()->left->length) {
todo.push_back(SubRange(node->concat()->left, pos, n));
} else if (pos >= node->concat()->left->length) {
@@ -1084,17 +1034,20 @@ Cord Cord::Subcord(size_t pos, size_t new_size) const {
size_t length = size();
if (pos > length) pos = length;
if (new_size > length - pos) new_size = length - pos;
+ if (new_size == 0) return sub_cord;
+
CordRep* tree = contents_.tree();
if (tree == nullptr) {
// sub_cord is newly constructed, no need to re-zero-out the tail of
// contents_ memory.
sub_cord.contents_.set_data(contents_.data() + pos, new_size, false);
- } else if (new_size == 0) {
- // We want to return empty subcord, so nothing to do.
- } else if (new_size <= InlineRep::kMaxInline) {
+ return sub_cord;
+ }
+
+ if (new_size <= InlineRep::kMaxInline) {
+ char* dest = sub_cord.contents_.data_.as_chars();
Cord::ChunkIterator it = chunk_begin();
it.AdvanceBytes(pos);
- char* dest = sub_cord.contents_.data_;
size_t remaining_size = new_size;
while (remaining_size > it->size()) {
cord_internal::SmallMemmove(dest, it->data(), it->size());
@@ -1103,10 +1056,17 @@ Cord Cord::Subcord(size_t pos, size_t new_size) const {
++it;
}
cord_internal::SmallMemmove(dest, it->data(), remaining_size);
- sub_cord.contents_.data_[InlineRep::kMaxInline] = new_size;
+ sub_cord.contents_.set_inline_size(new_size);
+ return sub_cord;
+ }
+
+ if (tree->IsBtree()) {
+ tree = tree->btree()->SubTree(pos, new_size);
} else {
- sub_cord.contents_.set_tree(NewSubRange(tree, pos, new_size));
+ tree = NewSubRange(tree, pos, new_size);
}
+ sub_cord.contents_.EmplaceTree(tree, contents_.data_,
+ CordzUpdateTracker::kSubCord);
return sub_cord;
}
@@ -1125,7 +1085,7 @@ class CordForest {
CordRep* node = pending.back();
pending.pop_back();
CheckNode(node);
- if (ABSL_PREDICT_FALSE(node->tag != CONCAT)) {
+ if (ABSL_PREDICT_FALSE(!node->IsConcat())) {
AddNode(node);
continue;
}
@@ -1140,9 +1100,9 @@ class CordForest {
concat_node->left = concat_freelist_;
concat_freelist_ = concat_node;
} else {
- Ref(concat_node->right);
- Ref(concat_node->left);
- Unref(concat_node);
+ CordRep::Ref(concat_node->right);
+ CordRep::Ref(concat_node->left);
+ CordRep::Unref(concat_node);
}
} else {
AddNode(node);
@@ -1175,7 +1135,7 @@ class CordForest {
void AddNode(CordRep* node) {
CordRep* sum = nullptr;
- // Collect together everything with which we will merge node
+ // Collect together everything with which we will merge with node
int i = 0;
for (; node->length > min_length[i + 1]; ++i) {
auto& tree_at_i = trees_[i];
@@ -1219,7 +1179,7 @@ class CordForest {
static void CheckNode(CordRep* node) {
ABSL_INTERNAL_CHECK(node->length != 0u, "");
- if (node->tag == CONCAT) {
+ if (node->IsConcat()) {
ABSL_INTERNAL_CHECK(node->concat()->left != nullptr, "");
ABSL_INTERNAL_CHECK(node->concat()->right != nullptr, "");
ABSL_INTERNAL_CHECK(node->length == (node->concat()->left->length +
@@ -1239,7 +1199,7 @@ class CordForest {
static CordRep* Rebalance(CordRep* node) {
VerifyTree(node);
- assert(node->tag == CONCAT);
+ assert(node->IsConcat());
if (node->length == 0) {
return nullptr;
@@ -1289,25 +1249,33 @@ bool ComputeCompareResult<bool>(int memcmp_res) {
} // namespace
-// Helper routine. Locates the first flat chunk of the Cord without
-// initializing the iterator.
+// Helper routine. Locates the first flat or external chunk of the Cord without
+// initializing the iterator, and returns a string_view referencing the data.
inline absl::string_view Cord::InlineRep::FindFlatStartPiece() const {
- size_t n = data_[kMaxInline];
- if (n <= kMaxInline) {
- return absl::string_view(data_, n);
+ if (!is_tree()) {
+ return absl::string_view(data_.as_chars(), data_.inline_size());
}
CordRep* node = tree();
- if (node->tag >= FLAT) {
- return absl::string_view(node->data, node->length);
+ if (node->IsFlat()) {
+ return absl::string_view(node->flat()->Data(), node->length);
}
- if (node->tag == EXTERNAL) {
+ if (node->IsExternal()) {
return absl::string_view(node->external()->base, node->length);
}
+ if (node->IsBtree()) {
+ CordRepBtree* tree = node->btree();
+ int height = tree->height();
+ while (--height >= 0) {
+ tree = tree->Edge(CordRepBtree::kFront)->btree();
+ }
+ return tree->Data(tree->begin());
+ }
+
// Walk down the left branches until we hit a non-CONCAT node.
- while (node->tag == CONCAT) {
+ while (node->IsConcat()) {
node = node->concat()->left;
}
@@ -1316,16 +1284,16 @@ inline absl::string_view Cord::InlineRep::FindFlatStartPiece() const {
size_t length = node->length;
assert(length != 0);
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
offset = node->substring()->start;
node = node->substring()->child;
}
- if (node->tag >= FLAT) {
- return absl::string_view(node->data + offset, length);
+ if (node->IsFlat()) {
+ return absl::string_view(node->flat()->Data() + offset, length);
}
- assert((node->tag == EXTERNAL) && "Expect FLAT or EXTERNAL node here");
+ assert(node->IsExternal() && "Expect FLAT or EXTERNAL node here");
return absl::string_view(node->external()->base + offset, length);
}
@@ -1505,48 +1473,47 @@ void Cord::CopyToArraySlowPath(char* dst) const {
}
}
-Cord::ChunkIterator& Cord::ChunkIterator::operator++() {
- assert(bytes_remaining_ > 0 && "Attempted to iterate past `end()`");
- assert(bytes_remaining_ >= current_chunk_.size());
- bytes_remaining_ -= current_chunk_.size();
-
- if (stack_of_right_children_.empty()) {
+Cord::ChunkIterator& Cord::ChunkIterator::AdvanceStack() {
+ auto& stack_of_right_children = stack_of_right_children_;
+ if (stack_of_right_children.empty()) {
assert(!current_chunk_.empty()); // Called on invalid iterator.
// We have reached the end of the Cord.
return *this;
}
// Process the next node on the stack.
- CordRep* node = stack_of_right_children_.back();
- stack_of_right_children_.pop_back();
+ CordRep* node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
// Walk down the left branches until we hit a non-CONCAT node. Save the
// right children to the stack for subsequent traversal.
- while (node->tag == CONCAT) {
- stack_of_right_children_.push_back(node->concat()->right);
+ while (node->IsConcat()) {
+ stack_of_right_children.push_back(node->concat()->right);
node = node->concat()->left;
}
// Get the child node if we encounter a SUBSTRING.
size_t offset = 0;
size_t length = node->length;
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
offset = node->substring()->start;
node = node->substring()->child;
}
- assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(node->IsExternal() || node->IsFlat());
assert(length != 0);
const char* data =
- node->tag == EXTERNAL ? node->external()->base : node->data;
+ node->IsExternal() ? node->external()->base : node->flat()->Data();
current_chunk_ = absl::string_view(data + offset, length);
current_leaf_ = node;
return *this;
}
Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
- assert(bytes_remaining_ >= n && "Attempted to iterate past `end()`");
+ ABSL_HARDENING_ASSERT(bytes_remaining_ >= n &&
+ "Attempted to iterate past `end()`");
Cord subcord;
+ auto constexpr method = CordzUpdateTracker::kCordReader;
if (n <= InlineRep::kMaxInline) {
// Range to read fits in inline data. Flatten it.
@@ -1565,14 +1532,34 @@ Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
}
return subcord;
}
+
+ if (btree_reader_) {
+ size_t chunk_size = current_chunk_.size();
+ if (n <= chunk_size && n <= kMaxBytesToCopy) {
+ subcord = Cord(current_chunk_.substr(0, n), method);
+ if (n < chunk_size) {
+ current_chunk_.remove_prefix(n);
+ } else {
+ current_chunk_ = btree_reader_.Next();
+ }
+ } else {
+ CordRep* rep;
+ current_chunk_ = btree_reader_.Read(n, chunk_size, rep);
+ subcord.contents_.EmplaceTree(rep, method);
+ }
+ bytes_remaining_ -= n;
+ return subcord;
+ }
+
+ auto& stack_of_right_children = stack_of_right_children_;
if (n < current_chunk_.size()) {
// Range to read is a proper subrange of the current chunk.
assert(current_leaf_ != nullptr);
- CordRep* subnode = Ref(current_leaf_);
- const char* data =
- subnode->tag == EXTERNAL ? subnode->external()->base : subnode->data;
+ CordRep* subnode = CordRep::Ref(current_leaf_);
+ const char* data = subnode->IsExternal() ? subnode->external()->base
+ : subnode->flat()->Data();
subnode = NewSubstring(subnode, current_chunk_.data() - data, n);
- subcord.contents_.set_tree(VerifyTree(subnode));
+ subcord.contents_.EmplaceTree(VerifyTree(subnode), method);
RemoveChunkPrefix(n);
return subcord;
}
@@ -1580,10 +1567,10 @@ Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
// Range to read begins with a proper subrange of the current chunk.
assert(!current_chunk_.empty());
assert(current_leaf_ != nullptr);
- CordRep* subnode = Ref(current_leaf_);
+ CordRep* subnode = CordRep::Ref(current_leaf_);
if (current_chunk_.size() < subnode->length) {
- const char* data =
- subnode->tag == EXTERNAL ? subnode->external()->base : subnode->data;
+ const char* data = subnode->IsExternal() ? subnode->external()->base
+ : subnode->flat()->Data();
subnode = NewSubstring(subnode, current_chunk_.data() - data,
current_chunk_.size());
}
@@ -1593,20 +1580,20 @@ Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
// Process the next node(s) on the stack, reading whole subtrees depending on
// their length and how many bytes we are advancing.
CordRep* node = nullptr;
- while (!stack_of_right_children_.empty()) {
- node = stack_of_right_children_.back();
- stack_of_right_children_.pop_back();
+ while (!stack_of_right_children.empty()) {
+ node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
if (node->length > n) break;
// TODO(qrczak): This might unnecessarily recreate existing concat nodes.
// Avoiding that would need pretty complicated logic (instead of
- // current_leaf_, keep current_subtree_ which points to the highest node
+ // current_leaf, keep current_subtree_ which points to the highest node
// such that the current leaf can be found on the path of left children
// starting from current_subtree_; delay creating subnode while node is
// below current_subtree_; find the proper node along the path of left
// children starting from current_subtree_ if this loop exits while staying
// below current_subtree_; etc.; alternatively, push parents instead of
// right children on the stack).
- subnode = Concat(subnode, Ref(node));
+ subnode = Concat(subnode, CordRep::Ref(node));
n -= node->length;
bytes_remaining_ -= node->length;
node = nullptr;
@@ -1615,20 +1602,20 @@ Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
if (node == nullptr) {
// We have reached the end of the Cord.
assert(bytes_remaining_ == 0);
- subcord.contents_.set_tree(VerifyTree(subnode));
+ subcord.contents_.EmplaceTree(VerifyTree(subnode), method);
return subcord;
}
// Walk down the appropriate branches until we hit a non-CONCAT node. Save the
// right children to the stack for subsequent traversal.
- while (node->tag == CONCAT) {
+ while (node->IsConcat()) {
if (node->concat()->left->length > n) {
// Push right, descend left.
- stack_of_right_children_.push_back(node->concat()->right);
+ stack_of_right_children.push_back(node->concat()->right);
node = node->concat()->left;
} else {
// Read left, descend right.
- subnode = Concat(subnode, Ref(node->concat()->left));
+ subnode = Concat(subnode, CordRep::Ref(node->concat()->left));
n -= node->concat()->left->length;
bytes_remaining_ -= node->concat()->left->length;
node = node->concat()->right;
@@ -1638,22 +1625,24 @@ Cord Cord::ChunkIterator::AdvanceAndReadBytes(size_t n) {
// Get the child node if we encounter a SUBSTRING.
size_t offset = 0;
size_t length = node->length;
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
offset = node->substring()->start;
node = node->substring()->child;
}
// Range to read ends with a proper (possibly empty) subrange of the current
// chunk.
- assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(node->IsExternal() || node->IsFlat());
assert(length > n);
- if (n > 0) subnode = Concat(subnode, NewSubstring(Ref(node), offset, n));
+ if (n > 0) {
+ subnode = Concat(subnode, NewSubstring(CordRep::Ref(node), offset, n));
+ }
const char* data =
- node->tag == EXTERNAL ? node->external()->base : node->data;
+ node->IsExternal() ? node->external()->base : node->flat()->Data();
current_chunk_ = absl::string_view(data + offset + n, length - n);
current_leaf_ = node;
bytes_remaining_ -= n;
- subcord.contents_.set_tree(VerifyTree(subnode));
+ subcord.contents_.EmplaceTree(VerifyTree(subnode), method);
return subcord;
}
@@ -1665,12 +1654,19 @@ void Cord::ChunkIterator::AdvanceBytesSlowPath(size_t n) {
n -= current_chunk_.size();
bytes_remaining_ -= current_chunk_.size();
+ if (stack_of_right_children_.empty()) {
+ // We have reached the end of the Cord.
+ assert(bytes_remaining_ == 0);
+ return;
+ }
+
// Process the next node(s) on the stack, skipping whole subtrees depending on
// their length and how many bytes we are advancing.
CordRep* node = nullptr;
- while (!stack_of_right_children_.empty()) {
- node = stack_of_right_children_.back();
- stack_of_right_children_.pop_back();
+ auto& stack_of_right_children = stack_of_right_children_;
+ while (!stack_of_right_children.empty()) {
+ node = stack_of_right_children.back();
+ stack_of_right_children.pop_back();
if (node->length > n) break;
n -= node->length;
bytes_remaining_ -= node->length;
@@ -1685,10 +1681,10 @@ void Cord::ChunkIterator::AdvanceBytesSlowPath(size_t n) {
// Walk down the appropriate branches until we hit a non-CONCAT node. Save the
// right children to the stack for subsequent traversal.
- while (node->tag == CONCAT) {
+ while (node->IsConcat()) {
if (node->concat()->left->length > n) {
// Push right, descend left.
- stack_of_right_children_.push_back(node->concat()->right);
+ stack_of_right_children.push_back(node->concat()->right);
node = node->concat()->left;
} else {
// Skip left, descend right.
@@ -1701,22 +1697,22 @@ void Cord::ChunkIterator::AdvanceBytesSlowPath(size_t n) {
// Get the child node if we encounter a SUBSTRING.
size_t offset = 0;
size_t length = node->length;
- if (node->tag == SUBSTRING) {
+ if (node->IsSubstring()) {
offset = node->substring()->start;
node = node->substring()->child;
}
- assert(node->tag == EXTERNAL || node->tag >= FLAT);
+ assert(node->IsExternal() || node->IsFlat());
assert(length > n);
const char* data =
- node->tag == EXTERNAL ? node->external()->base : node->data;
+ node->IsExternal() ? node->external()->base : node->flat()->Data();
current_chunk_ = absl::string_view(data + offset + n, length - n);
current_leaf_ = node;
bytes_remaining_ -= n;
}
char Cord::operator[](size_t i) const {
- assert(i < size());
+ ABSL_HARDENING_ASSERT(i < size());
size_t offset = i;
const CordRep* rep = contents_.tree();
if (rep == nullptr) {
@@ -1725,13 +1721,15 @@ char Cord::operator[](size_t i) const {
while (true) {
assert(rep != nullptr);
assert(offset < rep->length);
- if (rep->tag >= FLAT) {
+ if (rep->IsFlat()) {
// Get the "i"th character directly from the flat array.
- return rep->data[offset];
- } else if (rep->tag == EXTERNAL) {
+ return rep->flat()->Data()[offset];
+ } else if (rep->IsBtree()) {
+ return rep->btree()->GetCharacter(offset);
+ } else if (rep->IsExternal()) {
// Get the "i"th character from the external array.
return rep->external()->base[offset];
- } else if (rep->tag == CONCAT) {
+ } else if (rep->IsConcat()) {
// Recursively branch to the side of the concatenation that the "i"th
// character is on.
size_t left_length = rep->concat()->left->length;
@@ -1743,7 +1741,7 @@ char Cord::operator[](size_t i) const {
}
} else {
// This must be a substring a node, so bypass it to get to the child.
- assert(rep->tag == SUBSTRING);
+ assert(rep->IsSubstring());
offset += rep->substring()->start;
rep = rep->substring()->child;
}
@@ -1751,6 +1749,7 @@ char Cord::operator[](size_t i) const {
}
absl::string_view Cord::FlattenSlowPath() {
+ assert(contents_.is_tree());
size_t total_size = size();
CordRep* new_rep;
char* new_buffer;
@@ -1758,9 +1757,9 @@ absl::string_view Cord::FlattenSlowPath() {
// Try to put the contents into a new flat rep. If they won't fit in the
// biggest possible flat node, use an external rep instead.
if (total_size <= kMaxFlatLength) {
- new_rep = NewFlat(total_size);
+ new_rep = CordRepFlat::New(total_size);
new_rep->length = total_size;
- new_buffer = new_rep->data;
+ new_buffer = new_rep->flat()->Data();
CopyToArraySlowPath(new_buffer);
} else {
new_buffer = std::allocator<char>().allocate(total_size);
@@ -1771,29 +1770,35 @@ absl::string_view Cord::FlattenSlowPath() {
s.size());
});
}
- Unref(contents_.tree());
- contents_.set_tree(new_rep);
+ CordzUpdateScope scope(contents_.cordz_info(), CordzUpdateTracker::kFlatten);
+ CordRep::Unref(contents_.as_tree());
+ contents_.SetTree(new_rep, scope);
return absl::string_view(new_buffer, total_size);
}
/* static */ bool Cord::GetFlatAux(CordRep* rep, absl::string_view* fragment) {
assert(rep != nullptr);
- if (rep->tag >= FLAT) {
- *fragment = absl::string_view(rep->data, rep->length);
+ if (rep->IsFlat()) {
+ *fragment = absl::string_view(rep->flat()->Data(), rep->length);
return true;
- } else if (rep->tag == EXTERNAL) {
+ } else if (rep->IsExternal()) {
*fragment = absl::string_view(rep->external()->base, rep->length);
return true;
- } else if (rep->tag == SUBSTRING) {
+ } else if (rep->IsBtree()) {
+ return rep->btree()->IsFlat(fragment);
+ } else if (rep->IsSubstring()) {
CordRep* child = rep->substring()->child;
- if (child->tag >= FLAT) {
- *fragment =
- absl::string_view(child->data + rep->substring()->start, rep->length);
+ if (child->IsFlat()) {
+ *fragment = absl::string_view(
+ child->flat()->Data() + rep->substring()->start, rep->length);
return true;
- } else if (child->tag == EXTERNAL) {
+ } else if (child->IsExternal()) {
*fragment = absl::string_view(
child->external()->base + rep->substring()->start, rep->length);
return true;
+ } else if (child->IsBtree()) {
+ return child->btree()->IsFlat(rep->substring()->start, rep->length,
+ fragment);
}
}
return false;
@@ -1802,6 +1807,15 @@ absl::string_view Cord::FlattenSlowPath() {
/* static */ void Cord::ForEachChunkAux(
absl::cord_internal::CordRep* rep,
absl::FunctionRef<void(absl::string_view)> callback) {
+ if (rep->IsBtree()) {
+ ChunkIterator it(rep), end;
+ while (it != end) {
+ callback(*it);
+ ++it;
+ }
+ return;
+ }
+
assert(rep != nullptr);
int stack_pos = 0;
constexpr int stack_max = 128;
@@ -1809,7 +1823,7 @@ absl::string_view Cord::FlattenSlowPath() {
absl::cord_internal::CordRep* stack[stack_max];
absl::cord_internal::CordRep* current_node = rep;
while (true) {
- if (current_node->tag == CONCAT) {
+ if (current_node->IsConcat()) {
if (stack_pos == stack_max) {
// There's no more room on our stack array to add another right branch,
// and the idea is to avoid allocations, so call this function
@@ -1843,9 +1857,9 @@ absl::string_view Cord::FlattenSlowPath() {
}
}
-static void DumpNode(CordRep* rep, bool include_data, std::ostream* os) {
+static void DumpNode(CordRep* rep, bool include_data, std::ostream* os,
+ int indent) {
const int kIndentStep = 1;
- int indent = 0;
absl::InlinedVector<CordRep*, kInlinedVectorSize> stack;
absl::InlinedVector<int, kInlinedVectorSize> indents;
for (;;) {
@@ -1856,27 +1870,29 @@ static void DumpNode(CordRep* rep, bool include_data, std::ostream* os) {
*os << "]";
*os << " " << (IsRootBalanced(rep) ? 'b' : 'u');
*os << " " << std::setw(indent) << "";
- if (rep->tag == CONCAT) {
+ if (rep->IsConcat()) {
*os << "CONCAT depth=" << Depth(rep) << "\n";
indent += kIndentStep;
indents.push_back(indent);
stack.push_back(rep->concat()->right);
rep = rep->concat()->left;
- } else if (rep->tag == SUBSTRING) {
+ } else if (rep->IsSubstring()) {
*os << "SUBSTRING @ " << rep->substring()->start << "\n";
indent += kIndentStep;
rep = rep->substring()->child;
- } else { // Leaf
- if (rep->tag == EXTERNAL) {
+ } else { // Leaf or ring
+ if (rep->IsExternal()) {
*os << "EXTERNAL [";
if (include_data)
*os << absl::CEscape(std::string(rep->external()->base, rep->length));
*os << "]\n";
- } else {
- *os << "FLAT cap=" << TagToLength(rep->tag) << " [";
+ } else if (rep->IsFlat()) {
+ *os << "FLAT cap=" << rep->flat()->Capacity() << " [";
if (include_data)
- *os << absl::CEscape(std::string(rep->data, rep->length));
+ *os << absl::CEscape(std::string(rep->flat()->Data(), rep->length));
*os << "]\n";
+ } else {
+ CordRepBtree::Dump(rep, /*label=*/ "", include_data, *os);
}
if (stack.empty()) break;
rep = stack.back();
@@ -1908,7 +1924,7 @@ static bool VerifyNode(CordRep* root, CordRep* start_node,
ABSL_INTERNAL_CHECK(node->length != 0, ReportError(root, node));
}
- if (node->tag == CONCAT) {
+ if (node->IsConcat()) {
ABSL_INTERNAL_CHECK(node->concat()->left != nullptr,
ReportError(root, node));
ABSL_INTERNAL_CHECK(node->concat()->right != nullptr,
@@ -1920,13 +1936,13 @@ static bool VerifyNode(CordRep* root, CordRep* start_node,
worklist.push_back(node->concat()->right);
worklist.push_back(node->concat()->left);
}
- } else if (node->tag >= FLAT) {
- ABSL_INTERNAL_CHECK(node->length <= TagToLength(node->tag),
+ } else if (node->IsFlat()) {
+ ABSL_INTERNAL_CHECK(node->length <= node->flat()->Capacity(),
ReportError(root, node));
- } else if (node->tag == EXTERNAL) {
+ } else if (node->IsExternal()) {
ABSL_INTERNAL_CHECK(node->external()->base != nullptr,
ReportError(root, node));
- } else if (node->tag == SUBSTRING) {
+ } else if (node->IsSubstring()) {
ABSL_INTERNAL_CHECK(
node->substring()->start < node->substring()->child->length,
ReportError(root, node));
@@ -1955,7 +1971,7 @@ static bool VerifyNode(CordRep* root, CordRep* start_node,
while (true) {
const CordRep* next_node = nullptr;
- if (cur_node->tag == CONCAT) {
+ if (cur_node->IsConcat()) {
total_mem_usage += sizeof(CordRepConcat);
const CordRep* left = cur_node->concat()->left;
if (!RepMemoryUsageLeaf(left, &total_mem_usage)) {
@@ -1969,9 +1985,21 @@ static bool VerifyNode(CordRep* root, CordRep* start_node,
}
next_node = right;
}
+ } else if (cur_node->IsBtree()) {
+ total_mem_usage += sizeof(CordRepBtree);
+ const CordRepBtree* node = cur_node->btree();
+ if (node->height() == 0) {
+ for (const CordRep* edge : node->Edges()) {
+ RepMemoryUsageDataEdge(edge, &total_mem_usage);
+ }
+ } else {
+ for (const CordRep* edge : node->Edges()) {
+ tree_stack.push_back(edge);
+ }
+ }
} else {
// Since cur_node is not a leaf or a concat node it must be a substring.
- assert(cur_node->tag == SUBSTRING);
+ assert(cur_node->IsSubstring());
total_mem_usage += sizeof(CordRepSubstring);
next_node = cur_node->substring()->child;
if (RepMemoryUsageLeaf(next_node, &total_mem_usage)) {
@@ -1998,14 +2026,14 @@ std::ostream& operator<<(std::ostream& out, const Cord& cord) {
}
namespace strings_internal {
-size_t CordTestAccess::FlatOverhead() { return kFlatOverhead; }
-size_t CordTestAccess::MaxFlatLength() { return kMaxFlatLength; }
+size_t CordTestAccess::FlatOverhead() { return cord_internal::kFlatOverhead; }
+size_t CordTestAccess::MaxFlatLength() { return cord_internal::kMaxFlatLength; }
size_t CordTestAccess::FlatTagToLength(uint8_t tag) {
- return TagToLength(tag);
+ return cord_internal::TagToLength(tag);
}
uint8_t CordTestAccess::LengthToTag(size_t s) {
ABSL_INTERNAL_CHECK(s <= kMaxFlatLength, absl::StrCat("Invalid length ", s));
- return AllocatedSizeToTag(s + kFlatOverhead);
+ return cord_internal::AllocatedSizeToTag(s + cord_internal::kFlatOverhead);
}
size_t CordTestAccess::SizeofCordRepConcat() { return sizeof(CordRepConcat); }
size_t CordTestAccess::SizeofCordRepExternal() {