SWIG (Simplified Wrapper and Interface Generator) Version 1.3.20 (In progress) ============================ Due to the size of the CHANGES file, please add change entries to the file CHANGES.current. It will be merged into this file before release. -- Dave Version 1.3.19 (March 28, 2003) =============================== 03/28/2003: beazley Variety of minor bug fixes to the 1.3.18 release including: - Segmentation fault with %extend directive. - Typemap variable substitution bug. - Expression evaluation bug. - Large memory leak with template expansion. Version 1.3.18 (March 23, 2003) =============================== 03/21/2003: beazley Fixed two problems with the %extend directive, overloading, and template expansion. See the 'template_extend_overload' and 'template_extend_overload_2' tests in Examples/test-suite for details. 03/20/2003: cheetah (William Fulton) [C#] Added some typemaps as suggested by Andreas Schoerk for handling parameters that are passed as pointers or by reference. These have been put in typemaps.i. 03/20/2003: beazley Fixed a C++ scoping bug related to code like this: class Foo { public: int Foo::bar(); }; Previously, SWIG just tossed out the Foo::bar() declaration. Now, the declaration is wrapped provided that the prefix is exactly the same as the current scope (including any enclosing namespaces). Reported by Bruce Lowery. 03/20/2003: beazley Incorporated [ 696516 ] Enabling exception processing for data member access. In some compilers, attribute access can generate exceptions. However, SWIG ordinarily assumes that no exceptions will be raised. To disable this, use the %feature("allowexcept"). For example: %feature("allowexcept") Foo::x; ... class Foo { public: int x; /* Exception handling enabled */ ... }; Patch contributed by Yakov Markovitch. 03/20/2003: beazley Incorporated Patch. [ 701860 ] Improve Performance (python proxies). Gives a performance boost to proxy class code and the management of the .this and .thisown attributes. Contributed by Mike Romberg. 03/19/2003: cheetah (William Fulton) [C# and Java] Added missing vararg support. 03/18/2003: mrose (Mark Rose) Removed code related to tagging individual methods for directors. The concept of having directors for some but not all virtual methods of a class is deeply flawed. The %feature("nodirector") tag is also gone. Directors are off by default. To enable them for a class, issue %feature("director") classname; which will create director methods for every virtual method in the hierarchy of the class. 03/17/2003: beazley Fixed a subtle problem with passing arguments of type function. For example: int foo(int x(int, int)); or typedef int binop_t(int, int); int foo(binop_t x); In old versions, this would produce code that wouldn't compile. Now, SWIG merely adds an extra pointer, making these declarations the same as: int foo(int (*x)(int, int)); typedef int binop_t(int, int); int foo(binop_t *x); Reported by Garth Bushell. 03/17/2003: mrose (Mark Rose) Fixed the return statement for director base class calls that have no return value. 03/15/2003: beazley Fixed a problem with const smart-pointer wrapping. For example: class Foo { public: int x; void bar() const; void spam(); }; class Blah { ... const Foo *operator->(); ... }; In this case, only "x" and "bar" are visible from Blah (since application of spam violates constness). Moreover, access to "x" is read-only. 03/15/2003: mrose (Mark Rose) Cleaned up two signed versus unsigned comparisons in python/std_vector.i. 03/15/2003: cheetah (William Fulton) [C#] Global variables are wrapped using properties instead of get and set methods. Member variable wrapping bug fixes, for example wrapping pointers work now. Typemaps are used for all variable wrapping to generate the property code. 03/13/2003: mrose (Mark Rose) Fixed a bug in the virtual method unrolling for directors. The order of unrolling is now from base to derived, to ensure that the most derived implementation of a director method is found. Director methods for pure virtual methods now throw DIRECTOR_PURE_VIRTUAL_EXCEPTION if _up is set. 03/12/2003: cheetah (William Fulton) [C#] Polymorphism fix: virtual functions now use the appropriate keyword in the C# proxy class, virtual or override. Some 'using System;' statement fixes needed by the Mono compiler. 03/11/2003: beazley Fixed subtle bug in the application of SwigValueWrapper<> to template classes with default constructors. Reported by Bruce Lowery. 03/11/2003: beazley The $descriptor(type) variable is now expanded in code supplied to %extend. This is useful for certain kinds of advanced wrapping (especially container classes). 03/11/2003: luigi Support for std::map. (a) Integration with scripting language (a la std::vector) for Python, Ruby, MzScheme, and Guile; (b) Simple wrapper for other languages 03/10/2003: beazley Fixed problem with escape sequences in string and character constants. SWIG wasn't parsing certain octal codes correctly. 03/07/2003: beazley Fixed a variety of subtle preprocessor problems reported by Sebastien Recio. (a) Empty preprocessor values no longer generate "bad constant value" errors. For example: #define FOO #define FOO BAR (b) Macro names can now span multiple lines (technically valid, although questionable practice). For example: #define A_LONG_MACRO_\ NAME 42 (c) Whitespace is no longer required before certain macro values. For example: #define FOO"Hello" #define BAR\ "Hello" 03/07/2003: ljohnson (Lyle Johnson) [Ruby] Added missing long long and unsigned long long typemaps in the Lib/ruby/typemaps.i library file. 03/07/2003: mrose (Mark Rose) Added Examples/python/callback to demostrate how directors can be used to implement callbacks in Python Added Examples/python/extend to demonstrate virtual method calls from C++ to Python (really the same as the callback example, just a different context). Added four tests for very basic director functionality. These have runtime tests under python. The Python module now emits #define SWIG_DIRECTORS near the top of the output file if directors are enabled. This is useful for disabling parts of tests in target languages that don't support directors. 03/06/2003: mrose (Mark Rose) Added a section to Doc/Manual/Python.html on cross language polymorphism (directors). 03/06/2003: mrose (Mark Rose) The short-lived "-fdirectors" command line option has been removed. To enable directors, instead use the extended %module directive as follows: %module(directors="1") modulename 03/06/2003: cheetah (William Fulton) The long long typemaps have been rewritten so that they can be more easily used with non ISO compilers, like Visual C++. For example if you are wrapping the Windows 64 bit type __int64 the long long typemaps can be used with %apply: %apply long long { __int64 }; __int64 value1(__int64 x); __int64 will now appear in the generated code instead of long long. 03/06/2003: beazley *** DEVELOPER CHANGE *** Swig module mutation has been changed slightly. When a language class method wants to save node attributes, it now uses one of the following functions: Swig_require() Swig_save() The first argument to these functions is a namespace in which saved attributes are placed. For example,this code Node *n; Swig_save("cDeclaration",n,"type","parms","name",NIL); saves the attributes as "cDeclaration:type", "cDeclaration:parms", and so forth. If necessary, a language module can refer to old values by using this special namespace qualifier. In addition to this, a special attribute "view" contains the name of the last namespace used to save attributes. In the above example, "view" would have the value "cDeclaration". The value of "cDeclaration:view" would have the previous view and so forth. Swig_restore(n) restores a node to the state before the last Swig_require() or Swig_save() call. Note: This change makes it easier for language modules to refer to old values of attributes. 03/06/2003: mrose (Mark Rose) Merged the cross-language polymorphism patch. When enabled, C++ "proxy" classes (called directors) are generated for each specified C++ class. Directors pass method calls from C++ to Python, similar to the way the usual proxy (shadow) classes pass method calls from Python to C++. Together, these two types of proxies allow C++ classes that are extended in Python to behave just like ordinary C++ classes and be used in C++ like native objects. This feature is still very experimental and is disabled by default. To enable director support, specify '-fdirectors' on the SWIG command line or in the SWIG_FEATURES environment variable. In the interface file, add %feature("director") to generate directors for all classes that have virtual methods. See http://stm.lbl.gov/~tm2/swig/ProxyDoc.html for more details. 03/03/2003: beazley Fixed a small glitch in typemap local variable replacement. If you had a typemap like this: %typemap(in) type ($1_type temp) { ... temp = ...; ... } and no occurrence of "$1_type" appeared in the body, then the local variable type wouldn't be substituted. 03/03/2003: cheetah (William Fulton) [C#] New version of the CSharp module which is typemap based. It also uses ECMA C# and no longer uses Microsoft Visual C++.NET glue. This means that it will work on non-Windows platforms. Contributed by Neil Cawse. 02/27/2003: beazley Fixed [ 653548 ] error parsing casting operator definition. SWIG now ignores casting operators declared outside of a class. For example: inline A::operator char *() { ... } Bug reported by Martin Casado. 02/27/2003: beazley Added support for anonymous bit-fields. For example: struct Foo { int x : 4; int : 4; int y : 8; }; Anonymous bit-fields are ignored by SWIG. Problem reported by Franz Höpfinger. 02/26/2003: cheetah (William Fulton) [Java] Better typemaps in the Examples/java/typemap example and also fixes subtle bug when using the StringBuffer typemaps more than once. 02/26/2003: beazley Fixed [ 642112 ] Constants char bug. 02/26/2003: beazley Fixed [ 675337 ] Partial template specialization not entirely working. There was a subtle problem related to the naming and ordering of template partial specialization arguments. Matching worked okay, the resulting templates weren't expanded correctly. 02/25/2003: beazley Fixed problem with parsing (and generating code) for references to arrays. For example: int foo(int (&x)[10]); 02/25/2003: beazley Fixed [ 635347 ] Compilation warning from libpy.c. Reported by Daniel L. Rall. 02/25/2003: beazley Fixed a subtle problem with virtual method implementation checking and typedef. typedef int *intptr; struct A { virtual int *foo() = 0; }; struct B : public A { virtual intptr foo() { }; }; SWIG was treating these declarations as different even though they are the same (via typedef). 02/25/2003: ljohnson (Lyle Johnson) [Ruby] Added range checking for the NUM2USHRT macro, per [ 675353 ]. 02/24/2003: beazley Fixed a subtle problem with the code that determined if a class is abstract and can be instantiated. If you had classes like this: struct A { virtual int foo(int) = 0; }; struct B : virtual A { virtual int foo(int); }; struct C : virtual A { }; /* Note order of base classes */ struct D : B, C { }; /* Ok */ struct E : C, B { }; /* Broken */ then SWIG determined that it could instantiate D(), but not E(). This inconsistency arose from the depth-first search of the inheritance hierarchy to locate the implementations of virtual methods. This problem should now be fixed---SWIG will attempt to locate any valid implementation of a virtual method by traversing over the entire hierarchy. 02/22/2003: cheetah (William Fulton) [Java] Fix for using enum typemaps. The Java final static variable type can be set using the jstype typemap, enabling enums to be mapped to something other than int. Bug reported by Heiner Petith. 02/21/2003: songyanf (Tiger) Added CSharp (C#) module prototype i.e. csharp.cxx & csharp.h at Source/Modules/. They are for test usage only now and need improvement. The interface also need to be modified. *** NEW FEATURE *** 02/20/2003: songyanf (Tiger) Fixed problem with typedef with -fvirtual. Similar as beazley's modification today. 02/20/2003: beazley Added support for gcc-style variadic preprocessor macros. Patch [ 623258 ] GCC-style vararg macro support. Contributed by Joe Mason. 02/20/2003: beazley Fixed [ 605162 ] Typemap local variables. Reported by Lyle Johnson. 02/20/2003: beazley Fixed problem with abstract classes and typedef. For example: class Foo { public: virtual void foo(int x) = 0; }; typedef int Integer; class Bar : public Foo { public: virtual void foo(Integer x); }; SWIG was getting confused about the latter method---making Bar abstract. Reported by Marcelo Matus. 02/19/2003: cheetah (William Fulton) [Java] %javaconst(flag) can also be used on enums as well as constants. This feature enables true Java compiler constants so that they can be used in Java switch statements. Thanks to Heiner Petith for patches. 02/19/2003: songyanf (Tiger) Modified -fcompact feature to deal with PP lines 02/18/2003: beazley Fixed [ 689040 ] Missing return value in std_vector.i. Reported by Robert H. de Vries. 02/18/2003: beazley Fixed a few evil scoping problems with templates, namespaces, and the %extend directive. Problem reported by Luigi Ballabio. 02/18/2003: cheetah (William Fulton) [Ruby] Improved support for Visual C++ and other native Windows compilers. It is no longer necessary to specify "/EXPORT:Init_", where is the swig module name when linking using these native Windows compilers. 02/15/2003: songyanf (Tiger) Added -fvirtual option. Reduce the lines and size of the wrapper file by omitting redifined virtual function in children classes. Modified -compact option to -fcompact option Added -small option. -small = -fvirtual -fcompact And it can be extended by future feature options, which are used to reduce wrapper file szie. Added SWIG_FEATURES environment variable check. To dynamically set the feature options such as -fcompact & -fvirtual *** NEW FEATURE *** 02/13/2003: lenz Updated Doc/Manual/Perl5.html to talk about C++ compile problems configure.in now checks for PERL5_CCFLAGS Runtime/Makefile.in and Example/Makefile.in now use PERL5_CCFLAGS Added Lib/perl5/noembed.h which contains all the known macro conflicts 02/12/2003: beazley Fixed [ 685410 ] C++ Explicit template instantiation causes SWIG to exit. Fixes a syntax error with declarations similar to this: template class std::vector; SWIG now ignores the instantiation and generates a warning message. We might do more later. Reported by Thomas Williamson. 02/11/2003: cheetah (William Fulton) Rewrote bool typemaps to remove performance warning for compiling generated code under Visual C++. 02/11/2003: cheetah (William Fulton) Fix for wrapping reference variables (const non-primitive and all non-const types) for example: int& i; Class& c; const Class& c; 02/11/2003: beazley Fixed more very subtle preprocessor corner cases related to recursive macro expansion. For example: #define cat(x,y) x ## y cat(cat(1,2),3) // Produces: cat(1,2)3 #define xcat(x,y) cat(x,y) xcat(xcat(1,2),3) // Produces 123 See K&R, 2nd Ed. p. 231. 02/10/2003: cheetah (William Fulton) Fixed [ 683882 ] - patch submitted by F. Postma for SWIG to compile on HP-UX. 02/10/2003: beazley Fixed subtle preprocessor argument expansion bug. Reported by Marcelo Matus. 02/10/2003: songyanf Added -compact option. Reduce the lines and size of the wrapper file by omitting comments and combining short lines. *** NEW FEATURE *** 02/07/2003: beazley Fixed [ 651355 ] Syntax error with cstring.i Reported by Omri Barel. 02/07/2003: beazley Fixed [ 663632 ] incompatibility with standard cpp. This is a refinement that fixes this problem: // Some macro with an argument #define FOO(x) x int FOO; /* Not a macro---no arguments */ 02/05/2003: beazley Fixed [ 675491 ] parse error with global namespace qualification. Submitted by Jeremy Yallop. 02/04/2003: beazley Fixed bug in varargs processing introduced by the numinputs typemap parameter. 01/08/2003: ttn [xml] Fix string-replacement ordering buglet. Thanks to Gary Herron. 12/23/2002: cheetah (William Fulton) Further build changes: - The SWIG executable is now built using a single Makefile. - This makefile is generated by Automake (Source/Makefile.am). - Dependency tracking and tags support are in this makefile. - Automake 1.7.2 and Autoconf 2.54 minimum versions are needed to build SWIG from CVS. - Running ./autogen.sh now installs Autoconf/Automake support files into Tools/config and these files are no longer stored in CVS. - Bug fixes in 'make install' for systems using .exe executable extension and ./configure --with-release-suffix=whatever 12/16/2002: cheetah (William Fulton) More build changes: - Autoconf's AC_CANONICAL_HOST replaces proprietary approach for detecting build host. - Autoconf support files moved to Tools/config. 12/16/2002: cheetah (William Fulton) Modifications to run on MacOS, submitted by Bernard Desgraupes. Mainly ensuring generated files are output in the appropriate directory for some modules. 12/11/2002: cheetah (William Fulton) Various build modifications and bug fixes: - Simplification of version string. Use autoconf's PACKAGE_VERSION instead. - Build time removed from SWIG version. - Using standard autoconf config header generation. - Updated old autoconf macros as reported by autoupdate. - Removed $prefix in autoconf from search paths as autoconf won't expand them. - Subtle bug fix where 'make prefix=/somewhere; make clean; make prefix=/somwhere/else' produced an executable using the incorrect library directories. - Added -ldflags commandline option for MzScheme, Ocaml, Pike and PHP. - Fixed reporting of compiler used when using -version commandline option. - SWIG web address added to -version commandline option. 12/11/2002: beazley Minor fix to Tcl dynamic cast typemaps. Reported by Kristopher Blom. 12/10/2002: beazley Fixed subtle template argument replace bug. Reported by Chris Flatley. 12/10/2002: beazley Reverted CHANGES 09/03/2002, preprocessor argument evaluation. Arguments are not evaluated during collection, K&R, p. 230. 12/06/2002: beazley Fixed [ 649022 ] Compilation problems with KAI/KCC 12/02/2002: beazley SWIG 'rel-1-3' CVS branch merged back into the main branch. Version 1.3.17 (November 22, 2002) ================================== 11/19/2002: beazley Fixed [ 613922 ] preprocessor errors with HAVE_LONG_LONG. 11/19/2002: beazley Fixed [ 615480 ] mzscheme SWIG_MustGetPtr_. 11/19/2002: beazley Fixed [ 635119 ] SWIG_croak causes compiler warning. 11/16/2002: cheetah (William Fulton) [Java] Added typemaps for pointers to class members. 11/15/2002: cheetah (William Fulton) [Java] Bug fix: Overloaded C++ functions which cannot be overloaded in Java once again issue a warning. 11/14/2002: cheetah (William Fulton) [Java] Handling of NULL pointers is improved. A java null object will now be translated to and from a NULL C/C++ pointer by default. Previously when wrapping: class SomeClass {...}; void foo(SomeClass *s); and it was called from Java with null: modulename.foo(null) a Java NullPointerException was thrown. Extra typemaps had to be written in order to obtain a NULL pointer to pass to functions like this one. Now the default wrapping will detect 'null' and translate it into a NULL pointer. Also if a function returns a NULL pointer, eg: SomeClass *bar() { return NULL; } Then this used to be wrapped with a SomeClass proxy class holding a NULL pointer. Now null is returned instead. These changes are subtle but useful. The original behaviour can be obtained by using the original typemaps: %typemap(javaout) SWIGTYPE { return new $&javaclassname($jnicall, true); } %typemap(javaout) SWIGTYPE *, SWIGTYPE &, SWIGTYPE [] { return new $javaclassname($jnicall, $owner); } %typemap(javagetcptr) SWIGTYPE, SWIGTYPE *, SWIGTYPE &, SWIGTYPE [] %{ protected static long getCPtr($javaclassname obj) { return obj.swigCPtr; } %} *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 11/12/2002: beazley Fixed problem with abstract methods and signatures. For example: class abstract_foo { public: virtual int meth(int meth_param) = 0; }; class abstract_bar : public abstract_foo { public: int meth(int meth_param_1, int meth_param_2) { return 0; } }; In this case, abstract_bar is still abstract. Fixes [ 628438 ] Derived abstract class not abstract. Reported and patched by Scott Michel. 11/11/2002: beazley Fixed a matching problem with typemaps and array dimensions. For example, if you had this: typedef char blah[20]; and a typemap: %typemap() char [ANY] { ... $1_dim0 ... } then $1_dim* variables weren't be expanded properly. It should work now. Problem reported by Pankaj Kumar Goel. 11/07/2002: mkoeppe Added an experimental new module that dumps SWIG's parse tree as (Common) Lisp s-expressions. The module is invoked with SWIG's -sexp command-line switch. The output can be read into Common Lisp. There is (prototype) example Lisp code that generates Foreign Function Interface definitions for use with Kevin Rosenberg's UFFI. *** EXPERIMENTAL NEW FEATURE *** 11/07/2002: mkoeppe Removed duplicate declaration of "cpp_template_decl" in parser.y; bison 1.75 complained. 11/06/2002: cheetah (William Fulton) [Java] Default primitive array handling has changed like arrays of classes. C primitive arrays are no longer wrapped by a Java array but with a pointer (type wrapper class). Again the changes have been made for efficiency reasons. The original typemaps have been moved into arrays_java.i, so the original behaviour can be obtained merely including this file: %include "arrays_java.i" The array support functions are no longer generated by default. They are only generated when including this file, thus this often unused code is only generated when specifically requiring this type of array support. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 11/05/2002: ljohnson (Lyle Johnson) [Ruby] Added support for nested module declarations (as was previously added for the Perl module). So a %module directive of the form: %module "Outer::Inner::Foo" will nest everything as (in Ruby code): module Outer module Inner module Foo # stuff goes here end end end 11/05/2002: mkoeppe [MzScheme] Add an argument (-declaremodule) that generates code to correctly declare a primitive module extension. Patch submitted by Bruce Butterfield. 11/02/2002: cheetah (William Fulton) [Java] Added patch submitted by Michael Cahill to remove unused parameter warnings for the jenv and cls parameters. This patch also also allows one to use "void" in the jni typemap for any type without code being generated attempting to return a value. 10/29/2002: cheetah (William Fulton) [Java] Array handling is different. Arrays of classes are no longer wrapped with proxy arrays, eg wrapping class X {...}; X foo[10]; used to be wrapped with these Java getters and setters: public static void setFoo(X[] value) {...} public static X[] getFoo() {...} This approach is very inefficient as the entire array is copied numerous times on each invocation of the getter or setter. These arrays are now wrapped with a pointer so it is only possible to access the first array element using a proxy class: public static void setFoo(X value) {...} public static X getFoo() {...} Arrays of enums have also been similarly changed. This behaviour is now like the other SWIG language's implementation and the array library should be used to access the other elements. The original behaviour can be achieved using the macros and typemaps in arrays_java.i, for example: %include "arrays_java.i" JAVA_ARRAYSOFCLASSES(X) class X {...}; X foo[10]; *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 10/29/2002: cheetah (William Fulton) [Java] Two new typemaps javain and javaout for generating the proxy class and type wrapper class method calls to the JNI class. The new typemaps are really used for transforming the jstype (used in proxy class and type wrapper classes) to the jtype (used in the JNI class) and visa versa. A javain typemap is required whenever an in typemap is written and similarly javaout for an out typemap. An example is probably best to show them working: %typemap(javain) Class "Class.getCPtr($javainput)" %typemap(javain) unsigned short "$javainput" %typemap(javaout) Class * { return new Class($jnicall, $owner); } %inline %{ class Class {}; Class * bar(Class cls, unsigned short ush) { return new Class(); }; %} The generated proxy code is then: public static Class bar(Class cls, int ush) { return new Class(exampleJNI.bar(Class.getCPtr(cls), ush), false); } Some new special variables have been introduced in order to use these typemaps. Here $javainput has been replaced by 'cls' and 'ush'. $jnicall has been replaced by the native method call, 'exampleJNI.bar(...)' and $owner has been replaced by 'false'. $javainput is analogous to the $input special variable. It is replaced by the parameter name. $jnicall is analogous to $action in %exception. It is replaced by the call to the native method in the JNI class. $owner is replaced by either true if %newobject has been used otherwise false. The java.swg file contains default javain and javout typemaps which will produce the same code as previously. This change is only of concern to those who have written their own typemaps as you will then most likely have to write your own javain and javaout typemaps. The javaout typemap also makes it possible to use a Java downcast to be used on abstract proxy base classes. See the Java documentation on dynamic_cast. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 10/24/2002: ttn [Methodology] Upgaded to libtool 1.4.3, presumably w/ better support for newish platforms (like MacOS X). 10/21/2002: ttn Fixed Runtime/Makefile.in bug -- thanks to Richard Calmbach. 10/18/2002: ttn Fixed typo in doh.h -- thanks to Max Horn. Version 1.3.16 (October 14, 2002) ================================= 10/13/2002: beazley Fixed bug with %extend directive and %feature reported by William Fulton. 10/13/2002: beazley Added OpenVMS build directory (vms). Contributed by Jean-François Pieronne. 10/09/2002: cheetah (William Fulton) [Java] Added throws clause to the native functions in the JNI class. The throws clause is the same as the one generated for proxy functions and module class functions. 09/27/2002: beazley Fixed some problems with the %import directive and classes that were defined but not wrapped. Problem reported by Leslie Brooks, Gerry Woods, and others. 09/23/2002: cheetah (William Fulton) [Java] Some error checking added: 1) OutOfMemoryException check in the char * typemaps. 2) As SWIG treats pointers, references and passing by value all the same, it is possible to pass a NULL pointer to a function that expects an object passed by value or by reference. A NullPointerException is now thrown under this scenario. 09/20/2002: ttn [Methodology] Reworked "make clean" and "make install" to be more table driven. [Docs] Explain how to extend "make install" w/ extra-install.list. 09/15/2002: beazley Deprecation of the "ignore" typemap. The "ignore" typemap has been deprecated in favor of a generalization of the "in" typemap. To ignore an argument, use something like this instead: %typemap(in,numinputs=0) int *output (int temp) { $1 = &temp; } This change fixes a number of subtle bugs related to the interaction of the "in" and "ignore" typemaps (which were supposed to be mutually exclusive). The use of the numinputs argument is reserved for future expansion. Currently, values >1 will generate an error. However, future releases of SWIG may utilize that to support multi-input typemaps. %typemap(ignore) still works, but generates a warning message and is translated to %typemap(in,numinputs=0). *** POTENTIAL INCOMPATIBILITY *** *** NEW FEATURE *** 09/15/2002: beazley Fixed segmentation fault for unnamed structures. For example: typedef struct { } *blah; Reported by Roger Gibson. Note: we might be able to generate wrappers in special cases. 09/13/2002: beazley Minor modification to generated wrapper functions. Pointer arguments are now always set to an initial value of 0. Simplifies typemap writing and cleanup code (since you can rely on zero-value initialization). This also greatly reduces the need to ever write an "arginit" typemap. 09/12/2002: beazley Minor enhancement to smart-pointer support. If operator->() is part of an ignored base class like this, %ignore Bar; class Foo { public: int blah(); }; class Bar { /* Ignored */ public: ... Foo *operator->(); ... }; class Spam : public Bar { }; then methods from Foo are still available. For example, >>> s = Spam() >>> s.blah() 0 >>> The only catch is that the operator->() itself is not available (since it wasn't wrapped). Therefore, there won't be any __deref__() operation unless it is explicitly added to Spam (either using %extend or just placing operator->() in the definition of Spam). 09/11/2002: ttn [Methodology] Reworked "make check" to be more table driven. [Docs] Docuemented methodology in Manual/Extending.html. 09/11/2002: ttn [Docs] Prefixed Manual/*.html with "" to pander dotingly to (over-)sensitive editors. 09/10/2002: ttn [Guile] Converted Examples/guile/simple "make check" behavior to actually check execution results. Reduced iteration counts so that the test doesn't take too long. 09/10/2002: beazley SWIG-1.3.15 released. Version 1.3.15 (September 9, 2002) ================================== 09/09/2002: beazley Fixed nasty runtime type checking bug with subtypes and inheritance and templates. 09/09/2002: cheetah (William Fulton) [Java] Java exception classes for a method's throws clause can be generated by specifying them in a comma separated list in the throws attribute in any one of the following typemaps: in, out, check, freearg, argout and throws. A classic example would be to convert C++ exceptions into a standard Java exception: %typemap(throws, throws="java.io.IOException") file_exception { jclass excep = jenv->FindClass("java/io/IOException"); if (excep) jenv->ThrowNew(excep, _e.what()); return $null; // or use SWIG_fail } class file_exception {...}; void open(const char *filename) throw(file_exception); The Java method will then be declared with a throws clause: public static void open(String filename) throws java.io.IOException {...} 09/08/2002: mkoeppe * [Guile] Improved the documentation system. The arglist no longer gets cluttered with type specification, making it more readable. (Also the ILISP function C-u M-x `arglist-lisp' RET works better this way.) The types of arguments are explained in an extra sentence after the arglist. There are now two documentation-related typemap arguments: %typemap(in, doc="$NAME is a vector of integers", arglist="$name") int *VECTOR { ... } The "arglist" texts of all arguments of a function make up its arglist in the documentation. The "doc" texts of all arguments are collected to make a sentence that describes the types of the arguments. Reasonable defaults are provided. As usual, $name is substituted by the name of the argument. The new typemap variable $NAME is like $name, but marked-up as a variable. This means that it is upper-cased; in TeXinfo mode ("-procdocformat texinfo") it comes out as @var{name}. The directives %values_as_list, %values_as_vector, %multiple_values now also have an effect on the documentation. (This is achieved via the new pragmas return_nothing_doc, return_one_doc, return_multi_doc.) Documentation has also improved for variables that are wrapped as procedures-with-setters (command-line switch "-emit-setters"). * [Guile] Emit constants as _immutable_ variables. (This was broken recently.) 09/07/2002: mkoeppe [Guile] Updated the typemaps in list-vector.i. 09/07/2002: mkoeppe Short-circuit the typechecks for overloaded functions. (The changes in code generation are visible in the new testcase "overload_complicated".) 09/06/2002: cheetah (William Fulton) [Java] Solution for [ 596413 ] New typemap so that the Java proxy classes and type wrapper classes wrapper constructor modifier can be tailored by users. The default value is protected. Normally SWIG generates a constructor like this which can only be accessed within one package: protected Bar(long cPtr, boolean cMemoryOwn) { ... } If you are using SWIG across multiple packages or want to use this constructor anyway, it can now be accessed outside the package. To modify use for example: %typemap(javaptrconstructormodifiers) SWIGTYPE "public" to change to public for all proxy classes and similarly for all type wrapper classes: %typemap(javaptrconstructormodifiers) SWIGTYPE, SWIGTYPE *, SWIGTYPE &, SWIGTYPE [] "public" 09/06/2002: cheetah (William Fulton) [Java] Added throws typemaps for the Java module. C++ exceptions get converted into java.lang.RuntimeException Java exceptions. Warning: This may change from java.lang.Runtime exception in the future. 09/05/2002: cheetah (William Fulton) [Java] Fix for variables declared as references. 09/05/2002: beazley Fixed [ 605162 ] Typemap local variables. Reported by Lyle Johnson. 09/05/2002: ljohnson (Lyle Johnson) [Ruby] More updates to the Ruby module documentation, including a new typemap example that demonstrates how to collect key-value pairs from an argument list into a Hash. 09/05/2002: beazley Fixed bug with template expansion and constructors. template class Foo { public: Foo() { } }; The extra in the constructor was carried through in the name--causing runtime problems in generated modules. Reported by Jordi Arnabat Benedicto. 09/05/2002: mkoeppe [Guile] Support overloading. 09/04/2002: ljohnson (Lyle Johnson) [Ruby] Updated typemaps for long long and unsigned long long types to use Ruby 1.7 support for these types when available. 09/04/2002: ljohnson (Lyle Johnson) [Ruby] Added output typemaps for const reference to primitive types. 09/04/2002: mkoeppe [Guile] Fix pass-by-value typemaps. Reported by Arno Peters via Debian bugtracking (#156902), patch by Torsten Landschoff . 09/03/2002: samjam (Sam Liddicott) Better reference support. Functions that want a void** can take a NULL by reference and the void* will be made for you and then passed-by-reference Also all integer-class native types can be passed by reference where an int* or int& etc is needed 09/03/2002: beazley Changed the evaluation order of preprocessor macro arguments. Arguments are now expanded by the preprocessor *before* they are passed to macro expansion. This fixes a subtle expansion bug reported by Anthony Heading. 09/03/2002: beazley Fixed the file include order (again, apparently). See 2/27/99. 09/02/2002: beazley [Perl] Better exception handling support. Since Perl error handling relies on setjmp/longjmp, wrapper functions have been modified slightly to provide an extra block scope: XS(foo) { char _swigmsg[SWIG_MAX_ERRMSG] = ""; const char *_swigerr = _swigmsg; { /* Normal wrapper function here */ ... SWIG_croak("An error occurred\n"); ... XSRETURN(argvi); /* Successful return */ fail: /* cleanup code */ } croak(_swig_err); } The macro SWIG_croak(x) sets the value of _swigerr to x and executes a "goto fail". The whole wrapper function is enclosed block scope to provide proper cleanup of C++ objects. Since croak executes a longjmp(), there is no way to properly reclaim resources if this executes in the same scope as the wrapper function. The _swigmsg[] variable is normally unused, but can be used to store small error messages using sprintf or snprintf. It has a capacity of at least 256 bytes (SWIG_MAX_ERRMSG). 09/02/2002: beazley [Tcl] Added better support for exceptions. Instead of returning TCL_ERROR, use the macro SWIG_fail to return with an error. This ensures that arguments are properly cleaned up. Exception specifiers are now handled by default. 09/02/2002: ljohnson (Lyle Johnson) [Ruby] The type-checking system for the Ruby module has had a flaw in that some types which should be considered equivalent weren't. This bug was best demonstrated by the inherit_missing.i test suite case, which defines a base class "Foo" that is subclassed by "Bar". The "Foo" class isn't actually wrapped (i.e. it's not directly accessible from Ruby) but we'd still like to be able to pass "Bar" instances to functions expecting Foos and have that work; it wasn't. The revised implementation (similar to that used for some other language modules) adds a new instance variable (__swigtype__) to each object that indicates its SWIG type; that is, each "Bar" instance will now have a string instance variable called "__swigtype__" whose value is "_p_Bar". Unless developers were taking advantage of this low-level implementation detail, they shouldn't notice any compatibility problems; nevertheless, I'm marking it as a "potential incompatibility". *** POTENTIAL INCOMPATIBILITY *** 09/01/2002: ljohnson (Lyle Johnson) [Ruby] Fixed SF Bug #603199. 08/08/2002: cheetah (William Fulton) [Java] Added OUTPUT, INPUT and INOUT typemaps in typemaps.i for C++ references. 08/27/2002: mkoeppe [Guile] Fixed error in "lib_std_vector" testcase and compiler warning in "lib_cdata" testcase. 08/27/2002: ljohnson (Lyle Johnson) [Ruby] Added the "%mixin" directive, which allows the user to specify a comma-separated list of module names to mix-in to a class. So, for example, if you'd like to specify that Ruby's Enumerable module should be mixed-in to your class Foo, you'd write: %mixin Foo "Enumerable"; or to specify that the modules Fee, Fie and Fo should be mixed in to Foo: %mixin Foo "Fee,Fie,Fo"; *** NEW FEATURE *** 08/27/2002: ljohnson (Lyle Johnson) [Ruby] Modified the %alias directive so that multiple aliases can be specified for an instance method by using a comma-separated list of aliases. 08/27/2002: ljohnson (Lyle Johnson) [Ruby] Added "throw" typemaps for the Ruby module. 08/26/2002: beazley Two new command line options for printing dependencies. 'swig -M' lists all file dependencies. 'swig -MM' lists dependencies, but excludes files in the SWIG library. Example: % swig -M -python example.i example_wrap.cxx: \ /u0/beazley/Projects/lib/swig1.3/swig.swg \ /u0/beazley/Projects/lib/swig1.3/python/python.swg \ example.i \ example.h % swig -MM -python example.i example_wrap.cxx: \ example.i \ example.h *** NEW FEATURE *** 08/26/2002: beazley Fixed [ 597599 ] union in class: incorrect scope. Reported by Art Yerkes. 08/26/2002: beazley Fixed [ 600132 ] Default argument with namespace. Reported by Shibukawa Yoshiki. 08/24/2002: beazley Automatic C++ exception handling enabled for all language modules. This is pretty simple. If you have a class like this: class Foo { }; class Bar { public: void blah() throw(Foo); } then the generated wrapper code looks like this: wrap_Bar_blah() { ... try { arg1->blah(); } catch (Foo &_e) { /* "throw" typemap code inserted. $1 = _e */ } catch (...) { throw; } } The "throw" typemap can be used to raise an error in the target language. It can do anything. Here is a very simple example: %typemap("throw") Foo { PyErr_SetString(PyExc_RuntimeError, "Foo exception"); return NULL; } To make this work in each language module, simply define a few default "throw" typemaps for SWIGTYPE, SWIGTYPE *, int, const char *, and a few common exception types. That's all there is to it. Automatic exception handling can be disabled using -noexcept or setting the NoExcept global variable to 1. *** NEW FEATURE *** 08/23/2002: beazley [ Python ] Automatic translation of C++ exception specifications into error handling code. For example: class Foo { }; class Bar { public: void blah() throw(Foo); } In this case, Foo is wrapped as a classic-style class (compatible with exception handling). Furthermore, you can write Python code like this: b = Bar() try: b.blah(); except Foo,e: # Note use of exception class here! # Handle Foo error ... The object "e" in the exception handler is just a wrapped Foo object. Access it like a normal object. If an exception is not wrapped as a class, a RuntimeError exception is raised. The argument to this exception is the exception object. For example: class Bar { public: void blah() throw(int); } b = Bar() try: b.blah(); except RuntimeError,e: print e.args[0] # Integer exception value Comments: - If a class is used as an exception, it *must* be wrapped as a Python classic-style class (new classes don't work). - Automatic exception handling is compatible with %exception. - Use -noexcept to turn off this feature. - The newly introduced "throw" typemap is used to raise Python errors (naturally). *** EXPERIMENTAL NEW FEATURE *** 08/23/2002: beazley Information from throw() specifiers is now stored in the parse tree. For example: class Foo { public: int blah() throw(spam,bar); } The stored information is fully corrected for namespaces and works with templates. Uses will follow. 08/22/2002: beazley Exception handling code is no longer applied to member access function. For example, in this code %exception { try { $action } catch(whatever) { ... } } class Foo { public: int x; ... } The exception handling code is not applied to accessor functions for Foo::x. This should reduce the amount of extra code generated. Caveat: Exception handling code *is* used when attributes are accessed through a smart-pointer or a synthesized attributed added with %extend is used. 08/22/2002: beazley Made more patches to hopefully eliminate problems when compiling SWIG as a 64-bit executable. 08/22/2002: beazley Fixed a bug with const reference members, variables, and static members. For example: class Foo { public: static const int &ref; }; SWIG was trying to generate "set" functions which wouldn't compile. 08/21/2002: beazley Made the warning message for "Class X might abstract" off by default. Enable with -Wall. 08/21/2002: beazley Refined handling of const and non-const overloaded methods. If a class defines a method like this: class Foo { public: int bar(int); int bar(int) const; } Then the non-const method is *always* selected in overloading and the const method silently discarded. If running with -Wall, a warning message will be generated. 08/19/2002: beazley Better support for using declarations and inheritance. Consider this: class Foo { public: int blah(int x); }; class Bar { public: double blah(double x); }; class FooBar : public Foo, public Bar { public: char *blah(char *x); using Foo::blah; using Bar::blah; }; Now SWIG wraps FooBar::blah as an overloaded method that uses all accessible versions of blah(). See section 15.2.2 in Stroustrup, 3rd Ed. SWIG also supports access change through using declarations. For example: class Foo { protected: int x; int blah(int x); }; class Bar : public Foo { public: using Foo::x; using Foo::blah; }; Caveat: SWIG does not actually check to see if declarations imported via 'using' are in the inheritance hierarchy. If this occurs, the wrapper code won't compile anyways---not sure it's worth worrying about. 08/18/2002: beazley Modified overloading dispatch to not include nodes with an "error" attribute. A language module can set this if a node couldn't be wrapped and you don't want it included in the dispatch function. 08/18/2002: beazley Enhancement to overloaded function dispatch. The dispatcher is now aware of inheritance relationships. For example: class Foo { }; class Bar : public Foo { }; void spam(Foo *f); void spam(Bar *b); In this case, the dispatcher re-orders the functions so that spam(Bar *b) is checked first---it is more specific than spam(Foo *f). 08/17/2002: beazley Added -Werror command line option. If supplied, warning messages are treated as errors and SWIG will return a non-zero exit code. 08/17/2002: beazley Fixed [ 596135 ] Typedef of reference can't compile. For example: typedef int &IntRef; void foo(IntRef i); SWIG-1.3.14 generated code that wouldn't compile. Version 1.3.14 (August 12, 2002) ================================ 08/11/2002: mmatus Static const members initialized during declaration, and only them, ie: struct A { static const int a = 1 ; // this one static const int b; // not this one }; are emitted like constants (equivalent to enums or explicit %constant). This is because they cannot be added directly to 'cvar' since they lack the needed reference (well, you can force them to have a real reference, but in an ugly way which goes completely again the original purpose of initialize them during declaration, you also have to deal with extra linking matters, and it take a while to figure out what is the problem and how to solve it). Please test it with your preferred target language, and not only the code generation, but really run the example in the test-suite (static-const-member-2.i) because the problem and the solution cannot be "fully" appreciated until you try to load the module and run it. In some target languages (python specially), this can produces a difference in the way that the static constant members 'a' and 'b' are internally wrapped. Hopefully, they still can be accessed in the same way. 08/11/2002: mmatus [python] Now static const members can be accessed in a more natural way, ie, if you have struct A { typedef unsigned int viewflags; static const viewflags forward_field = 0; static const viewflags backward_field; }; now you can do: print A.backward_field and also a = A() print a.forward_field Note that if the static const members don't have an initializer (like backward_field), still you can access them in the same way in the python side, but the implementation is a quite different: backward_field will still appear in the cvar entity, and also, you are responsible to initialize it in some code unit, and link it properly. forward_field, by the other hand, will not appear in the cvar entity but only as a A member, similar to what happen with enum or %constant members. 08/11/2002: mmatus [python] Common code in the __setattr__/__getattr__ now goes to two "free" methods at the beginning of the proxy file, from where each class use it. This change reduces the size of the proxy file, specially if you wrap a lot of small classes in one module (up to 33% in some cases), making it faster to load too. 08/09/2002: beazley [Perl5] If a function that returns char * returns NULL, undef is returned to the Perl interpreter. 08/09/2002: beazley Fix to conversion operators and namespaces. For example: namespace ns { struct Foo { }; struct Bar { operator Foo*(); }; } In the wrapper code, SWIG was using ->operator Foo*() when it should have been using ->operator ns::Foo*(). Note: if using %rename with a conversion operator, you might have to do this: %rename(toFooPtr) ns::operator ns::Foo*(); // ^^^^ note extra qualifier namespace ns { ... 08/09/2002: beazley [Python] Minor enhancement to 'const' variable declarations. Normally const declarations are wrapped as read-only variables accessible only through the cvar attribute (see SWIG.html for a discussion of why). However, in many programs, "const" declarations may just be constants---making the cvar. access awkward. To fix this, "const" declarations are now available both through cvar. and as a simple name. For example: const int FOO = 42; In Python: >>> print example.cvar.FOO 42 >>> print example.FOO 42 Note: There are cases where the value of a "const" variable might change. For example: char *const BAR = "Hello World"; In this case, the pointer itself can not change, but the data being pointed to could be modified. In these situations, cvar.BAR should be accessed to obtained the current value. 08/08/2002: beazley [Python] Fixed generation of the proxy code (.py files) to more closely follow the order of declarations as they appear in the .i file. In the past, all of the class wrappers appeared first, followed by function stubs, inserted Python code, and other details. 08/08/2002: cheetah (William Fulton) [Java] Proxy method _delete() changed to delete(). There shouldn't ever be a wrapped function called delete() as it is a C++ keyword and there is no such thing as a member function in C. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** Backwards compatibility can be achieved by adding the function back in for all proxy classes: %typemap(javacode) SWIGTYPE %{ public void _delete() { delete(); } %} Java backwards compatibility summary ------------------------------------ There are a number of changes that have been made in improving the Java module for ver 1.3.14. If at all possible change your code to take advantages of the improvements. If you were using proxy classes you may not notice any backwards compatibility issues. Here is an example which will help with most backwards compatibility problems where it is not possible to modify the code that uses the generated output: Replace: %module modulename With: %module (jniclassname="modulename") modulename; %typemap(javacode) SWIGTYPE %{ public long getCPtr$javaclassname() { return swigCPtr; } public void _delete() { delete(); } %} %pragma(java) jniclassclassmodifiers="public"; The proxy constructors that took parameters (long cPtr, boolean cMemoryOwn) were public and are now protected. If you were making use of these then you'll have to modify your code and the best solution would be to use the new type wrapper classes. The other main areas are the pragmas and global variable wrapping. Replace the pragmas with one of the new directives or typemaps mentioned below and use %rename on the variables. If you were not using proxy classes, you will have to define a jstype typemap as well as a jtype typemap. 08/08/2002: cheetah (William Fulton) [Java] Fix for wrapping two dimension array variables. 08/07/2002: beazley [Python,Tcl] Object management now has a much better sense of ownership. Ownership bits is changed whenever an object is stored in a global variable or structure member. For example: struct Foo { int val; Foo *next; }; Now in Python >>> f = Foo() >>> f.thisown 1 >>> g = Foo() >>> g.next = f # Assign a pointer >>> f.thisown # Notice ownership change 0 >>> This scheme is mostly a conservative heuristic designed to provide segmentation faults. It could cause a memory leak if ownership is changed unnecessarily. In this case, you can either write a typemap (that doesn't change ownership), or manually set the thisown attribute back to 1. 08/07/2002: beazley [Tcl] Major usability improvements to the object interface. Suppose you had code like this: struct Foo { int x; int spam(); }; void blah(Foo *f); In past versions of SWIG, you could create objects and use them like this: % Foo f % f configure -x 3 % f spam 37 The only problem is that if you tried to call blah(), it didn't work: % blah f Type Error. Expected _p_Foo % Instead, you had to do this: % blah [f cget -this] SWIG now automatically extracts the -this pointer, avoiding this problem. This means that saying "blah f" is perfectly legal and everything will still work normally. Caveat: Since pointer strings start with a leading underscore (_), don't use this in object names. For example: % Foo _f % blah _f # Potential crash Objects now have a -thisown attribute that shows the ownership. This builds upon the CHANGES 11/24/2001 entry. 08/07/2002: samjam, Sam Liddicott Properly implemented pointer system using php resources. Still need to work out whether or not to let script-users call destructors directly 08/06/2002: beazley Upgraded mzscheme module to support version 201 and added overloading support. 08/05/2002: beazley Added parsing support for extra grouping (in very limited cases). For example: typedef int (FuncPtr)(int, double); *** EXPERIMENTAL *** 08/03/2002: ljohnson (Lyle Johnson) [Ruby] Updates to typemaps.i as those done previously for Perl, Python and Tcl modules. Now supports reference types with INPUT, OUTPUT and INOUT typemaps. 08/02/2002: beazley New library file cstring.i added. Provides macros for manipulating char * data. 08/02/2002: beazley Deprecated the %new directive. Use %newobject instead. For example: %newobject foo; ... char *foo(); %newobject follows the same rules as %rename, %ignore, %feature, etc. *** POTENTIAL INCOMPATIBILITY *** 08/01/2002: cheetah (William Fulton) [Java] New attribute 'jniclassname' for the module directive allows a way of changing the JNI class name from the default which uses the modulename with JNI appended after it. %module (jniclassname="name") modulename If 'name' is the same as 'modulename' then the module class name gets changed from 'modulename' to modulenameModule. 08/01/2002: beazley Fixed problem with file include order. Language specific directories should take precedence over generic directories. For example: "swig_lib/python/foo.i" should be loaded before "swig_lib/foo.i". I thought this was the case already, but apparently it has been broken for quite some time. 08/01/2002: beazley Added std_deque.i library file. Work in progress. 08/01/2002: beazley [Python,Tcl,Perl] Improvements to typemaps.i. INPUT/INOUT typemaps perform better error checking. Typemaps are now supplied for references like int &OUTPUT, double &INOUT, etc. 08/01/2002: beazley [Python] Deprecated the T_* and L_* typemaps in typemaps.i. Multiple return values are always placed in a tuple. Deprecated the BOTH typemaps. This is now INOUT (e.g., int *INOUT). *** POTENTIAL INCOMPATIBILITY FOR PYTHON MODULE *** 08/01/2002: beazley Deprecated the array.i, carray.i, and timer.i library files. 08/01/2002: beazley Deprecated the pointer.i library file. Use cpointer.i instead. *** POTENTIAL INCOMPATIBILITY *** 08/01/2002: cheetah (William Fulton) [Java] For consistency the global variable getters and setters use the JavaBean property design pattern like member variables always have. This means if you are wrapping a variable called foo, the getter is called getFoo() and the setter is called setFoo(). Before the recent changes to the Java module the getters and setters were called get_foo() and set_foo(). If you really want the original function names use the %rename directive like this: %rename(_foo) Foo; 07/31/2002: beazley Fixed casting problem with multiple inheritance. If you had this, class foo {}; class bar : public foo {}; class baz : public foo {}; class spam : public bar, public baz {}; then the wrappers wouldn't compile due to an ambiguous cast. Reported by Art Yerkes. 07/30/2002: cheetah (William Fulton) [Java] Due to new static typechecking all pointers held in a Java long are part of the internal workings and this pointer value in the Java long has become abstracted data. The type wrapper constructor and getCPtr() methods are as such protected. If you need to mess around with pointers from Java or for example create a proxy class or type wrapper class around a null pointer, add a function/constructor to do so with the %javacode typemap. You can also make getCPtr() public again with the %javagetcptr typemap. 07/30/2002: cheetah (William Fulton) [Java] Fixes for %typemap(ignore). In particular when ignoring the last parameter in a function. Also for all parameters in constructors. These mods have also fixed multi-argument typemaps for proxy classes - SF 581791. 07/30/2002: cheetah (William Fulton) [Java] %newobject (replacement for %new) now implemented for Java. 07/29/2002: beazley Fixed problem with typemap copies, %apply, and %clear inside C++ namespaces. 07/28/2002: cheetah (William Fulton) [Java] The JNI class now has package access as the class modifier has been changed from "public" to nothing. This has been done as this class is now more for the internal workings of SWIG since the module class has static type checking for all types. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** Backwards compatibility can be achieved by using the %jniclassclassmodifier pragma to change it back to "public". 07/28/2002: cheetah (William Fulton) [Java] Proxy/Shadow classes are generated by default. The -proxy and -shadow command line options are deprecated. If you want to use the low-level functional interface then use the new -noproxy commandline option. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 07/28/2002: cheetah (William Fulton) [Java] Remaining pragmas shakeup. These were the remaining pragmas and their new names where changed: modulebase modulecode moduleclassmodifiers moduleimport => moduleimports moduleinterface => moduleinterfaces The moduleimports works slightly differently to how the moduleimport pragma worked. Now it actually takes code which gets placed before the class definition so the whole import statement has to be given, for example: %pragma(java) moduleimports=%{ import java.io.*; import java.math.*; %} The moduleinterfaces is slightly different to the old moduleinterface in that if more than one interface is required they must be comma separated in one use of the pragma, for example: %pragma(java) moduleinterfaces="Serializable, MyInterface" These last two pragmas are consistent with the javainterfaces and javaimports typemap. A similar set of pragmas has been introduced, namely: jniclassbase jniclasscode jniclassclassmodifiers jniclassimport jniclassinterface These work in the same way as their module counterparts. Note that previously the moduleXXX pragmas worked on the old module class which is now called the JNI class (the class with the native functions). The jniclassXXX pragmas now work on the new module class (the class that has all the global functions and global variable getters and setters when using proxy classes, plus all other remaining functions when using the low-level procedural interface). In summary the contents of the pragmas make up a class like this: class modulename extends implements { ... SWIG generated functions ... } } *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 07/28/2002: cheetah (William Fulton) [Java] Deprecated modulemethodmodifiers pragma and replaced with a better %feature based directive called %javamethodmodifiers. A useful example would be for synchronisation in multi-threaded apps: %javamethodmodifiers foo(int a) "public synchronized"; Changes this function from the default ("public") to "public synchronized". *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 07/26/2002: beazley Several directives now allow optional configuration parameters. These include: %module(name="value", name="value", ...) modulename %import(name="value", ...) "filename.i" %extend(name="value", ...) classname { ... } These currently have no effect and are reserved for future expansion. 07/26/2002: beazley Enhancements to smart-pointer handling. SWIG only provides extra support for a smart-pointer if operator->() returns a proper pointer. For example: Foo *operator->(); If operator->() returns an object by value or reference, then SWIG examines the returned object to see if it also implements operator->(). If so, SWIG chases operator->() until it can find one that returns a pointer. This allows cases like this to work: class Foo { public: void blah(); }; class Bar { ... Foo *operator->(); ... }; class Spam { ... Bar operator->(); ... }; For example: >>> s = Spam() >>> s.blah() # Invokes Foo::blah() The s.blah() call actually invokes: ((s.operator->()).operator->())->blah(); 07/26/2002: beazley Fixed a bug with typedef and references. For example: typedef Foo & FooRef; FooRef blah(); Previous versions of SWIG generated code that wouldn't compile. 07/25/2002: beazley Wrapping of static methods has been improved in proxy classes. In older versions of SWIG, if you had this: class Foo { public: static void bar(); }; The static method was only available as a function Foo_bar(). For example: >>> Foo_bar() Now, the static method can also be invoked through an instance like this: >>> f = Foo() >>> f.bar() # Invokes static method This works with all versions of Python. Additionally, for Python-2.2, the static method can be invoked as: >>> Foo.bar() The old-style function is still support for backwards compatibility. If you care about making your code across different versions of Python, either use Foo_bar() or access the method through an instance. 07/25/2002: beazley Changes to the Python module. Proxy classes now utilize new Python-2.2 features including properties and static methods. However, these features are supported in a way that provides backwards compatibility with older Python versions. In other words, proxy classes work with all versions of Python and only use new features when running on Python-2.2. 07/25/2002: beazley Modified %extend so that overloaded methods can be added. For example: %extend Foo { void bar(int x) { }; void bar(char *s) { }; ... } This works with both C++ *and* C. 07/24/2002: cheetah (William Fulton) [Java] More new typemaps so that the Java proxy classes and type wrapper classes can be further tailored by users. These are the default code for generating the finalize() methods (proxy classes only) and the getCPtr() methods for proxy classes and type wrapper classes: %typemap(javafinalize) SWIGTYPE %{ protected void finalize() { _delete(); } %} %typemap(javagetcptr) SWIGTYPE, SWIGTYPE *, SWIGTYPE &, SWIGTYPE [] %{ public static long getCPtr($javaclassname obj) { return obj.swigCPtr; } %} The javagetcptr typemap will enable users to handle Java null by overriding this typemap - a requested feature. The -nofinalize commandline option has been deprecated. The javafinalize typemap is more powerful as it will allow the removal of the finalize methods for all or any one or more particular proxy class. 07/23/2002: cheetah (William Fulton) [Java] The getCPtrXXX() function has been changed to a static function and is now of the form: protected static long getCPtr(XXX obj) {...} This is a requested change which will allow Java null pointers to be used as null can be passed in for obj. However, to achieve this the appropriate code must be written using the new javagetcptr typemap directive. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** Backwards compatibility can be achieved by adding this function back in using the new javacode typemap: %typemap(javacode) SWIGTYPE %{ // SWIG-1.3.12 and SWIG-1.3.13 public long getCPtr$javaclassname() { return swigCPtr; } // SWIG-1.3.11 and earlier public long getCPtr() { return swigCPtr; } %} 07/23/2002: cheetah (William Fulton) [Java] New directive to control constant code generation - %javaconst. The default handling for handling constants is to get the value through a JNI call, eg #define YELLOW 5 #define BIG 1234LL results in: public final static int YELLOW = modulename.get_YELLOW(); public final static long BIG = modulename.get_BIG(); Earlier versions of the Java module initialised the value using the C value: public final static int YELLOW = 5; public final static long BIG = 1234LL; This works in most cases, but the value for BIG won't compile as 1234LL is not valid Java code and this is one of the reasons why the default is now to get the values through a JNI call. The side effect is that these 'constants' cannot be used in switch statements. The %javaconst directive allows one to specify the way the constant value is initialised and works like other %feature directives, eg %javaconst(0); // all constants from this point on are initialised using the C value %javaconst(1) BIG; // just BIG initialised using JNI call (must be parsed before BIG is defined) 07/23/2002: beazley *** IMPORTANT CHANGES TO THE PYTHON MODULE *** (1) The Python module now enables shadow/proxy classes by default. This means that two files are always created by SWIG. For instance, if you have this: // file: foo.i %module foo ... Then swig generates two files "foo_wrap.c" and "foo.py". (2) The name of the low-level C extension module has been changed to start with a leading underscore. This means that you have to compile the module as follows: $ cc -c -I/usr/local/include/python2.2 foo_wrap.c $ cc -shared foo_wrap.o $(OBJS) -o _foo.so ^^^^ note extra underscore This naming scheme is consistent with other Python modules that utilize extension code. For instance, the socket module consists of "_socket.so" and "socket.py". In previous versions of SWIG, the shared object file was named "foocmodule.so". (3) A new directive can be used to insert Python code into the corresponding .py file. For example: %pythoncode %{ def foo(): print "Hello World" %} This directive allows you to create modules as a mix of C and Python. Python code is seamlessly added to the module. (4) The -shadow command line option is deprecated. This is turned on by default. (5) To disable the generation of the extra python file, use the "-noproxy" command line option. *** POTENTIAL INCOMPATIBILITY *** This change will likely break the build environment of projects that utilize shadow classes. To fix this, you probably only need to change the name of the target .so file. For example, if you have Makefile information like this: TARGET = examplecmodule.so Just change it to: TARGET = _example.so *** DOCUMENTATION UPDATE *** The file Doc/Manual/Python.html has been updated to describe these changes. 07/23/2002: beazley Added -noextern option. If supplied, SWIG will not generate extra extern declarations. This is sometimes an issue on non-unix platforms. 07/23/2002: beazley Added a warning for ignored friend functions. 07/23/2002: beazley Fixed [ 574498 ] -proxy and %include "pointer.i" clash. Reported by David Creasy. 07/23/2002: beazley Fixed [ 576103 ] global destruction warning with shadow. Perl extensions should no longer report the warning "Use of uninitialized value during global destruction." when running with "perl -w". Reported by Brett Williams. 07/23/2002: beazley In C++ mode, SWIG now always defines namespace std. By default, it's empty. However, this will silence errors from programs that include statements such as "using namespace std;". This fixes Bug [ 584017 ] using namespace std generates error. Reported by Joseph Winston. 07/22/2002: beazley Added a new warning message for %apply. If you use %apply but no typemaps are defined, you will get a warning message. This should help with problems like this: %apply char *OUTPUT { ... }; In old versions of SWIG, this silently did nothing. Now you get an error like this: file:line. Warning. Can't apply (char *OUTPUT). No typemaps are defined. 07/22/2002: cheetah (William Fulton) [Java] Started Java pragma deprecation. Replacements use %typemap based directives and enable proxy classes and the new type wrapper classes to be tailored in various ways. These are the new typemaps: %typemap(javabase) - base (extends) for Java class %typemap(javaclassmodifiers) - class modifiers for the Java class: default is "public" %typemap(javacode) - java code is copied verbatim to the Java class %typemap(javaimports) - import statements for Java class %typemap(javainterfaces) - interfaces (extends) for Java class And these are the %pragma directives being deprecated: allshadowbase allshadowclassmodifiers allshadowcode allshadowimport allshadowinterface shadowbase shadowclassmodifiers shadowcode shadowimport shadowinterface Note that it is possible to target a particular proxy class: %typemap(javaimports) Foo "import java.util.*"; or a particular type wrapper class: %typemap(javaimports) double* "import java.math.*"; Note that $javaclassname in these typemaps are substituted with either the proxy classname when using proxy classes or the SWIGTYPE class name. 07/18/2002: cheetah (William Fulton) [Java] Java module overhaul to implement static type checking of all types. 1) Changes when using Java Proxy classes ---------------------------------------- Previously when wrapping global functions: class SomeClass{}; void foo(SomeClass* s); SomeClass* bar(); The native method prototypes used a long for pointers and looked like this: public class modulename { ... public final static native void foo(long jarg1); public final static native long bar(); } and unlike member functions of a C++ class there was no wrapper around the native calls to make the use of them more user friendly. They would be used from Java like this: SomeClass s = new SomeClass(modulename.bar(), false); modulename.foo(s.getCPtrSomeClass()); Note that the following will have the same effect, but then it would not have been possible to call any proxy member functions in SomeClass: long s = modulename.bar(); modulename.foo(s); Now wrapper functions are generated: public class modulename { public static void foo(SomeClass s) { // calls the native function } public static SomeClass bar() { // calls the native function } } Which means these functions can now be used more naturally with proxy classes: SomeClass s = modulename.bar(); modulename.foo(s); 2) Changes when not using Java Proxy classes -------------------------------------------- The so called low-level interface was rather low-level indeed. The new static type checking implementation makes it less so but it remains a functional interface to the C/C++ world. Proxy classes are the obvious way to use SWIG generated code, but for those who want a functional interface all non-primitive types now have a simple Java class wrapper around the C/C++ type. Pointers and references to primitive types are also wrapped by type wrapper classes. The type wrapper classnames are based on the SWIG descriptors used by the other language modules. For example: C/C++ type Java type wrapper class name ---------- ---------------------------- int* SWIGTYPE_p_int double** SWIGTYPE_p_p_double SomeClass* SWIGTYPE_p_SomeClass SomeClass& SWIGTYPE_p_SomeClass SomeClass SWIGTYPE_p_SomeClass Note that everything wrapped by SWIG is accessed via a pointer even when wrapping functions that pass by value or reference. So the previous example would now be used like this: SWIGTYPE_p_SomeClass s = example.bar(); example.foo(s); Note that typedefs that SWIG knows about are resolved, so that if one has class Foo{}; typedef Foo Bar; then any use of Bar will require one to use SWIGTYPE_p_Foo; Some considerations: Make sure you make a firm decision to use either proxy classes or the functional interface early on as the classnames are different. 3) Pointers and non-parsed types -------------------------------- Sometimes SWIG cannot generate a proxy class. This occurs when the definition of a type is not parsed by SWIG, but is then used as a variable or a parameter. For example, void foo(Snazzy sds); If SWIG has not parsed Snazzy it handles it simply as a pointer to a Snazzy. The Java module gives it a type wrapper class around the pointer and calls it SWIGTYPE_p_Snazzy. In other words it handles it in the same manner as types are handled in the low-level functional interface. This approach is used for all non-proxy classes, eg all pointer to pointers and pointers to primitive types. 4) Backwards compatibility ----------------------- Backwards compatibility is not an issue if you have been using proxy classes and no global variables/functions. Otherwise some changes will have to be made. The native methods still exist but they are now in a JNI class, which is called modulenameJNI. As this class is really part of the internal workings, it should not be required so the class has become protected. Some pragmas/directives will hopefully be added to help with backwards compatibility. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 07/18/2002: beazley Modified wrapping of uninstantiated templates returned by value. Just to be safe, they are now wrapped by SwigValueWrapper<> just in case they don't define a default constructor. This would be used if you had code like this Foo blah(); void moreblah(Foo x); but you didn't instantiate Foo using %template. We should probably add a warning for this. 07/17/2002: beazley Added an error check to detect shadowed template paramaters. For example: template class Foo { public: int T; }; This results in an error, not a warning. This warning is also needed to fix some rather insidious problems like this: struct T { int blah; }; template class Foo { public: typedef T Traits; // Which T is this???? }; In this case, the template parameter T shadows the outer structure (which is what you want). 07/16/2002: beazley Improved support for templates with integer arguments. SWIG is much more aware of situations such as this: const int Size = 100; %template(Foo100) Foo<100>; void bar(Foo *x); // Knows that Foo is the same as Foo<100>; 07/15/2002: beazley Fixed bug with %feature/%ignore/%rename and namespaces. For example: %ignore Foo::Bar namespace Foo { class Bar { ... }; } Reported by Marcelo Matus. 07/09/2002: beazley Added parsing support for constructors that try to catch exceptions in initializers. For example: class Foo { Bar b; public: Foo(int x) try : b(x) { ... } catch(int) { ... } } This has no effect on the generated wrappers. However, the try and catch parts of the declaration are ignored. See Stroustrup, 3rd Ed, section 14.4.6.1 for details. 07/06/2002: beazley Fixed bug in template symbol table management. This fixes two bugs. First, mixing abstract methods, templates, and inheritance no longer generates a failed assertion. template class A { public: virtual void foo() = 0; }; template class B : public A { }; %template(A_int) A; %template(B_int) B; This fix also fixes a subtle problem with default values and templates. For example: template struct B { typedef unsigned int size_type; static const size_type nindex = static_cast(-1); void foo(size_type index = nindex); }; Bugs reported by Marcelo Matus. 07/05/2002: ljohnson (Lyle Johnson) [Ruby] Changed the definition of the SWIG_ConvertPtr() function for the SWIG/Ruby runtime support so that it looks like the Python version. If the last argument (flags) is non-zero, SWIG_ConvertPtr() will raise an exception for type mismatches as before. If flags is zero, this function will return -1 for type mismatches without raising an exception. *** POTENTIAL INCOMPATIBILITY FOR RUBY MODULE *** 07/04/2002: beazley Overloaded functions/methods/constructors now work in many language modules. The support is completely transparent--just call the function normally and SWIG will dispatch to the correct implementation. There are a variety of issues associated with this. Please refer to the overloading section of Doc/Manual/SWIGPlus.html for details. *** NEW FEATURE *** 07/04/2002: beazley Fixed a bug with namespaces, enums, and templates. For example: namespace hello { enum Hello { Hi, Hola }; template struct traits { typedef double value_type; }; traits::value_type say_hi() { return traits::value_type(1); } } SWIG wasn't generating wrappers that properly qualified traits. Reported by Marcelo Matus. 06/30/2002: beazley Supplied array variable typemaps for Tcl module. If you have a variable like this: int foo[10]; then a set function like this is generated: void foo_set(int *x) { memmove(foo,x,10*sizeof(int)); } 06/30/2002: beazley New %fragment directive. When writing typemaps, it can be easy to get carried away and write a lot of code. However, doing so causes tremendous code bloat. A common way to solve this is to write helper functions. For example: %{ void some_helper_function() { ... } %} %typemap(in) type { some_helper_function(...); } The only problem with this is that the wrapper file gets polluted with helper functions even if they aren't used. To fix this, a new fragment directive is available. For example: %fragment("type_helper","header") %{ void some_helper_function() { ... } %} %typemap(in, fragment="type_header") type { some_helper_function(...); } In this case, the code fragment is only emitted if the typemap is actually used. A similar capability is provided for declaration annotation and the %feature directive. For example: %feature("fragment","type_header") SomeDeclaration; The first argument to %fragment is the fragment name. The second argument is the file section where the fragment should be emitted. The primary use of this directive is for writers of language modules and advanced users wanting to streamline typemap code. *** EXPERIMENTAL NEW FEATURE *** 06/30/2002: beazley Supplied memberin typemaps for all arrays in an attempt to eliminate confusion about their use. 06/29/2002: beazley Experimental support for smart-pointers. When a class defines operator->() like this class Foo { ... Bar *operator->(); ... }; SWIG locates class Bar and tries to wrap its member variables and methods as part of Foo. For example, if Bar was defined like this: class Bar { public: int x; int spam(); }; You could do this (in the target language): f = Foo() f.x = 4 # Accesses Bar::x f.spam() # Accesses Bar::spam The primary use of this feature is to emulate the behavior of C++ smart-pointers---which allow attributes to accessed transparently through operator->. This feature is supported automatically in SWIG---no special directives are needed. To disable this behavior. Use %ignore to ignore operator->. *** NEW FEATURE *** 06/26/2002: beazley Deprecated the %except directive. %exception should be used instead. 06/25/2002: beazley Major cleanup of the modules directory. Eliminated most header files, consolidated module code into single files. 06/24/2002: beazley Reworked the instantiation of language modules. All language modules must now define a factory function similar to this: extern "C" Language * swig_python(void) { return new PYTHON(); } This function is then placed in a table and associated with a command line option in swigmain.cxx. This approach has a number of benefits. It decouples the SWIG main program from having to know about the class definitions for each module. Also, by using a factory function, it will be easier to implement dynamic loading of modules (simply load the file and invoke the factory function). 06/24/2002: beazley Fixed syntax error for reference conversions. For example: operator Foo &(); 06/24/2002: beazley Fixed syntax error for operator new[] and operator delete[]. 06/24/2002: beazley Fixed code generation problem for constants and default arguments involving templates. 06/19/2002: ljohnson (Lyle Johnson) [Ruby] Fixed a bug for the '-feature' command line argument; that setting was effectively being ignored and so the feature name was always set equal to the module name. 06/17/2002: beazley Fixed problems with static members and enums in templates. Version 1.3.13 (June 17, 2002) ============================== 06/16/2002: beazley Fixed a bug with __FILE__ expansion in the preprocessor. On Windows, the backslash (\) is now converted to (\\) in the string literal used for __FILE__. Reported by Steve Glaser. 06/14/2002: beazley Fixed warning message about 'name private in this context'. The warning is only generated for public methods. Reported by Scott Michel. 06/14/2002: beazley Fixed some problems related to template instantiation and namespaces. When SWIG expands a template, it does so with fully resolved types. For example, if you have this: template class foo { }; typedef double Double; %template(foo_d) foo; then, it is handled as foo in the typesystem. This fixes a number of subtle problems with inheritance and templates. 06/14/2002: ljohnson (Lyle Johnson) [Ruby] Added missing bool typemaps for INPUT, OUTPUT and INOUT in Lib/ruby/typemaps.i. 05/29/2002: cheetah (William Fulton) [Java] Fix for a couple of broken pragmas. 05/29/2002: cheetah (William Fulton) Fix for unnecessary cast when wrapping global variable where the type is not parsed by SWIG - Java variables example failure as reported by Larry Virden. 06/10/2002: beazley Modified %template to allow for empty instantiations. %template() foo; This registers foo with the type system, but doesn't wrap it (same as %ignore). This may only be a temporary measure. SWIG might be able to automatically instantiate templates in certain cases. 06/10/2002: beazley Fixed function prototype problems with Tcl 8.4 06/09/2002: beazley Fixed problem with templates and location of base classes. This one is a little mind-bending, but here is an example that illustrates: template struct traits { typedef ArgType arg_type; typedef ResType res_type; }; template struct Function { }; template struct Class : Function::arg_type, typename traits::res_type> { }; %template(traits_dd) traits ; %template(Function_dd) Function ; %template(Class_dd) Class ; In this example, the base class of 'Class' is determined from the Function template, but the types are obtained through typedefs. Because of this, SWIG could not locate the wrapped base class (Function). Should be fixed in 1.3.13 even though I can think of a million other things that might also be broken. 06/07/2002: beazley Fixed a problem with conversion operators. If you had an operator like this, operator double() const; SWIG was ommitting the "const" qualifier. This affected %rename and other directives. Reported by Zhong Ren. 06/07/2002: beazley Lessened the strictness of abstract class checking. If you have code like this: class Foo { public: virtual int method() = 0; }; class Bar : public Foo { public: Bar(); ~Bar(); }; SWIG will go ahead and generate constructor/destructors for Bar. However, it will also generate a warning message that "Bar" might be abstract (since method() isn't defined). In SWIG-1.3.12, SWIG refused to generate a constructor at all. 06/07/2002: beazley Change to %template directive. If you specify something like this: %template(vi) std::vector; It is *exactly* the same as this: namespace std { %template(vi) vector; } SWIG-1.3.12 tried to instantiate the template outside of the namespace using some trick. However, this was extremely problematic and full holes. This version is safer. 06/07/2002: beazley Fixed bug with scope qualification and templates. For example: A::DD Before, this was separated as scopes A, and DD. Fixed now. 06/06/2002: beazley Allow the following syntax: class A { }; struct B : A { ... }; A base class without a specifier is assumed to be public for a struct. 06/06/2002: beazley Fixed syntax error with template constructor initializers. Reported by Marcelo Matus. 06/06/2002: beazley Fixed bug with default template arguments. Reported by Marcelo Matus. 06/05/2002: beazley Fixed subtle problems with %rename directive and template expansion. Code like this should now work: %rename(blah) foo::method; ... template class foo { public: void method(); }; %template(whatever) foo; 06/05/2002: beazley Resolved some tricky issues of multi-pass compilation and and inheritance. The following situation now generates an error: class Foo : public Bar { ... }; class Bar { ... }; The following code generates a warning about incomplete classes. class Bar; class Foo : public Bar { }; The following code generates a warning about an undefined class. class Foo : public Bar { }; // Bar undefined This fixes a failed assertion bug reported by Jason Stewart. 06/05/2002: ljohnson [Ruby] Added a warning message for the Ruby module about the lack of support for multiple inheritance. Only the first base class listed is used and the others are ignored. (Reported by Craig Files). 06/03/2002: beazley Fixed a bug with struct declarations and typedef. For example: typedef struct Foo Foo; struct Foo { ... }; A few other subtle struct related typing problems were also resolved. Version 1.3.12 (June 2, 2002) ============================= 05/30/2002: beazley Fixed problem related to forward template class declarations and namespaces. Bug reported by Marcelo Matus. 05/30/2002: beazley Added 'make uninstall' target. Contributed by Joel Reed. 05/29/2002: beazley Fixed rather insidious bug with %rename, %feature and template specialization. For example: %exception vector::__getitem__ { ... some exception ... } template class vector { ... T __getitem__(int index); // Fine ... }; template<> class vector { ... T __getitem__(int index); // Oops. ... }; Now, the %exception directive (and other features) should correctly apply to both vector and specializations. 05/29/2002: beazley Subtle changes to %template() directive. Template arguments are now reduced to primitive types in template matching. For example: template class vector { ... partial specialization ... } typedef int *IntPtr; // Gross typedef // Gets the above partial specialization %template(vectorIntPtr) vector; This change is extremely subtle, but it fixes a number of potential holes in Luigi's STL library modules. For example: typedef int Integer; %template(vectori) vector; 05/29/2002: beazley Fixed rather insidious typemap bug related to const. const was being discarded through typedefs. 05/29/2002: ljohnson (Lyle Johnson) [Ruby] Added input typemaps for const references to primitive types (in Lib/ruby/ruby.swg). 05/29/2002: cheetah (William Fulton) [Java] The java arrray support functions are enclosed by a SWIG_NOARRAYS #define. Useful if not using arrays and it is desirable to minimise the amount of compiled code. 05/29/2002: cheetah (William Fulton) [Java] Enums were not renamed when using %name or %rename fix. 05/28/2002: ljohnson [Ruby] Modified the name of the wrapper functions for the "new" singleton method and "initialize" instance method for consistency with the other language modules. The wrapper name for the function that implements "new" is alloc_classname and the wrapper name for the function that implements "initialize" is new_classname. 05/27/2002: beazley Changes to runtime. Pointer conversion/creation functions now almost always have an extra "flags" argument. For example: SWIG_ConvertPtr(obj, void **, swig_type_info *ty, int flags); ^^^^^^^^^^ This extra parameter is reserved for future expansion and will be used for more control over pointers in future versions. 05/27/2002: beazley Fix for C++ classes with private assignment operators. It is now possible to safely return objects like this by value. Caveat: the class must provide a copy constructor. 05/26/2002: beazley -proxy option added to many language modules. This is the same as -shadow. We are merely changing terminology. 05/26/2002: beazley [perl] Fixed some inconsistencies in the -package option. -package merely sets the package name to be used on the wrappers. It does not change the name of the shared library file or the name of the generated .pm file. This was broken at some point, but works again now. 05/25/2002: beazley [perl] Fixed [ 475452 ] memory leak in return-by-value. Problem related to static member variables returning newly allocated objects. Reported by Roy Lecates. 05/25/2002: beazley [perl] Fixed [ 513134 ] %BLESSEDMEMBERS isn't always right. Reported by Fleur Diana Dragan. 05/25/2002: beazley Fixed [ 540735 ] -importall and the -I option. 05/25/2002: beazley [guile] Fixed [ 532723 ] Default arg for char* can SegV. Error in guile module. Reported by Brett Williams. 05/25/2002: beazley Subtle change to typemap application code. The "freearg" typemap must exactly match up with the "in" or "ignore" typemap. For example: %typemap(in) (char *data, int len) { ... }; %typemap(freearg) char *data { ... }; void foo(char *data, int len); In this case, the "in" typemap is applied, but the freearg typemap is not. This is because the freearg typemap doesn't match up with the input argument sequence. 05/25/2002: beazley Fixed [ 548272 ] Default argument code missing braces. Reported by Brett Williams. 05/25/2002: beazley Fixed [ 547730 ] SwigValueWrapper needed for constructors. Reported by William Fulton. 05/25/2002: beazley Undefined identifiers now evaluate to 0 when evaluating preprocessor expressions. For example: #if !FOO ... #endif where FOO is undefined or set to some non-numeric value. Fixes [ 540868 ] #if defined whatever - not parsed. Reported by Adam Hupp. 05/24/2002: beazley SWIG now ignores the C++ 'export' keyword. 05/23/2002: beazley Some refinement of type-name mangling to account for pointers, arrays, references, and other embedded type constructs. 05/23/2002: beazley Initial attempt at supporting template partial specialization. At the very least, it is parsed and the classes are stored. Matching of instantiations to specialized version is more limited and based on the SWIG default typemap rules: SWIGTYPE * SWIGTYPE [] SWIGTYPE & Now, why in the world would you want to use this feature? Other than allowing for slightly modified class APIs, this capability is primarily used to provide advanced wrapping support for STL-like objects. It can also be mixed with typemaps. Here is an example: /* Generic version */ template class vector { %typemap(in) vector * { // A container of objects } }; /* Partial specialization (pointers) */ template class vector { %typemap(in) vector * { // A container of pointers to objects. } }; /* Specialization (integers). */ template<> class vector { %typemap(in) vector * { // A container of integers. } }; *** EXPERIMENTAL FEATURE *** 05/23/2002: beazley Enhancement to typemaps. Normally, typemap variables are renamed to avoid conflicts. For example: %typemap(in) int * (int temp) { $1 = &temp; } This results in code that creates and uses variables "temp1","temp2", "temp3" and so forth depending on how many times the typemap is used. Sometimes you want a single variable instead. To do that, using the following naming scheme: %typemap(in) int *(int _global_temp) { } Is this case, a single variable _global_temp is emitted in the wrapper functions. It is shared across all typemaps. Repeated typemaps do not replicate the variable---they use the first one emitted. *** NEW FEATURE *** 05/23/2002: beazley Minor enhancement to typemaps. If you have this code, %typemap(in) Foo (int somevar = 3) { ... } the default value for somevar is now emitted into the wrapper code. 05/22/2002: beazley Fixed %extend to be better behaved in namespaces. If you have code like this: namespace foo { struct bar { %extend { void blah(); }; }; } SWIG matches the blah() method to a C function named void foo_bar_blah(foo::bar *self). This is consistent with the non-namespace version. Bug reported by Marcelo Matus. 05/22/2002: beazley New library files: cpointer.i, carrays.i, cmalloc.i. These provide access to C pointers and memory allocation functions. See Doc/Manual/Library.html for details. 05/22/2002: cheetah (William Fulton) [Java] C type char no longer maps to Java type byte, but to Java type char. It is now treated as a character rather than a signed number. This fits in with the other language modules and is a more natural mapping as char* is mapped as a string of characters. Note that the C signed char type is still mapped to a Java byte. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 05/22/2002: cheetah (William Fulton) [Java] Improved constants wrapping. Constants (#define and %constant) values are now obtained through a JNI call. Previously the value was compiled as Java code, but this didn't work for all cases, eg #define 123ULL. 05/22/2002: beazley Fixed bogus error message with %extend directive and C++ access specifiers. Reported by Marcelo Matus. 05/22/2002: beazley Namespaces and enums now work correctly. For example: namespace Foo { enum Bar { A, B }; } Bug reported by Marcelo Matus. 05/21/2002: beazley The %types directive can now be used to specify inheritance relationships in the runtime type system. For example, %types(Foo = Bar); specifies that Foo isa Bar. Using this is potentially quite dangerous. However, this is useful in certain cases (and in the SWIG library). 05/20/2002: beazley %nodefault and %makedefault directives now require a trailing semicolon. For example: %nodefault; ... %makedefault; In addition both directives can take a class name. For example: %nodefault Foo; class Foo { /* No default constructor/destructor */ }; class Bar { /* Default constructor/destructor generated */ }; *** POTENTIAL INCOMPATIBILITY *** If you don't use the trailing semicolon, things will mysteriously break. 05/20/2002: beazley More improvements to type system handling. SWIG now correctly handles template names and parameters in a namespace. For example: namespace foo { template class bar { }; typedef int Integer; void blah(bar *x); }; In the generated code, all of the typenames are properly qualified. 05/17/2002: cheetah (William Fulton) [Java] deprecated broken -jnic and -jnicpp commandline options. The C or C++ JNI calling convention is now determined from the -c++ commandline option. 05/16/2002: cheetah (William Fulton) [Java] The JCALL macros which exist so that the same typemaps can be used for generating both the C and C++ JNI calling conventions no longer appear in the generated code. This is because the output is now passed through the SWIG preprocessor which does the macro expansion for either C or C++ (depending on whether -c++ is passed on the SWIG commandline). The generation of the functions used in the array typemaps have been adjusted to take account of this. The side effect is that any typemaps which contained JCALL macros within %{ %} brackets will have to be moved within {} brackets so that the SWIG preprocessor can expand the macros. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 05/13/2002: beazley Class templates may now be used as template parameters. For example: template class C> class Foo { ... }; template class Bar { ... }; %template(Fooi) Foo; SWIG doesn't really do anything special with this---it's just another way of specifying a template parameter. 05/13/2002: beazley Minor refinement of template support. Template parameter names are no longer required for types. For example: template class Foo { }; Obviously, names are required for template; 05/12/2002: beazley New macro expansion in typemaps. The sequence: $descriptor(type) Will expand into the SWIG type descriptor structor for the given type. Type may be any abstract datatype. For example: $descriptor(int *) $descriptor(int (*)(int,double)) $descriptor(vector *) Caveat: It is *NOT* currently legal to use other typemap substitution variables in the macro. For example $descriptor($1_type). The primary purpose of this modification is to better support typemaps for container objects or to allow typemaps that might be performing type conversions. *** NEW FEATURE *** 05/11/2002: beazley The wrapping of references to primitive types has been changed as follows: Arguments of type 'const primitive &' are now passed by value as opposed to pointers. Return values of type 'const primitive &' are returned as values instead of pointers. 'primitive' is any one of int, short, long, long long, char, float, double, bool (as well as unsigned variants). This change is being made to better support C++ wrapping--especially code that makes use of templates and the STL. 05/11/2002: beazley The %template directive can now be used to access templates in a namespace. For example: namespace std { template class complex { T re, im; public: complex(T _r = T(), T _i = T()) : re(_r), im(_i) { } T real() { return re; } T imag() { return im; } }; } %template(complex) std::complex; Note: There are some very subtle namespace/symbol table management issues involved in the implementation of this. It may not work in certain cases. 05/10/2002: beazley Member template constructor support added. For example: template struct pair { _T1 first; _T2 second; pair() : first(_T1()), second(_T2()) { } template pair(const pair<_U1,_U2> &x); }; To instantiate the template, use %template and %extend. For example, this expands the constructor into a default copy constructor: %extend pair { %template(pair) pair<_T1,_T2>; } Highly experimental. Other uses may be broken. 05/10/2002: beazley The %extend (%addmethods) directive no longer works unless it appears in the public section of a class. An error message is now generated (as opposed to a segmentation fault). 05/09/2002: beazley New %warnfilter() directive. This directive attaches a warning filter to specific declarations and has the same semantics as %rename, %ignore, %feature, and so forth. For example: %warnfilter(501) foo; // Suppress overloaded warning int foo(int); int foo(double); or %warnfilter(501) Object::foo(double); class Object { public: int foo(int); int foo(double); }; This feature only suppresses warnings in later stages of code generation. It does not suppress warnings related to preprocessing or parsing. *** NEW FEATURE *** 05/09/2002: beazley SWIG now supports C99 variadic preprocessor macros. For example: #define debugf(fmt,...) fprintf(stderr,fmt,__VA_ARGS__) The argument "..." is used to indicate variable arguments which are all placed into the special argument name __VA_ARGS__ in the macro expansion. SWIG also implements the GNU (##) extension for swallowing the preceding comma when __VA_ARGS__ is empty. For example: #define debugf(fmt,...) fprintf(stderr,fmt, ##__VA_ARGS__) Here is how this is expanded: debugf("%d", 3) --> fprintf(stderr,"%d",3) debugf("Hello") --> fprintf(stderr,"Hello" ) (notice the deleted comma). *** NEW FEATURE *** 05/08/2002: samjam (Sam Liddicott) Many changes to php module. Shadow classes are now implemented entirely in native C and no need for php-code shadow wrappers Populated template config.m4 and Makefile.in as needed by phpize are generated. 05/08/2002: ljohnson (Lyle Johnson) [Ruby] A copy constructor is now turned into a "clone" instance method (see Dave's change for copy constructors dated 4/7/2002). This seems like the appropriate thing to do for Ruby code. 05/08/2002: ljohnson (Lyle Johnson) [Ruby] Fixed [ 553864 ] Inline destructor code not written. 05/08/2002: beazley %ignore behaves better with constructors, destructors, and the type system in general. For constructors and destructors, %ignore now suppresses the creation of a default constructor or destructor. For example: %ignore ~Foo; class Foo { public: Foo(); ~Foo(); ... }; In SWIG-1.3.11, ~Foo() simply "disappeared" and the code generator created a wrapper for a default destructor (as if it was never declared in the interface). In SWIG-1.3.12, %ignore suppresses the creation of a destructor if one is actually defined. Similarly, even though a declaration is ignored, information may still be needed to properly handle types. For example, here is a very subtle error that is fixed by this change: %ignore std::string; // Prevent class wrapping namespace std { class string { ... }; %typemap(in) string * { ... } } void foo(std::string *s); // Broken. Before this fix, %ignore would cause the class definition to disappear. This, in turn, would cause the typemap to be misapplied. 05/08/2002: beazley Minor changes to %rename, %ignore, %feature, and related directives for better support of destructors. Destructors can now be precisely tagged. For example: %ignore Foo::~Foo; %feature("action") ~Bar { ... } *Developer warning* Operations such as renaming and feature attachment for classes used to be applied to destructors as well. For instance, if you did this: %rename(Bar) Foo; The operation applied to the class itself, the constructor, and the destructor. This is no longer the case. Now such operations will only apply to the class and the constructor. Note: if you were relying on this for class renaming, be aware that renamed classes should really only be handled at the level of the class itself and not the level of individual declarations in the class (although they can be renamed individually if needed). As far as I know, the Language class is already taking care of this case correctly. 05/07/2002: beazley New set of tests. The Examples/test-suite/errors directory contains tests that try to exercise all of SWIG's error and warning messages. 05/07/2002: beazley Start of a warning framework. Warning messages are now assigned numeric values that are shown in warning messages. These can be suppressed using the -w option. For example: swig -w302 example.i swig -w302,305 example.i Alternatively, the #pragma preprocessor directive can be used to disable this: #pragma SWIG nowarn=302 #pragma SWIG nowarn=302,305 Note: Since SWIG is a multi-pass compiler, this pragma should only be used to change global settings of the warning filter. It should not be used to selectively enable/disable warnings in an interface file. The handling of #pragma occurs in the C++ preprocoessor and affects all subsequent stages of compilation. The -Wall option turns on all warnings and overrides any filters that might have been set. Warnings can be issued from an interface using %warn. For example: %warn "110:%section is deprecated" The first part of a warning message is an optional warning number. A complete set of warning numbers is found in Source/Include/swigwarn.h. *** NEW FEATURE *** 05/07/2002: beazley Internal parsing change. Directives to include files now use brackets [ ... ] instead of { ... }. %includefile "foo.i" [ ... ] The use of { ... } was a bad choice because they were included implicitly by the preprocessor and made it impossible to properly detect legitimate missing '}' errors. 04/16/2002- 05/02/2002: beazley SWIG European Tour: Paris-Amsterdam-Bath. 04/23/2002: beazley The %addmethods directive has been renamed to %extend. For example: class Foo { ... }; %extend Foo { int blah() { ... }; int bar() { ... }; ... }; Motivation: the %addmethods directive can be used for many other tasks including adding synthesized attributes, constructors, and typemaps. Because of this, "addmethods" is somewhat misleading. %extend more precisely describes this operation---extension of a class or structure. *** POTENTIAL INCOMPATIBILITY *** %addmethods still works via a macro definition. However, a warning message may be generated. Errors involving %addmethods will actually refer to the %extend directive. 04/23/2002: beazley Further refinement of the type system. Typedef now propagates through functions, pointers to functions, and pointers to member functions. For example: typedef int Integer; void foo(int (*x)(int), Integer (*y)(Integer)); In this case, arguments 'x' and 'y' have exactly the same type (and would obviously accept objects of either type). Similarly, consider this: class Foo { }; typedef Foo Bar; void bar(int (Foo::*x)(int), int (Bar::*y)(int)); In this case, arguments x and y are the same type (via typedef). 04/22/2002: beazley SWIG now generates a warning message if any part of an expression involves values from a private part of a class. For example: class Foo { private: static int X; public: void blah(int a, int b = X); // Warning }; In this case, the default argument is ignored. There are workarounds, but they are rather clumsy. For instance, you might do this: %feature("action") blah(int,int) { if ($nargs == 1) { result = blah(arg1); } else { result = blah(arg1,arg2); } } void blah(int a, int b = 0); 04/21/2002: beazley Use of the %inline directive inside a namespace is forbidden and now generates an error message. This is not allowed since the inlined code that is emitted is not placed inside a namespace. This confuses other stages of parsing. 04/21/2002: beazley Some bug fixes to casting operations and expression parsing. Due to some parsing issues, it is not currently possible to use casts for all possible datatypes. However, the common cases work. 04/20/2002: beazley (Amsterdam) Member templates now work. Simply use the %template directive inside a class or %addmethods to create instantiations (see Doc/Manual/SWIGPlus.html). Supporting this was easy---earlier changes to templates made it possible using only a two-line modification to the parser and a few minor modifications elsewhere. Hmmm, come to think of it, the smoke was rather thick in that Internet "cafe". *** NEW FEATURE *** 04/19/2002: beazley (TGV) Improved handling of non-type template parameters. For example: vector; Simple numbers and strings can be used with the %template directive as well. For example: %template(vecint100) vector; Note: Arithmetic expressions are not currently allowed. Default template arguments now work and do not have to be given to %template. 04/18/2002: beazley (Paris) Change in internal template handling. Template parameters are now fully integrated into the type system and are aware of typedefs, etc. This builds upon the change below. *** DEVELOPER WARNING *** Word of caution to language module writers. The "name" parameter of certain parse tree nodes (classes, functions, etc.) may be parameterized with types. This parameterization is done using SWIG type-strings and not the underlying C version. For example, int max(int *,int *) has a name of "max<(p.int)>". If you use the name directly, you may get syntax errors in the generated code. To fix this, use SwigType_namestr(name) to convert a parameterized name to a C name with valid syntax. The internal version is used to reduce template types to a common representation and to handle issues of typedef. 04/16/2002: beazley (somewhere over the Atlantic) Enhancement of typedef resolution. The type system is now aware of template arguments and typedef. For example: typedef int Integer; foo(vector *x, vector *y); In this case, vector and vector are the same type. There is some interaction between this mechanism and the implementation of typemaps. For example, a typemap defined for vector * would apply to either type. However, a typemap for vector * would only apply to that type. Typedefs and typemaps and matched by left-most expansion. For example: vector --> vector --> vector 04/24/2002: cheetah (William Fulton) [Java] Changes to Java shadow classes. Overcomes a bug where the module assumed that a pointer to a derived class could be used in place of a pointer to a base class. Thanks to Stephen McCaul for analysing the bug and submitting patches. A consequence is that the getCPtr() method in each shadow class has disappeared and has been replaced with a getCPtrXXX(), where XXX is the shadow class name. If you have code that previously used getCPtr(), and the associated class is wrapping a C struct or a C++ class that is not involved in an inheritance chain, just use the new method. If however, the class is involved in an inheritance chain, you'll have to choose which pointer you really want. Backwards compatibility has been broken as not using the correct pointer can lead to weird bugs through ill-defined behaviour. If you are sure you want the old methods, you could add them back into all shadow classes by adding this at the beginning of your interface file: %pragma(java) allshadowcode=%{ public long getCPtr(){ return swigCPtr; } %} Please see entry dated 07/23/2002 to see how to do this after the deprecation of the allshadowcode pragma. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 04/13/2002: beazley Fixed problem with default arguments and references. Declarations such as this should now work: void foo(const string &x = "Hello"); 04/12/2002: beazley Added typemap $* substitutions for typemaps involving arrays. Requested by William Fulton. 04/11/2002: beazley Template specialization is now supported. For example: template<> class vector { ... }; When the %template directive is used, it will use a specialization if one is defined. There are still some limitations. Partial specialization is not supported. A template of type does not match all pointers. *** NEW FEATURE *** 04/11/2002: beazley Major change to template wrapping internals. Template declarations are no longer processed as macros but now result in real parse-tree nodes. The %template directive expands these nodes into a specific instantiation. This change enables a number of new and interesting capabilities: Directives such as %rename, %feature, and %addmethods can now be applied to uninstantiated templates. For example: %rename(barsize) vector::bar(char *buf, int len); ... template class vector { public: ... void bar(char *buf); void bar(char *buf, int len); // Renamed ... }; %template(intvector) vector; // Renaming carries through By parsing templates into an internal data structure, it will be possible to support specialization (and maybe partial specialization). This is highly experimental and a work in progress. *** POTENTIAL INCOMPATIBILITY *** In SWIG-1.3.11, template declarations were simply processed as weird macros. No other information was retained. This made it impossible to support more advanced features and complicated many other parts of the implementation. 04/09/2002: beazley Change to template class wrapping. There were a variety of "issues" with the old approach related to parsing, the type system, and namespaces. These changes are meant to rectify some of these problems: A specific instantiation of a template can now be specified by including the class inline like this: class vector { public: vector(); ~vector(); ... whatever ... }; This is template specialization, but partial specialization is not yet implemented. The %template directive has been modified to expand roughly as follows: %template(vecint) vector; becomes %rename(vecint> vector; class vector { public: vector(); ... }; Note that this simply builds upon the code above (templates included inline). This modified approach to wrapping fixes some subtle type issues. For instance, you can now define typemaps and typedefs like this: %typemap(in) vector * { ... } typedef vector intvector; ... void blah(intvector *v); // Gets the above typemap This did not work in SWIG-1.3.11 due to a peculiarity of the template implementation. %template(name) no longer installs the template as a class with name "name". This might break %addmethods as described in the manual. For example: %template(vecint) vector; %addmethods vecint { // Fails. vecint not a class ... }; To fix this, just use the template name instead: %addmethods vector { ... } Note: This technique might be a way to implement some bizarre template specialization techniques. For example: %addmethods vector { // Only applied if vector instantiated later %typemap(in) vector * { ... } ... }; *** POTENTIAL INCOMPATIBILITY *** 04/08/2002: beazley Fixed [ 540868 ] #if defined whatever - not parsed. SWIG should now correctly handle preprocessor directives like this: #if defined __cplusplus ... #endif Note: was implemented previously, but there was a minor bug. Reported by Adam Hupp. 04/07/2002: beazley %readonly and %readwrite are deprecated due to a change in the implementation. Instead of being pragmas, mutability is now controlled as a "feature" using the following two directives: %immutable; int x; // read-only variable int y; // read-only variable %mutable; int z; // Modifiable %immutable and %mutable are much more powerful than their older counterparts. They can now pinpoint a specific declaration like this: %immutable x; /* Any x */ %immutable Foo::x; /* x in class Foo */ In fact, the matching algorithm is the same as for %rename, %ignore, and other directives. This means that the declaration %immutable Foo::x; would not only apply to class Foo but to all derived classes as well. *** POTENTIAL INCOMPATIBILITY *** %immutable and %mutable must be terminated by a semi-colon. This differs slightly from the older %readonly and %readwrite directives. Since %immutable and %mutable can be applied to declarations the semicolon is needed to distinguish between a global feature and one targeted to a single declaration. Note: this incompatibility is the primary reason for changing the name of the directive. 04/07/2002: beazley New handling of copy constructors. If a class defines constructors like this: class Foo { public: Foo(); Foo(const Foo &); // Copy constructor ... }; SWIG now generates a function copy_Foo() for the copy constructor. In previous verions, this generated a name-clash and an error message. To preserve backwards compatibility, SWIG does not change the behavior if %rename is used to resolve the name conflict. However, if no name resolution is made, this new approach is used. Copy constructors may be handled as a special case in the target language. However, this is up to the language module itself. 04/07/2002: beazley The %template directive is now namespace aware. This allows code like this: namespace foo { template max(T a, T b) { return a > b ? a : b; } } using namespace foo; %template(maxint) max; // Ok namespace bar { using foo::max; %template(maxdouble) max; // Ok } Caveat: the template name supplied to %template must be defined in the same scope in which the %template directive appears. This code is illegal: %template(maxint) foo::max; 04/07/2002: beazley Minor enhancement to preprocessor. The preprocessor can now perform string comparison. For example: #define A "hello" ... #if A == "hello" ... #endif The primary use of this is in SWIG macros. For example: %define FOO(x) #if #x == "int" /* Special handling for int */ ... #endif %enddef Normal users can probably safely ignore this feature. However, it may be used in parts of the SWIG library. 04/07/2002: beazley Further refinement of default constructor/destructor wrapper generation. SWIG is now much more aware of pure virtual methods. For instance: class A { /* Abstract */ public: virtual void method1() = 0; virtual void method2() = 0; }; class B : public A { /* Abstract */ public: virtual void method1() { }; }; class C : public B { /* Ok */ public: virtual void method2() { }; }; In this case, SWIG will only generate default constructors for C. Even though B looks fine, it's missing a required method and is abstract. 04/04/2002: beazley Subtle change to structure data member access. If you have a structure like this: struct Foo { Bar b; }; The accessor functions for b are generated as follows: (1) If b is *not* defined as a structure or class: Bar Foo_b_get(Foo *self) { return self->b; } void Foo_b_set(Foo *self, Bar value) { self->b = value; } (2) If b *is* defined as a structure or class: Bar *Foo_b_get(Foo *self) { return &self->b; } void Foo_b_set(Foo *self, Bar *value) { self->b = *value; } See the "Structure data members" section of Doc/Manual/SWIG.html for further details. *** POTENTIAL INCOMPATIBILITY *** This may break interfaces that relied on a lot of a undeclared structure and class names. To get the old behavior, simply use a forward declaration such as "struct Bar;" 04/04/2002: beazley C++ namespace support added. SWIG supports all aspects of namespaces including namespace, using, and namespace alias declarations. The default behavior of SWIG is to flatten namespaces in the target language. However, namespaces are fully supported at the C++ level and in the type system. See Doc/Manual/SWIGPlus.html for details on the implementation. 04/02/2002: cheetah (William Fulton) [Java] Sun has modified javac in jdk1.4 to no longer compile an import of an unnamed namespace. To fix this SWIG no longer generates the import for packageless classes. http://developer.java.sun.com/developer/bugParade/bugs/4361575.html As reported SF #538415. 03/27/2002: ljohnson (Lyle Johnson) [Ruby] Added support for pointer-to-member, similar to that for the Python module. Remarkably similar. Also added a new example for this (Examples/ruby/mpointer), which is remarkably similar to the Python example of the same name. 03/26/2002: ljohnson (Lyle Johnson) [Ruby] Made a few minor edits to the "Advanced Topics" chapter of the SWIG manual and added a new major section about how to create multi-module Ruby packages with SWIG. 03/26/2002: ljohnson (Lyle Johnson) [Ruby] Removed all of the old Ruby pragmas. If any of this functionality is truly missed we can resurrect it, preferably with some kind of feature-based directive. 03/25/2002: ljohnson (Lyle Johnson) [Ruby] Fixed SWIG exception library support for Ruby, which has apparently been broken for some time. Luckily, no one seems to have noticed. 03/23/2002: beazley C++-namespace support in SWIG directives. %addmethods: The %addmethods directive now accepts a fully qualified classname and can be used inside C++ namespace declarations. For example: // Attaches to the class Foo::Bar below %addmethods Foo::Bar { int somemethod() { ... } }; namespace Foo { class Bar { public: ... }; // Attaches to the class Bar above %addmethods Bar { int othermethod() { ... }; } } %feature, %rename, %ignore, %exception, and related directives: Namespaces are fully integrated into the the renaming and declaration matcher. For example: %rename(display) Foo::print; // Rename in namespace Foo %ignore Foo::Bar::blah; // Ignore a declaration %rename directives can be placed inside namespace blocks as well. For example: namespace Foo { %rename(display) print; // Applies to print below void print(); }; Most other SWIG directives should work properly inside namespaces. No other changes are needed. 03/22/2002: beazley Some changes to internal symbol table handling. SWIG no longer manages structures and unions in a separate namespace than normal declarations like ANSI C. This means you can't have a structure with the same name as a function. For example: struct Foo { ... } int Foo() { ... } This approach is more like C++. It's not clear that SWIG ever really supported the ANSI C anyways---using the same name would almost certainly generate a name-clash in the target language. 03/22/2002: ljohnson (Lyle Johnson) [Ruby] Fixed [ 517302 ] for handling of renamed overloaded constructors. Now, renamed overloaded constructors are converted into class singleton methods (basically acting as "factory" methods). 03/21/2002: beazley Fixed [ 532957 ] %ignore parse error and casting operator. Reported by William Fulton. 03/18/2002: beazley (** ADVANCED USERS ONLY **) Added support for dynamic casting in return values. A somewhat common problem in certain C++ programs is functions that hide the identity of underlying objects when they are returned from methods and functions. For example, a program might include some generic method like this: Node *getNode(); However, Node * may just be base class to a whole hierarchy of different objects. Instead of returning this generic Node *, it might be nice to automatically downcast the object into the appropriate type using some kind dynamic cast. Assuming you understand the peril involved, a downcast can now be performed using the following function in the run-time type checker: swig_type_info *SWIG_TypeDynamicCast(swig_type_info *, void **ptr); This function checks to see if the type can be converted to another type. If so, a different type descriptor (for the converted type) is returned. This type descriptor would then be used to create a pointer in the target language. To use this, you would write a typemap similar to this: %typemap(out) Node * { swig_type_info *ty = SWIG_TypeDynamicCast($1_descriptor, (void **) &$1); $result = SWIG_NewPointerObj($1, ty); } Alternatively, %typemap(out) Node * = SWIGTYPE *DYNAMIC; To make the typemap have any effect, you have to write a supporting function that knows how to perform downcasting. For example: %{ static swig_type_info * Node_dynamic_cast(void **ptr) { Node **nptr = (Node **) ptr; Element *e = dynamic_cast(*nptr); if (e) { *ptr = (void *) e; return SWIGTYPE_p_Element; } Data *d = dynamic_cast(*nptr); if (d) { *ptr = (void *) d; return SWIGTYPE_p_Data; } return 0; } %} There is no restriction on how types are determined. dynamic_cast<> uses C++ RTTI. However, if you had some other mechanism for determining the type, you could use that here. Note: it is important to save the new pointer value back into the argument as shown. When downcasting, the value of the pointer could change. Finally, to make the casting function available, you have to register it with the run-time type checker. Put this macro in your interface file. DYNAMIC_CAST(SWIGTYPE_p_Node, Node_dynamic_cast); Note: this feature does not introduce a performance penalty on normal SWIG operation. The feature is only enabled by writing a new typemap that explicitly calls SWIG_TypeDynamicCast() to make a conversion. Examples/test-suite/dynamic_cast.i contains a simple example. This feature is not supported in the Java module due to differences in the type-checking implementation. *** EXPERIMENTAL FEATURE *** 03/17/2002: beazley Small change to type-name handling of unnamed structures and typedef. If a structure of this form appears: typedef struct { ... } Foo; Then 'Foo' is used as the proper typename for the structure. Furthermore, Foo can now be used as a name in C++ inheritance. SWIG was already kind of doing this, but this modification refines the implementation to more closely follow the C++ ARM, section 7.1.3, p. 106. This fixes a couple of obscure corner cases. 03/16/2002: beazley Modified C++ inheritance with a few enhancements. First, type information needed for casting and type-equivalence is generated even when base-classes aren't defined in the interface. For example: class Foo : public Bar { /* Bar unspecified */ public: ... }; void blah(Bar *b); In this case, the blah() function still accepts Foo * even though nothing is really known about Bar. Previous SWIG versions would just generate a type error. Inheritance has also been modified to work through typedef. For example: class Bar { }; typedef Bar OtherBar; class Foo: public OtherBar { } In this case, the base class of OtherBar is correctly resolved back to Bar. The use of the name OtherBar is lost in this resolution (the wrappers will simply use Bar instead of the typedef name OtherBar). 03/13/2002: beazley %typemap, %apply, and related directives can now appear inside class definitions. 03/13/2002: beazley Fixed a variety of problems related to compiling SWIG on 64-bit platforms. 03/12/2002: beazley Fixed problem with "ignore" and "in" typemaps. Local variables associated with "in" were being added to the wrapper function even though they were never used. Mostly harmless, but it would lead to a variety of compilation warnings. 03/12/2002: beazley Some changes to the internal type system and handling of nested C++ types. In previous versions of SWIG, if you had the following: class Foo { public: typedef int Blah; }; class Bar : public Foo { public: void somemethod(Blah x); }; The argument type in somemethod() would implicitly be set to Bar::Blah. Although this is technically allowed, it breaks typemaps. For example: %typemap(in) Foo::Blah { ... } doesn't match like you expect. This has been changed in SWIG-1.3.12. Now, types are expanded using the class in which they were defined. So, the argument type in somemethod() will be Foo::Blah---since the type Blah was defined in Foo. 03/10/2002: beazley Fixed some subtle type scoping problems with typedef and C++ classes. For example: typedef int Blah; class Bar { public: typedef double Blah; void foo(Blah x, ::Blah y); ... } 03/10/2002: beazley Highly experimental change to handle variable length arguments. First, there is no portable or reliable way to wrap a varargs function in full generality. However, you *can* change the function signature using %varargs. %varargs(char *) fprintf; ... void fprintf(FILE *f, char *fmt, ...); In this case, the variable length parameter "..." is simply replaced by the parameters given in %varargs. This results in a function like this: void fprintf(FILE *f, char *fmt, char *s); More than one argument can be used and default values can be defined. For example, this code specifies a maximum of four arguments. %varargs(char *x1 = 0, char *x2 = 0, char *x3 = 0, char *x4 = 0) fprintf; *** EXPERIMENTAL NEW FEATURE *** 03/10/2002: beazley Change to handling of variable length arguments. varargs is now handled as a proper parameter and is passed to the code generator. However, it still can't be handled correctly (and will generate a typemap warning). This change has been made to better incorporate variable length arguments with other directives such as %ignore, %rename, %feature, and so forth. 03/10/2002: beazley Fixed [ 522555 ] Syntax error parsing "define" construct. SWIG is a little more restrictive in determining #define statements that will be wrapped as constants. Also added a better parser error rule for handling bad constants. 03/08/2002: cheetah (William Fulton) [Java] Bug fix: Classes renamed with %rename that are derived from another class generate more appropriate shadow class code. 03/08/2002: cheetah (William Fulton) [Java] Fixed SF [ #523632 ] and [ #513335 ] both reported by Israel Tanner. Support for types that are used which are in a typedef. The appropriate shadow class name is generated. Also generated correct shadow classname when a templated class is used within another templated class. See the cpp_typedef.i testcase. 03/08/2002: cheetah (William Fulton) [Java] Bug fix: No type was generated in shadow classes for types that weren't wrapped by SWIG. The type is treated as a raw pointer, ie no shadow class. 02/22/2002: beazley Refined the matching algorithm used by %rename, %ignore, and %feature. If a type signature is supplied, it must exactly match that used in the declaration---including any use of const. For example: %rename(foo1) foo(int); %rename(bar1) bar(int) const; class Blah { public: void foo(int); // Matched --> foo1 void foo(int) const; // Not matched void bar(int); // Not matched void bar(int) const; // Matched --> bar1 } In previous versions, a non-const specification would match both the non-const and const declarations. However, the whole point of %rename and related directives is that they be able to precisely pinpoint exact declarations in an interface. This fixes the problem. 02/21/2002: beazley Reworked the handling of default constructor and destructors. SWIG now makes a preliminary pass over the parse tree to discover which classes support default allocation. This fixes a number of very subtle issues in code generation and call/return by value. 02/18/2002: cheetah (William Fulton) Improved support on Cygwin: Perl, Python, Tcl, Ruby and Java should work out of the box, barring the runtime library. Removed dllwrap and replaced with newly working gcc -shared instead for Cygwin. All this will require the new improved binutils 20010802 and later, but the latest Cygwin is usually the best recommendation. 02/15/2002: beazley Fixed some problems related to wrapping of global variables and Perl shadow classes. Reported by Chia-liang Kao. 02/15/2002: ljohnson (Lyle Johnson) [Ruby] Made a fix to the code generation for C++ class constructors so that we get both a "new" singleton method and an "initialize" instance method for each class. This change enables developers to derive new Ruby classes from SWIG-wrapped C++ classes and then override their initialize methods to provide subclass-specific instance initialization. 02/15/2002: ljohnson (Lyle Johnson) [Ruby] Massive documentation update for the Ruby module, contributed by Craig Files. 02/14/2002: ljohnson (Lyle Johnson) [Ruby] Bug fix: An error in the SWIG runtime support for Ruby was causing several of the examples to fail. Reported by William Fulton. 02/14/2002: ljohnson (Lyle Johnson) [Ruby] Bug fix: Enumerations defined within a class (such as those seen in the Examples/ruby/enum example) were not being exported with the correct names. Reported by William Fulton. 02/13/2002: ljohnson (Lyle Johnson) [Ruby] Added a warning message when we run across overloaded class constructors for C++ code, that this is currently not supported (even if the overloads have been %renamed). For an example of where this doesn't work, see Examples/ruby/operator. 02/13/2002: ljohnson (Lyle Johnson) [Ruby] Added an "ignored" warning message when the parser runs across an operator!=() declaration for C++ code. 02/11/2002: ljohnson (Lyle Johnson) [Ruby] Added the "import", "import_template", "operator" and "template" examples. 02/11/2002: ljohnson (Lyle Johnson) [Ruby] Added multi-module support. 02/09/2002: ljohnson (Lyle Johnson) [Ruby] Added the missing "#define SWIG_NOINCLUDE" at the top of the wrapper code when the '-c' option is used. 02/09/2002: ljohnson (Lyle Johnson) Corrected a minor off-by-one error for the size of the swig_types[] array that's generated in the wrapper code. 02/08/2002: beazley Fixed SF [ #515058 ] Wrong code for C++ templates. Reported by Israel Taller. Version 1.3.11 (January 31, 2002) ================================= 01/30/2002: beazley Fix to pass/return by value for C++ objects that define no default constructor. Changes to the typemap system made it impossible to wrap C++ objects with no default constructor. This has been fixed, but the solution involves some clever template magic contributed by William Fulton. Please see the comments in the file Lib/swig.swg for further details. This solution is experimental and may be refined in a future release. 01/30/2002: beazley Global variables and member data of type "const char *" can be set, but the old value is silently discarded without any garbage collection. This may generate a memory leak. This change is needed to more safely handle variables like this: const char *foo = "Hello World\n"; In this case, it's not safe to free the old value. However, SWIG can dynamically allocate a new value and make foo point to it. To fix this memory leak, you can probably do this: %clear const char *foo; %apply char * {const char *foo}; *** POTENTIAL INCOMPATIBILITY *** 01/30/2002: beazley Two minor typemap enhancements have been added. First, typemaps can issue a warning message by including a special warning attribute. For example: %typemap(in,warning="I'm going to do something dangerous") ... The warning message will show up whenever the typemap is applied. Second, a typemap can force a no-match by defining %typemap(in) sometype "pass"; If this is used, the typemap system will *not* record a typemap match for "sometype". This can be used to block selected typemaps. For example, if you wanted to disable a typemap feature for some type, you could do this. // Do not allow global variables of type 'const char *' to be set. %typemap(varin) const char * "pass"; It might also be possible to use this to do subtle and strange things with typemaps. For example, if you wanted to make 'blah *' an output value and 'const blah *' an input parameter, you might do this: %typemap(ignore) blah *(blah temp) { $1 = &temp; } %typemap(argout) blah * { ... return a value ... } /* Block unqualified typemaps defined above */ %typemap(ignore) const blah * "pass"; %typemap(argout) const blah * "pass"; %typemap(in) const blah * { ... get input value ... } (This potential applications of typemaps suggested by Greg Stein). *** NEW FEATURE *** 01/29/2002: cheetah (william fulton) [Java] Bug fix: No enumerations were wrapped when the -shadow commandline option was not specified. Reported by Israel Taller. 01/28/2002: cheetah (william fulton) [Java] Global arrays are successfully wrapped. In fact they started mostly working in SWIG-1.3.10. 01/28/2002:richardp Added first attempt at C++ and -shadow support for PHP4 module, please test and mail me if any problems/ideas on improving it. There is a known problem with uninitialized member variables, please see Examples/php4/sync/README for details. Also more PHP documentation added to Doc/Manual/Php.html 01/27/2002:beazley The ANSI C size_t type is now recognized as an integer by default. 01/26/2002:beazley long long and unsigned long long support added to many language modules. This is not a portable feature and will require compiler support for the long long type. In target languages that do not support long long (e.g., Tcl and Perl), numbers are converted to a string of digits. This prevents their use in arithmetic calculations, but still allows values to be set from a string. long long support requires the use of the strtoll() and strtoull() functions as well as the 'lld' and 'llu' format specifiers of sprintf(). 01/26/2002:beazley Fixed [ #501827 ] Delete method is not called. The Tcl module wasn't correctly calling destructors when they were defined using %addmethods. This has been fixed. Reported by Reinhard Fobbe. 01/26/2002: beazley Better support for long long and unsigned long long. Typemaps have been included in a number of modules for handling these types. In addition, the parser has been modified to accept long long literals such as 1234LL and 1234ULL. 01/27/2002: cheetah (william fulton) [Java] A C char[] is mapped to a Java String which is the default SWIG handling of char[] and char*. It used to be mapped to byte[]. Note that a C signed char[] array is mapped to byte[]. *** POTENTIAL INCOMPATIBILITY *** 01/25/2002: beazley Fixed a problem with return-by-value, C++, and objects that define no default constructor. Reported by Joel Reed. 01/25/2002: cheetah (william fulton) [Java] Overhaul of the Java module. The C code generation is now done from typemaps. 01/24/2002: cheetah (william fulton) [Java] Support for arrays of enum pointers 01/20/2002: cheetah (william fulton) [Java] Error checking for null Java objects being passed to native functions. Exception thrown now whereas before the JVM crashed. 01/18/2002: cheetah (william fulton) [Java] Corrected behaviour for functions that take arrays. For example, when this c function: void arrayfn(int array[]); is wrapped the corresponding native function public final static native void arrayfn(int[] array); is produced. Previously if the C function made any changes to the array elements, these were not reflected back into the Java array. This has now been corrected so that the changes are propogated back to Java and the calling function will see these changes. This is how pure Java functions work, ie arrays are passed by reference. 01/15/2002:mkoeppe [Guile] New file cplusplus.i with C++ typemaps contributed by Marcio Luis Teixeira . 01/11/2002: cheetah (william fulton) [Java] Changed mapping of C long to Java type. Was mapped to Java long, now mapped to Java int. If you want the previous mapping to Java long use this approach in your interface file: %clear long; %typemap(jni) long "jlong" %typemap(jtype) long "long" %typemap(jstype) long "long" %clear long[ANY]; %typemap(jni) long[ANY] "jlongArray" %typemap(jtype) long[ANY] "long[]" %typemap(jstype) long[ANY] "long[]" %typemap(in) long[ANY] {write me for array support} %typemap(out) long[ANY] {write me for array support} %typemap(argout) long[ANY] {write me for array support} %typemap(freearg) long[ANY] {write me for array support} *** POTENTIAL INCOMPATIBILITY *** This new mapping is more appropriate when interfacing to 32 bit applications which are used in the current 32-bit JVMs. For future 64-bit JVMs you may have to change these mappings - eg on Unix LP64 systems, but not on Microsoft 64bit Windows which will be using a P64 IL32 model. This may be automated in a future version of SWIG. 01/10/2002:beazley Fixed [ 501677 ] %init block in wrong place. Reported by Luigi Ballabio. 01/09/2002: cheetah (william fulton) [Java] Default support for the long long type. signed long long is mapped to a Java long. unsigned long long is mapped to BigInteger. 01/09/2002:beazley Experimental change to parser to better support mixing of int, long, short, unsigned, float, and double. The parser should now support types like this: short unsigned int int unsigned short unsigned short int unsigned int short This change also enables a type of 'long double' (previously unsupported) to be used. *** NEW FEATURE *** 01/05/2002: cheetah (william fulton) [Java] Casting fix for when function return type is a pointer as reported by Gary Pennington 2002-01-05. The upper 32bits of the 64 bit jlong will have contained junk for 32bit pointers. 01/05/2002: cheetah (william fulton) [Java] Better pointer handling in Java is possible as the INPUT, OUTPUT and INOUT typemaps have been added into typemaps.i. 01/05/2002: cheetah (william fulton) [Java] $null can be used in input typemaps to return early from JNI functions that have either void or a non-void return type. Example: %typemap(check) int * %{ if (error) { SWIG_exception(SWIG_IndexError, "Array element error"); return $null; } %} If the typemap gets put into a function with void as return, $null will expand to nothing: void jni_fn(...) { if (error) { SWIG_exception(SWIG_IndexError, "Array element error"); return ; } ... } otherwise $null expands to zero, where javareturntype is either a pointer or a primitive type: javareturntype jni_fn(...) { if (error) { SWIG_exception(SWIG_IndexError, "Array element error"); return 0; } ... } 01/02/2002: cheetah (william fulton) [Java] The Java module incorrectly used argout typemaps for strings. This is now corrected and the code now resides in the freearg typemap. The argout array typemaps have been split into argout and freearg typemaps. This correction may require some user written typemaps to be modified. *** POTENTIAL INCOMPATIBILITY *** 12/28/2001: cheetah (william fulton) [Java] Multi typemaps now working for Java see multimap example. [Java] Fix for recently introduced bug - freearg typemap code was appearing before the function call. 12/28/2001: cheetah (william fulton) [Java] JCALL macro for JNI calls that work in both C and C++ typemaps have been replaced with JCALL0, JCALL1, JCALL2, JCALL3 and JCALL4 macros. *** POTENTIAL INCOMPATIBILITY *** 12/22/2001:beazley Resolved some inconsistent behavior with %rename and class renaming. If you specify the following: %rename(Foo) Bar; class Bar { public: Bar(); ~Bar(); } Then the %rename directive applies to the class itself, the constructor, and the destructor (all will be renamed to Foo). If a class defines more than one constructor, the overloaded variants can still be renamed by specifying parameters to %rename. For example: %rename(Bar_copy) Bar(Bar &); class Bar { public: Bar(); Bar(Bar &); ~Bar(); }; There are still some odd corner cases. If you specify %rename(Foo) ::Bar; then only the name of the class is changed and the constructor/destructor names are left unmodified. If you specify %rename(Foo) *::Bar; then the names of the constructor/destructor functions are modified but the name of the class is not. 12/21/2001: cheetah (william fulton) [Java] jni, jtype and jstype typemaps no longer hardcoded but real typemaps. New variable substitution, $javaclassname, can be used in the jstype typemaps. It is replaced with the Java shadow class name where applicable. [Java] Fix for recently introduced bug to do with inheritance when using %import. [Java] A few more bug fixes, todo with %rename and using the kind with the type, eg void fn(union uni myuni, struct str mystr, class cl mycl); 12/20/2001:beazley Fixed [ #494524 ] Preprocessor bug - apostrophe and #subst. 12/20/2001:beazley Added SWIG_VERSION preprocessor symbol. This is a hexadecimal integer such as 0x010311 (corresponding to SWIG-1.3.11). This can be used in the interface as follows: #if SWIG_VERSION >= 0x010311 /* Use some fancy new feature */ #endif Note: The version symbol is not defined in the generated SWIG wrapper file. *** NEW FEATURE *** 12/20/2001:mkoeppe [MzScheme]: Renamed mzswig_make_boolean to swig_make_boolean, as the latter is used in the typemaps. Reported by Luigi Ballabio. 12/17/2001:mkoeppe [Guile]: Rewrote list-vector.i using multi-dispatch typemaps. Updated pointer-in-out.i. Make the deprecated typemap-substitution of "$source" in "argout" work as before. 12/16/2001:mkoeppe [Guile]: Fixed macros %values_as_list, %values_as_vector, %multiple_values to use the proper %pragma syntax. New Guile example/test "multivalue"; new Guile run-test for test-suite item "list-vector" (currently broken). 12/14/2001:mkoeppe [Guile]: Fixed typemap-substition bug for "varin". Relaxed valid-identifier check to allow all R5RS identifiers. Version 1.3.10 (December 10, 2001) ================================== 12/08/2001:beazley Modified %typemap so that %{ ... %} can also be used as a code block (mostly for completeness). For example: %typemap(in) blah %{ ... %} This form does not introduce a new block scope. Also, the code enclosed in %{ ... %} is not processed by the preprocessor. 12/08/2001:beazley Fixed [ #459614 ] SWIG with multiple TCL interpreters. 12/08/2001:beazley Fixed [ #417141 ] rubydec.swg is wrong Reported by Paul Brannan. 12/08/2001:beazley Fixed [ #410557 ] Problem with %addmethods on NT. Reported by Magnus Ljung. 12/08/2001:beazley Fixed [ #445233 ] Enhancement: handle access change. SWIG now parses (but ignores) C++ access changes for the the following: class A { protected: void something() { } public: A() {} }; class B : private A { public: B() : A() { } protected: A::something; <---- Parsed, but ignored }; Suggested by Krzysztof Kozminski. 12/08/2001: cheetah (william fulton) Fix for Ruby to work using Visual C++. 12/06/2001:beazley Fixed [ #465687 ] unsigned short parameters fail. Reported by Gerald Williams. 12/06/2001:beazley Fixed SF [ #489594 ] PyString_FromString can't take NULL arg. Reported by John Merritt. SWIG now converts string values to Python using code like this: resultobj = result ? PyString_FromString(result) : Py_BuildValue(""); 12/06/2001:beazley Fixed SF [ #463561 ] Type conversions not generated. Reported by Gerald Williams. 12/04/2001:beazley Fixed SF [ #470217 ] Tcl default argument handling. Reported by Shaun Lowry. 12/04/2001:beazley Fixed SF [ #472088 ] defined(MACRO) expanded everywhere. Embedded preprocessor directives such as %#if defined(FOO) are not expanded by the SWIG preprocessor. Reported by Gerald Williams. 12/04/2001:beazley Fixed SF [ #476467 ] Problems with #define & commas. 12/04/2001:beazley Fixed SF [ #477547 ] wrong declaration of pointer functions. Bad prototypes in Lib/tcl/ptrlang.i. 12/04/2001:beazley Fixed SF [ #483182 ] Constants can take args by mistake. When swig -perl5 -const is used, constants are declared with a void prototype. For example: sub ICONST () { $examplec::ICONST } Patch submitted by Rich Wales. 12/03/2001:beazley New %exception directive. This is intended to replace %except. It works in exactly the same manner except it does not accept a language specifier. For example: %exception { try { $action } catch(SomeError) { error } } %exception is also name aware---allowing it to be applied to specific declarations in an interface. For example: %exception foo { ... exception for any function/method foo ... } %exception Foo::bar { ... exception for method bar in class Foo ... } %exception Foo::bar(double) { ... exception for method bar(double) in class Foo ... } The semantics of this name matching is exactly the same as for %rename. *** NEW FEATURE *** 12/03/2001:beazley Substantial cleanup of the Python shadow class code. Shadow classes used to be created in this rather complicated manner involving about a half-dozen strings created in bits and pieces. Shadow classes are now generated in a more straightforward manner--in the same order that appears in the interface file. *** POTENTIAL INCOMPATIBILITY *** The order in which declarations appear in the shadow file may differ. 12/03/2001:beazley The %insert directive (%{ ... %}, %runtime, %header, %wrapper, etc.) can now be used inside of a class definition. This has potential uses when generating shadow class code. For example: class Foo { ... %insert("shadow") %{ # Some python code def blah(self): print "I'm blah!" %} ... }; The support for class code insertion depends on the language module. However, the intent of this feature is to simplify the task of extending shadow class code. In the Python module, this inserts code with the proper level of indendation (regardless of what was used in the SWIG interface). *** NEW FEATURE *** 11/29/2001: cheetah (william fulton) Modifications for Java and Python modules to work on cygwin. Unfortunately a lot of the python module has started to produces code which cannot be auto-imported using cygwin libtools so most of it is still broken. 11/28/2001:beazley The %rename and %feature directive can now be used inside of a class definition. For example: class Foo { %rename(foo_i) foo(int); %rename(foo_d) foo(double); public: ... void foo(int); void foo(double); ... }; When used in this manner, the %rename directive only applies to members of the class in which it appears as well as all derived classes. In fact, this is really just the same as saying: %rename(foo_i) Foo::foo(int); %rename(foo_d) Foo::foo(double); class Foo { ... }; *** NEW FEATURE *** 11/26/2001:beazley Added the experimental %feature directive. %feature can be used to attach arbitrary string attributes to parse tree nodes. For example: %feature("except") blah { try { $function } catch (Error) { whatever; } } or %feature("set") *::x_set "x"; or %feature("blah") Foo::bar(int,double) const "spam"; The syntax is borrowed from the %rename directive. In fact, the exact same semantics apply (inheritance, matching, etc.). %feature is a very powerful low-level primitive that can be used to customize individual language modules and to provide hints to any stage of code generation. Features are attached to parse tree nodes as attributes with names like "feature:*" where * is replaced by the feature name (e.g., "feature:except", "feature:set", etc.). Language modules can then look for the features using a simple attribute lookup. %feature is intended to be a replacement for a number of older SWIG directives including %except and specialized pragmas. It is more powerful (due to its parameterized name matching) and it provides very precise control over how customization features are attached to individual declarations. There are future expansion plans that will build upon this capability as well. It's not certain that %feature will ever be used directly by SWIG users. Instead, it may be a low-level primitive that is used in high-level macro definitions. For instance, to support properties, you might define a macro like this: %define %property(name, setf, getf) %feature("set") setf #name; %feature("get") getf #name; %enddef Which allows a user to specify things like this: %property(p, get_p, set_p); class Blah { public: int get_p(); void set_p(int); }; *** EXPERIMENTAL NEW FEATURE *** 11/24/2001:beazley The Tcl module has been expanded with some new features for managing object ownership. For example: set c [Circle -args 20] $c area # Invoke a method $c -disown # Releases ownership of the object $c -acquire # Acquires ownership of the object If Tcl owns the object, its destructor is invoked when the corresponding object command is deleted in Tcl. To simplify the destruction of objects, the following syntax can be used: $c -delete # Delete an object This is an alternative for the more obscure variant of rename $c {} These features also add functionality at the C API level. The following functions manage ownership from C and can be used in typemaps. SWIG_Acquire(void *ptr); SWIG_Disown(void *ptr); A new function for constructing instances is also available: Tcl_Obj * SWIG_NewInstanceObj(Tcl_Interp *interp, void *ptr, swig_type_info *type, int own); When used in a typemap, this creates a pointer object and an interpreter command that can be used to issue methods and access attributes as shown above. *** NEW FEATURE *** 11/23/2001:beazley All Python-related %pragma operations have been eliminated. Most of these were written for older SWIG versions in order to compensate for limitations in earlier releases. In an effort to reduce the amount of code-clutter and potential for errors, it is easier to simply eliminate the pragmas and to start over (if needed). To be honest, I'm not even sure the pragmas worked in 1.3.9 and recent releases. Note: If you need to insert code into the shadow class file created by SWIG, simply use the %shadow directive like this: %shadow %{ def some_python_code(): print "blah!" %} *** POTENTIAL INCOMPATIBILITY *** 11/22/2001:beazley Sweeping changes to the way in which the Python module handles shadow classes. In early implementations, shadow classes were merely Python wrappers around typed pointer objects. However, some users actually wanted to receive the shadow class object in C. To accomodate this, the dereferencing of the "this" pointer in a shadow class was moved to C as described in CHANGES [8/8/99]. However, the process of returning pointers to Python was still somewhat problematic. Specifically, shadow classes never worked in situations such as these: - Use of any kind of output typemap ('out' or 'argout') - Global variables (broken as far as I can tell). In the past, some users have dealt with this by manually trying to create shadow class objects themselves from C/C++. However, this was difficult because the C wrappers don't really know how to get access to the corresponding Python class. The Python module has now been modified to automatically attach shadow class objects to pointers when they are returned to Python. This process occurs in the function SWIG_NewPointerObj() so the process is completely transparent to users. As a result, shadow classes are now more seamlessly integrated with typemaps and other features of SWIG. This change may introduce a number of incompatibilities. The SWIG_NewPointerObj() now takes an extra parameter "own" to indicate object ownership. This can be used to return a pointer to Python that Python should destroy. In addition, older code that tries to manually construct shadow class objects or which expects bare pointers may break---such pointers may already be encapsulated by a shadow class. *** POTENTIAL INCOMPATIBILITY *** 11/20/2001:beazley Modified the %insert directive to accept single braces { ... }. For example: %insert("header") { ... some code ... } This works exactly like %{ ... %} except that the code in the braces is processed using the preprocessor. This can be useful in certain contexts such as low-level code generation in language modules. *** NEW FEATURE *** 11/20/2001:beazley Command line options are now translated into preprocessor symbols. For example: ./swig -python -shadow -module blah interface.i Creates the symbols: SWIGOPT_PYTHON 1 SWIGOPT_SHADOW 1 SWIGOPT_MODULE blah Modules can look for these symbols to alter their code generation if needed. *** NEW FEATURE *** 11/20/2001:beazley Massive overhaul of the Perl5 module. A lot of code generation is now driven by tables and typemaps. The generated wrapper code also makes use of tables to install constants, variables, and functions instead of inlining a bunch of procedure calls. The separate variable initialization function is gone. Most code generation is controlled via the perl5.swg file in the library. *** POTENTIAL INCOMPATIBILITY *** 11/13/2001:beazley Added parsing support for the C++ typename keyword. Primarily this is added to better support templates. For example: template void blah(C& v) { typename C::iterator i = v.begin(); } Note: typename is supported in the parser in the same way as 'struct' or 'class'. You probably shouldn't use it anywhere except in templates. *** NEW FEATURE *** 11/11/2001:beazley Massive overhaul of the language module API. Most functions now use a common, very simple, API. There are also a number of interesting semantic side-effects of how code is actually generated. Details will be forthcoming in Doc/Manual/Extending.html. *** POTENTIAL INCOMPATIBILITY *** Language modules written for previous versions of SWIG will no longer work, 11/10/2001:beazley Fixed a very subtle bug due to unnamed class wrapping. For example, if you did this typedef struct { int x,y; } gdPoint, *gdPointPtr; void foo(gdPointPtr x); Then the foo function would get a type-error. The problem has to do with internal typedef handling and the fact that the typedef declarations after the struct appear later in the parse tree. It should work now. Problem reported by Vin Jovanovic. 11/09/2001:beazley Subtle change to "out" typemaps (and related variations). The name that is attached to the typemap is now the raw C identifier that appears on a declaration. This changes the behavior of member functions. For example: %typemap(out) int foo { ... } class Blah { public: int foo(); // typemap gets applied } Previous versions never really specified how this was supposed to work. In SWIG1.1, you could probably write a typemap for the wrapper name like this: %typemap(out) int Blah_foo { ... } However, this old behavior is now withdrawn and not supported. Just use the member name without any sort of special prefix. *** POTENTIAL INCOMPATIBILITY *** 11/06/2001:beazley Changes to Tcl module initialization: (1) SWIG now automatically includes the code needed to work with Tcl stubs. Simply compile with -DUSE_TCL_STUBS. (2) SWIG now automatically calls Tcl_PkgProvide to register a package name. The package name is the same as the name specified with the %module directive. The version number is set to "0.0" by default. To change the version number, use swig -pkgversion 1.2 interface.i. *** POTENTIAL INCOMPATIBILITY *** Modules that provided stubs and Tcl_PkgProvide on their own might break. Simply remove that code. 11/05/2001:beazley Changed code generation of constants in the Tcl module. Constants are now stored in a large table that get installed at module startup. There are also no longer any static variables so it should generate somewhat less code. 11/04/2001:beazley The "const" typemap has been renamed to "constant" in many language modules. "const" is a C keyword which made the handling of the typemap directive somewhat awkward in the parser. *** POTENTIAL INCOMPATIBILITY *** 11/04/2001:beazley %typemap directive can now accept nearly arbitrary keyword parameters. For example: %typemap(in,parse="i",doc="integer") int "..."; The purpose of the keyword parameters is to supply code generation hints to the target language module. The intepretation of the parameters is language specific. *** NEW FEATURE *** 11/04/2001:beazley Slight semantic change to internal call/return by value handling. In previous versions of SWIG, call-by-value was translated into pointers. For example: double dot_product(Vector a, Vector b); turned into this: double wrap_dot_product(Vector *a, Vector *b) { return dot_product(*a,*b); } This translation was normally performed by the SWIG core, outside of the control of language modules. However, a side effect of this was a lot of bizarre typemap behavior. For example, if you did something like this: %typemap(in) int32 { ... } You would find that int32 was transformed into a pointer everywhere! (needless to say, such behavior is unexpected and quite awkward to deal with). To make matters worse, if a typedef was also used, the pointer behavior suddenly disappeared. To fix this, the pointer transformation is now pushed to the language modules. This produces wrappers that look roughly like this: double wrap_dot_product(Vector *a, Vector *b) { Vector arg1 = *a; Vector arg2 = *b; return dot_product(arg1,arg2); } This change also makes it easy to define typemaps for arbitrary undefined types. For example, you can do this (and it will work regardless what int32 is): %typemap(in) int32 { $1 = (int32) PyInt_AsLong($input); } *** POTENTIAL IMCOMPATIBILITY *** This change may break call/return by value code generation in some language modules. 11/03/2001:beazley Changed the name of the default typemaps to the following: %typemap() SWIGTYPE { ... an object ... } %typemap() SWIGTYPE * { ... a pointer ... } %typemap() SWIGTYPE & { ... a reference ... } %typemap() SWIGTYPE [] { ... an array ... } %typemap() enum SWIGTYPE { ... an enum value ... } %typemap() SWIGTYPE (CLASS::*) { ... pointer to member ... } These types are used as the default for all types that don't match anything else. See CHANGES log entry for 8/27/2000 for the old behavior. The role of these types is also described in Doc/Manual/Typemaps.html *** POTENTIAL INCOMPATIBILITY *** 10/25/2001:beazley Modified Guile and Mzscheme modules to support multi-argument typemaps. 10/25/2001: cheetah (william fulton) [Java] Fix to handle pointers to arrays. 10/24/2001:beazley Defining a typemap rule for enum SWIGENUM can now be used to define default behavior for enum variables. 10/22/2001:beazley Ruby module modified to support multi-argument typemaps. 10/22/2001:beazley The Ruby module can now handle functions with an arbitrary number of arguments. Previous versions were limited to to functions with only 9 or 16 arguments depending on the use of default arguments. Note: from some inspection of the Ruby interpreter source, the new approach might be a little faster as well. 10/18/2001:beazley Fixed a bug with forward class declarations and templates. class Foo ; Bug reported by Irina Kotlova. 10/16/2001:beazley Support for multivalued typemaps added. The typemaps are specified using the syntax below. Within each typemap, variable substitution is handled as follows: %typemap(in) (int argc, char *argv[]) { $arg; // The input object in the target language $1; // C local variable for first argument $2; // C local variable for second argument // These variables refer to either argument $1_type, $1_ltype, $1_basetype, etc... (argc) $2_type, $2_ltype, $2_basetype, etc... (argv[]) // Array dimension of argv $2_dim0 } Basically any variable that was available in normal typemaps is available for either argument by prefacing the variable name by '$n_' where n is the argument position. Notes: (1) Multi-valued typemaps can only be applied to a single object in the target scripting language. For example, you can split a string into a (char *, int) pair or split a list into a (int, char []) pair. It is not possible to map multiple objects to multiple arguments. (2) To maintain compatibility with older SWIG versions, the variables such as $target and $type are preserved and are mapped onto the first argument only. (3) This should not affect compatibility with older code. Multi-valued typemaps are an extension to typemap handling. Single valued typemaps can be specified in the usual way. The old $source and $target variables are officially deprecated. Input variables are referenced through $arg$ and output values are reference through $result$. *** NEW FEATURE *** 10/16/2001:beazley Added parsing support for multivalued typemaps. The syntax is a little funky, but here goes: // Define a multivalued typemap %typemap(in) (int argc, char *argv[]) { ... typemap code ... } // Multivalued typemap with locals %typemap(in) (int argc, char *argv[])(int temp) { ... typemap code ... } // Copy a multivalued typemap %typemap(in) (int argcount, char **argv) = (int argc, char *argv[]); // Apply a multivalued typemap %apply (int argc, char *argv[]) { (int argcount, char **argv) }; Note: this extra parsing support is added for future extension. No language modules currently support multi-valued typemaps. 10/11/2001:beazley Modified the typemap matching code to discard qualifiers when checking for a match. For example, if you have a declaration like this: void blah(const char *x); The typemap checker checks for a match in the following order: const char *x const char * char *x char * If typedef's are involved, qualifier stripping occurs before typedef resolution. So if you had this, typedef char *string; void blah(const string x); typemap checking would be as follows: const string x const string string x string const char *x const char * char *x char * The primary reason for this change is to simplify the implementation of language modules. Without qualifier stripping, one has to write seperate typemaps for all variations of const and volatile (which is a pain). *** POTENTIAL INCOMPATIBILITY *** Typemaps might be applied in places where they weren't before. 10/9/2001: beazley SWIG now generates wrappers that properly disambiguate overloaded methods that only vary in constness. For example: class Foo { ... void blah(); void blah() const; ... }; To handle this, the %rename directive can be used normally. %rename(blah_const) blah() const; In the resulting wrapper code, method calls like this are now generated: (obj)->blah() // Non-const version ((Foo const *)obj)->blah() // const version This should force the right method to be invoked. Admittedly, this is probably obscure, but we might as well get it right. 10/8/2001: beazley The preprocessor now ignores '\r' in the input. This should fix the following bug: [ #468416 ] SWIG thinks macro defs are declarations? 10/8/2001: beazley Added support for ||, &&, and ! in constants. This fixes SF [ #468988 ] Logical ops break preprocessor. However, at this time, constants using these operators are not supported (the parser will issue a warning). 10/4/2001: beazley Added -show_templates command line option. This makes SWIG display the code it actually parses to generate template wrappers. Mostly useful for debugging. *** NEW FEATURE *** 10/4/2001: beazley Change to semantics of %template directive. When using %template, the template arguments are handled as types by default. For example: %template(vecint) vector; %template(vecdouble) vector; To specify a template argument that is *not* a type, you need to use default-value syntax. For example: %template(vecint) vector; %template(vecdouble) vector; In this case, the type name doesn't really matter--only the default value (e.g., 50, 100) is used during expansion. This differs from normal C++, but I couldn't figure out a better way to do it in the parser. Might implement an alternative later. *** POTENTIAL INCOMPATIBILITY *** 10/4/2001: beazley Major changes to template handling in order to provide better integration with the C++ type-system. The main problem is as follows: Suppose you have a template like this: template void blah(const T x) { stuff }; Now suppose, that you instantiate the template on a type like this in SWIG: %template(blahint) blah; In C++, this is *supposed* to generate code like this: void blah(int *const x) { stuff }; However, in SWIG-1.3.9, the template substitution gets it wrong and produces void blah(const int *x) { stuff }; (notice the bad placement of the 'const' qualifier). To fix this, the SWIG parser now generates implicit typedefs for template type arguments that produces code roughly equivalent to doing this: typedef int *__swigtmpl1; %template(blahint) blah<__swigtmpl1>; which generates code like this: void blah(const __swigtmpl1 x) { stuff }; Since this is correct in both C++ and SWIG, it provides the right semantics and allows everything to compile properly. However, to clean up the generated code a little bit, the parser keeps track of the template types and performs back-substitution to the original type when building the parse tree. Thus, even though the implicit typedef is used in the input and may appear in the generated wrapper file (for proper compilation), the parse tree will hide a lot of these details. For example: void blah(const __swigtmpl1 x) { stuff }; will look like it was declared as follows (which is what you want): void blah(int *const x) { stuff } The only place you are likely to notice the typedef hack is in bodies of template functions. For example, if you did this, template class blah { ... %addmethods { void spam() { T tempvalue; ... } } } you will find that 'T tempvalue' got expanded into some strange typedef type. This *still* compiles correctly so it's not a big deal (other than looking kind of ugly in the wrapper file). 10/4/2001: beazley Fixed some inheritance problems in Tcl Object interface. 10/1/2001: beazley Tcl module has changed to use byte-backed pointer strings. This implementation should be safe on 64-bit platforms. However, the order in which digits appear in pointer values no longer directly corresponds to the actual numerical value of a pointer (on little-endian machines, pairs of digits appear in reverse order). 10/1/2001: beazley Perl5 module is now driven by a configuration file 'perl5.swg' in the SWIG library. 10/1/2001: beazley The perl5 module no longer tries to apply the "out" typemap in code generated for magic variables. I'm surprised that this ever worked at all (since all of the code that was there was wrong anyways). Use the "varout" typemap to handle global variables. 10/1/2001: beazley Fixed a bug related to character array members of structures. For example: struct Foo { char name[32]; }; SWIG is normally supposed to return a string, but this was broken in 1.3.9. The reason it was broken was actually due to a subtle new feature of typemaps. When a data member is set to an array like this, the return type of the related accessor function is actually set to an array. This means that you can now write typemaps like this: %typemap(python,out) char [ANY] { $target = PyString_FromStringAndSize($source,$dim0); } This functionality can be used to replace the defunct memberout typemap in a more elegant manner. 9/29/2001: beazley Some further refinement of qualified C++ member functions. For example: class Foo { ... void foo() const; ... }; (i) The SWIG parser was extended slightly to allow 'volatile' and combinations of 'const' and 'volatile' to be used. This is probably rare, but technically legal. Only added for completeness. (ii) For the purposes of overloading, qualified and non-qualified functions are different. Thus, when a class has methods like this: void foo(); void foo() const; Two distinct methods are declared. To deal with this, %rename and similar directives have been extended to recognize const. Thus, one can disambiguate the two functions like this: %rename(fooconst) Foo::foo() const; or simply ignore the const variant like this: %ignore Foo::foo() const; Note: SWIG currently has no way to actually invoke the const member since the 'const' is discarded when generating wrappers for objects. 9/27/2001: beazley New directive. %namewarn can be used to issue warning messages for certain declaration names. The name matching is the same as for the %rename directive. The intent of this directive is to issue warnings for possible namespace conflicts. For example: %namewarn("print is a python keyword") print; The name matching algorithm is performed after a name has been resolved using %rename. Therefore, a declaration like this will not generate a warning: %rename("Print") print; ... void print(); /* No warning generated */ Since the warning mechanism follows %rename semantics, it is also to issue warnings for specific classes or just for certain member function names. (Dave - I've been thinking about adding something like this for quite some time. Just never got around to it) *** NEW FEATURE *** 9/27/2001: beazley Enhanced the %ignore directive so that warning messages can be issued to users. This is done using %ignorewarn like this: %ignorewarn("operator new ignored") operator new; The names and semantics of %ignorewarn is exactly the same as %ignore. The primary purpose of this directive is for module writers who want to ignore certain types of declarations, but who also want to alert users about it. A user might also use this for debugging (since messages will appear whenever an ignored declaration appears). *** NEW FEATURE *** 9/26/2001: beazley Super-experimental support for overloaded operators. This implementation consists of a few different parts. (i) Operator names such as 'operator+' are now allowed as valid declarator names. Thus the 'operator' syntax can appear *anyplace* a normal declarator name was used before. On the surface, this means that operators can be parsed just like normal functions and methods. However, it also means that operator names can be used in many other SWIG directives like %rename. For example: %rename(__add__) Complex::operator+(const Complex &); (ii) Operators are wrapped *exactly* like normal functions and methods. Internally, the operator name is used directly meaning that the wrapper code might contain statements like this: arg0->operator*((Complex const &)*arg1); This all seems to parse and compile correctly (at least on my machine). (iii) SWIG will no longer wrap a declaration if its symbol table name contains illegal identifier characters. If illegal characters are detected, you will see an error like this: Warning. Can't wrap operator* unless renamed to a valid identifier. The only way to fix this is to use %rename or %name to bind the operator to a nice name like "add" or something. Note: the legal identifier characters are determined by the target language. There are certain issues with friend functions and operators. Sometimes, friends are used to define mixed operators such as adding a Complex and a double together. Currently, SWIG ignores all friend declarations in a class. A global operator declaration can probably be made to work, but you'll have to rename it and it probably won't work very cleanly in the target language since it's not a class member. SWIG doesn't know how to handle operator specifications sometimes used for automatic type conversion. For example: class String { ... operator const char*(); ... }; (this doesn't parse correctly and generates a syntax error). Also: operators no longer show up as separate parse-tree nodes (instead they are normal 'cdecl' nodes). I may separate them as a special case later. See Examples/python/operator for an example. *** SUPER-EXPERIMENTAL NEW FEATURE *** Version 1.3.9 (September 25, 2001) ================================== 9/25/2001: beazley Fixed parsing problem with type declarations like 'char ** const'. SWIG parsed this correctly, but the internal type was represented incorrectly (the pointers and qualifiers were in the wrong order). 9/25/2001: beazley Withdrew experimental feature (noted below) that was causing serious parsing problems. Version 1.3.8 (September 23, 2001) ================================== 9/23/2001: beazley Included improved distutils setup.py file in the Tools directory (look for the setup.py.tmpl file). Contributed by Tony Seward. 9/23/2001: beazley Included two new RPM spec files in the Tools directory. Contributed by Tony Seward and Uwe Steinmann. 9/21/2001: beazley Fixed SF Bug [ #463635 ] Perl5.swg does not compile in Visual C++ 9/21/2001: beazley Two new directives control the creation of default constructors and destructors: %nodefault %makedefault These replace %pragma nodefault and %pragma makedefault. (old code will still work, but documentation will only describe the new directives). 9/21/2001: beazley Fixed SF Bug [ #462354 ] %import broken in 1.3.7. 9/20/2001: beazley Parser modified to ignore out-of-class constructor and destructor declarations. For example: inline Foo::Foo() : Bar("foo") { } inline Foo::~Foo() { } Suggested by Jason Stewart. *** EXPERIMENTAL FEATURE *** 9/20/2001: beazley Modified the parser to ignore forward template class declarations. For example: template class MapIter; Suggested by an email example from Irina Kotlova. 9/20/2001: beazley Fixed problem with undeclared tcl_result variable in the "out" typemap for Tcl. Reported by Shaun Lowry. 9/20/2001: beazley Incorporated changes to make SWIG work with ActivePerl. Contributed by Joel Reed. 9/20/2001: beazley Slight change to the parsing of C++ constructor initializers. For example: class Foo : public Bar { public: Foo() : Bar(...) {...} }; SWIG now discards the contents of the (...) regardless of what might enclosed (even if syntactically wrong). SWIG doesn't need this information and there is no reason to needless add syntax rules to handle all of the possibilities here. 9/20/2001: beazley Change to typemaps for structure members. If you have a structure like this: struct Vector { int *bar; }; The member name 'bar' is now used in any accessor functions. This allows the "in" typemap to be used when setting the value. For example, this typemap %typemap(python,in) int *bar { ... } now matches Vector::bar. It should be noted that this will also match any function with an argument of "int *bar" (so you should be careful). *** NEW FEATURE. POTENTIAL INCOMPATIBILITY *** 9/20/2001: beazley Fixed SF bug #462642 setting string values in structures 9/20/2001: beazley Fixed SF bug #462398 problem with nested templates. 9/20/2001: beazley Fixed SF bug #461626 problem with formatting and C++ comments. 9/20/2001: beazley Fixed SF bug #462845 Wrong ownership of returned objects. 9/19/2001: beazley Fixed SF bug #459367. Default constructors for classes with pure virtual methods. 9/19/2001: beazley Fixed problem with default arguments and class scope. For example: class Foo { public: enum bar { FOO, BAR }; void blah(bar b = FOO); ... } SWIG now correctly generates a default value of "Foo::FOO" for the blah() method above. This used to work in 1.1, but was broken in 1.3.7. Bug reported by Mike Romberg. Version 1.3.7 (September 3, 2001) ================================== 9/02/2001: beazley Added special %ignore directive to ignore declarations. This feature works exactly like %rename. For example: %ignore foo; // Ignore all declarations foo %ignore ::foo; // Only ignore foo in global scope %ignore Spam::foo; // Only ignore in class Spam %ignore *::foo; // Ignore in all classes %ignore can also be parameterized. For example: %ignore foo(int); %ignore ::foo(int); %ignore Spam::foo(int); %ignore *::foo(int); *** NEW FEATURE *** 9/02/2001: cheetah (william fulton) [Java] shadowcode pragma modified so that the code that is output in the shadow file is placed relative to where it is placed in the c/c++ code. This allows support for JavaDoc function comments. 9/01/2001: beazley Fixed SF Patch [ #447791 ] Fix for python -interface option. Submitted by Tarn Weisner Burton. 9/01/2001: beazley SWIG no longer generates default constructors/destructors for a class if it only defines a private/protected constructor or destructor or if any one of its base classes only has private constructors/destructors. This was reported in SF Patch [ #444281 ] nonpublic/default/inhereted ctor/dtor by Marcelo Matus. 9/01/2001: beazley Added patch to Perl5 module that allows constants to be wrapped as constants that don't require the leading $. This feature is enabled using the -const option. Patch contributed by Rich Wales. *** NEW FEATURE *** 8/31/2001: beazley Added parsing support for the 'volatile' type qualifier. volatile doesn't mean anything to SWIG, but it is needed to properly generate prototypes for declarations that use it. It's also been added to make the SWIG type system more complete. *** NEW FEATURE *** 8/30/2001: beazley Added support for parameterized %rename directive. *** This new feature can be used to greatly simplify the task of resolving overloaded methods and functions. *** In prior versions of SWIG, the %rename directive was used to consistently apply an identifier renaming. For example, if you said this: %rename foo bar; Every occurrence of 'foo' would be renamed to 'bar'. Although this works fine for resolving a conflict with a target language reserved word, it is useless for for dealing with overloaded methods. This is because all methods are simply renamed to the same thing (generating the same conflict as before). Therefore, the only way to deal with overloaded methods was to go through and individually rename them all using %name. For example: class Foo { public: virtual void bar(void); %name(bar_i) virtual void bar(int); ... }; To make matters worse, you had to do this for all derived classes too. class Spam : public Foo { public: virtual void bar(void); %name(bar_i) virtual void bar(int); ... }; Needless to say, this makes it extremely hard to resolve overloading without a lot of work and makes it almost impossible to use SWIG on raw C++ .h files. To fix this, %rename now accepts parameter declarators. The syntax has also been changed slightly. For example, the following declaration renames all occurrences of 'bar(int)' to 'bar_i', leaving any other occurrence of 'bar' alone. %rename(bar_i) bar(int); Using this feature, you can now selectively rename certain declarations in advance. For example: %rename(bar_i) bar(int); %rename(bar_d) bar(double); // Include raw C++ header %include "header.h" When %rename is used in this manner, all occurrence of bar(int) are renamed wherever they might occur. More control is obtained through explicit qualification. For example, %rename(bar_i) ::bar(int); only applies the renaming if bar(int) is defined in the global scope. The declaration, %rename(bar_i) Foo::bar(int); applies the renaming if bar(int) is defined in a class Foo. This latter form also supports inheritance. Therefore, if you had a class like this: class Spam : public Foo { public: void bar(int); } The Spam::bar(int) method would also be renamed (since Spam is a subclass of Foo). This latter feature makes it easy for SWIG to apply a consistent renaming across an entire class hierarchy simply by specifying renaming rules for the base class. A class wildcard of * can be used if you want to renaming all matching members of all classes. For example: %rename(bar_i) *::bar(int); will rename all members bar(int) that are defined in classes. It will not renamed definitions of bar(int) in the global scope. The old use of %rename is still supported, but is somewhat enhanced. %rename(foo) bar; // Renames all occurrences of 'bar'. %rename(foo) ::bar; // Rename all 'bar' in global scope only. %rename(foo) *::bar; // Rename all 'bar' in classes only. %rename(foo) Foo::bar; // Rename all 'bar' defined in class Foo. *** NEW FEATURE *** 8/30/2001: beazley Added support for data-member to member-function transformation. For example, suppose you had a structure like this: struct Vector { double x,y; }; Now suppose that you wanted to access x and y through a member function interface instead of the usual SWIG behavior. For example: f.set_x(3.4) # instead of f.x = 3.4 x = f.get_x() # instead of x = f.x To do this, simply use the new %attributefunc directive. For example: %attributefunc(get_%s,set_%s) struct Vector { double x,y; }; %noattributefunc The arguments to %attributefunc are C-style printf format strings that determine the naming convention to use. %s is replaced with the actual name of the data member. SWIG provides a number of printf extensions that might help. For example, if you wanted to title case all of the attributes, you could do this: %attributefunc(get%(title)s,set%(title)s); This will turn an attribute 'bar' to 'getBar()' and 'setBar()'. (someone requested this long ago, but I finally figured how to implement it in a straightforward manner). *** EXPERIMENTAL NEW FEATURE *** 8/30/2001: beazley SWIG now automatically generates default constructors and destructors if none are defined. This used to be enabled with a command line switch -make_default, but most people want these functions anyways. To turn off this behavior use the -no_default option or include the following pragma in the interface file: %pragma no_default; This may break certain interfaces that defined their own constructors/destructors using the same naming convention as SWIG. If so, you will get duplicate symbols when compiling the SWIG wrapper file. *** POTENTIAL INCOMPATIBILITY *** 8/29/2001: beazley Changes to Perl5 shadow class code generation. Iterators are no longer supported (FIRSTKEY, NEXTKEY). Also, attribute access has been changed to rely on inheritance in order to provide better behavior across modules. 8/28/2001: beazley Various obscure improvements to the type system and classes. Strange declarations like this are now wrapped correctly (i.e., the generated wrapper code doesn't cause the C++ compiler to die with a type error). class Foo { public: typedef double Real; Real foo(Real (*op)(Real,Real), Real x, Real y); }; Inheritance of types is also handled correctly. 8/28/2001: beazley Changes to class wrappers. When SWIG sees two classes like this, class X { public: void foo(); ... } class Y : public X { public: void bar(); ... } it now only generates two wrapper functions: X_foo(X *x) { x->foo(); } Y_bar(Y *y) { y->bar(); } Unlike SWIG1.15, the foo() method does *not* propagate to a wrapper function Y_foo(). Instead, the base class method X_foo() must be used. This change should not affect modules that use shadow classes, but it might break modules that directly use the low-level C wrappers. This change is being made for a number of reasons: - It greatly simplifies the implementation of SWIG--especially with anticipated future changes such as overloaded methods. - It results in substantially less wrapper code--especially for big C++ class hierarchies (inherited declarations are no longer copied into every single derived class). - It allows for better code generation across multiple SWIG generated modules (code isn't replicated in every single module). *** POTENTIAL INCOMPATIBILITY *** 8/22/2001: cheetah (william fulton) Provided some Windows documentation in the Win directory and some Visual C++ project files for running examples on Windows. 8/28/2001: mkoeppe [Guile] Handle renamed overloaded functions properly; thanks to Marc Zonzon for the patch. See the new test case name_cxx. 8/27/2001: mkoeppe [Tcl] Removed lots of warnings issued by the Sun Forte compilers, which were caused by mixing function pointers of different linkages (C++/C). 8/23/2001: mkoeppe Improved the MzScheme module by porting Guile's pointer type checking system and making type dispatch typemap-driven. 8/22/2001: beazley Entirely new symbol table processing. SWIG should be able to report much better error messages for multiple declarations. Also, the new symbol table allows for overloaded functions (although overloading isn't quite supported in the language modules yet). 8/22/2001: cheetah (william fulton) * [Java] %new support added. * [Java] Package JNI name refixed! 8/19/2001: beazley Python module modified to support pointers to C++ members. This is an experimental feature. *** NEW FEATURE *** 8/19/2001: beazley Added limited parsing and full type-system support for pointers to members. None of SWIG's language modules really know how to deal with this so this is really only provided for completeness and future expansion. Note: SWIG does not support pointers to members which are themselves pointers to members, references to pointers to members, or other complicated declarations like this. *** NEW FEATURE *** 8/19/2001: beazley SWIG is much better at parsing certain C++ declarations. Operators and friends generally don't cause anymore syntax errors. However, neither are really supported. 8/18/2001: beazley Added *highly* experimental support for wrapping of C++ template declarations. Since C++ templates are essentially glorified macros and SWIG has a fully operational C preprocessor with macro support, the parser now converts template declarations to macros. For example, a function template like this template T max(T a, T b); is internally converted into a macro like this: %define %_template_max(__name,T) %name(__name) T max(T a, T b); %enddef To instantiate a version of the template, a special %template declaration is used like this: %template(maxint) max; %template(maxdouble) max; The parameter to the %template directive must be proper C identifier that's used to uniquely name the resulting instantiation. When used, the the expanded macro looks like this: %name(maxint) int max(int a, int b); %name(maxdouble) double max(double a, double b); A similar technique is used for template classes. For instance: template class vector { T *data; int sz; public: vector(int nitems); T *get(int n); ... }; Gets converted into a macro like this: %define %_template_vector(__name, T) %{ typedef vector __name; %} class __name { T *data; int sz; public: __name(int nitems); T *get(int n); ... }; typedef __name vector; %enddef An a specific instantiation is created in exactly the same way: %template(intvec) vector; The resulting code parsed by SWIG is then: %{ typedef vector intvec; %} class intvec { int *data; int sz; public: intvec(int nitems); int *get(int n); ... }; typedef intvec vector; Note: the last typedef is non-standard C and is used by SWIG to provide an association between the name "intvec" and the template type "vector". CAUTION: This is an experimental feature and the first time SWIG has supported C++ templates. Error reporting is essential non-existent. It will probably break in certain cases. *** EXPERIMENTAL NEW FEATURE **** 8/15/2001: beazley Change to wrapping of multi-dimensional arrays. Arrays are now properly mapped to a pointer to an array of one less dimension. For example: int [10]; --> int * int [10][20]; --> int (*)[20]; int [10][20][30]; --> int (*)[20][30]; This change may break certain SWIG extensions because older versions simply mapped all arrays into a single pointer such as "int *". Although possibly unusual, the new version is correct in terms of the C type system. *** POTENTIAL INCOMPATIBILITY *** 8/06/2001: cheetah (william fulton) * [Java] Array setters generated for struct/class array members. 8/13/2001: beazley Many improvements to Tcl/Perl/Python modules to better work with multiple interface files and the %import directive. 8/13/2001: beazley Fixed up the behavior of %import in the Python module. SWIG no longer pollutes the module namespace by using 'from module import *' to refer to the other module. Instead, it does a proper 'import module'. Also, SWIG may work a lot better when importing modules that include references to other imported modules. 8/13/2001: mkoeppe Added new typemap substitutions, generalizing those of the Guile-specific 5/27/2001 changes: * $descriptor is the same as SWIGTYPE$mangle, but also ensures that the type descriptor of this name gets defined. * $*type, $*ltype, $*mangle, $*descriptor are the same as the variants without star, but they REMOVE one level of pointers from the type. (This is only valid for pointer types.) * $&type, $<ype, $&mangle, $&descriptor are the same as the variants without ampersand, but they ADD one level of pointers to the type. The Guile-specific substitution $basedescriptor was removed because it was useless. 8/12/2001: beazley The %extern directive is now deprecated and withdrawn. The purpose of this directive was to import selected definitions from other interface files and headers. However, the same functionality is better handled through %import. This leaves SWIG with two file inclusion directives: %include filename - Inserts into current interface %import filename - Import types and classes from another module *** POTENTIAL INCOMPATIBILITY *** 8/09/2001: beazley Added new support for wrapping C/C++ callback functions. A common problem with some C libraries is that many functions take a function pointer as an argument. For example: int do_op(..., int (*op)(int,int), ...); Unfortunately, the only way to call such a function is to pass it a function pointer of some compatible type. In previous versions of SWIG, you had to solve this problem with some really gross hacks. For example, if you wanted to use the following function as a callback, int foo(int, int); you had to install a pointer to it as a constant. For example: %constant int (*FOO)(int,int) = foo; or const int (*FOO)(int,int) = foo; or if you had a really old SWIG version: typedef int (*OP_FUNC)(int,int); int do_op(..., OP_FUNC, ...); const OP_FUNC FOO = foo; Now, you can do one of two things: %constant int foo(int,int); This creates a constant 'foo' of type int (*)(int,int). Alternatively, you can do this: %callback("%s") int foo(int,int); int bar(int,int); %nocallback In this case, the functions are installed as constants where the name is defined by the format string given to %callback(). If the names generated by the format string differ from the actual function name, both a function wrapper and a callback constant are created. For example: %callback("%(upper)s") int foo(int,int); int bar(int,int); %nocallback Creates two wrapper functions 'foo', 'bar' and additionally creates two callback constants 'FOO', 'BAR'. Note: SWIG still does not provide automatic support for writing callback functions in the target language. *** NEW FEATURE *** 8/06/2001: cheetah (william fulton) * struct nesting fixes as per SF bug #447488. 8/03/2001: beazley The %name directive now applies to constants created with #define and %constant. However, most language modules were never written to support this and will have to be modified to make it work. Tcl, Python, and Perl modules are working now. *** NEW FEATURE *** 8/03/2001: beazley Massive changes and simplification of C declaration parsing. Although SWIG is still not a full C parser, its ability to handle complex datatypes including pointers to functions and pointers to arrays has been vastly improved. 8/03/2001: cheetah (william fulton) * Distribution fixes: autoconf no longer needed to install SWIG. 8/02/2001: beazley Removed two undocumented parsing features. SWIG no longer supports out-of-class static function or variable declarations. For example: static int Foo::bar; This feature may return if there is sufficient demand. However, since SWIG is most often used with header files, it is more likely for these definitions to be included in the class definition. *** POTENTIAL INCOMPATIBILITY *** 8/02/2001: cheetah (william fulton) * Cleanup of the GIFPlot examples. Upgraded Java GIFPlot example. 8/01/2001: cheetah (william fulton) * [Java] Efficiency changes: _cPtr used where possible rather than getCPtr(). Bug fixes for inheritance - derived class sometimes didn't delete the c memory when _delete() was called. * [Java] Abstract c++ classes are wrapped with a java abstract shadow class. Also a pure virtual function is mapped with an abstract method. * The default output file has always been _wrap.c. It is now _wrap.cxx if the -c++ commandline option is passed to swig. This has been done as otherwise c++ code would appear in a c file. *** POTENTIAL INCOMPATIBILITY *** 7/31/2001: beazley Modified the %constant directive to be more C-like in syntax. The syntax is now: %constant NAME = VALUE; %constant TYPE NAME = VALUE; For example: %constant Foo *Bar = &Spam; A more subtle case is as follows: %constant int (*FOO)(int,int) = blah; *** POTENTIAL INCOMPATIBILITY *** Modules that were using the %constant directive directly will need to be modified. 7/30/2001: beazley Removed obscure and undocumented form of the %inline directive: %inline int blah(int a, int b) { ... } *** POTENTIAL INCOMPATIBILITY *** (note: this feature was never documented and is withdrawn) 7/30/2001: beazley Removed support for functions with no explicitly declared return type. For example: foo(int); In C, such functions were implicitly assumed to return an 'int'. In C++, this is illegal. Either way, it's considered bad style. Removing support for this in SWIG will simplify certain issues in parsing. *** POTENTIAL INCOMPATIBILITY *** 7/30/2001: mkoeppe * Partial merge from the CVS trunk. The Source/DOH directory and most of the Source/Swig directory is up-to-date now. * [Guile] %scheme is now a macro for %insert("scheme"). New syntax: %scheme "FILENAME"; New syntax: %scheme %{ SCHEME-CODE %} New macros %multiple_values, %values_as_list, %values_as_vector. 7/29/2001: beazley %readonly and %readwrite have been turned into SWIG pragmas. %pragma(swig) readonly and %pragma(swig) readwrite. Macros are used to provide backwards compatibility. 7/29/2001: beazley Minor changes to %pragma directive. %pragma must always be directed to a specific language. For example: %pragma(swig) make_default; %pragma(perl5) include = "blah.i"; Also extended the pragma directive to allow code blocks %pragma(foo) code = %{ ... some code ... %} *** POTENTIAL INCOMPATIBILITY *** 7/29/2001: beazley Change to the way 'const' variables are wrapped. In previous versions of SWIG, a 'const' variable was wrapped as a constant. Now, 'const' variables are wrapped as read-only variables. There are several reasons for making this change, mostly pertaining to subtle details of how 'const' actually works. This will probably break old interfaces that used 'const' to create constants. As a replacement, consider using this: const int a = 4; ===> %constant int a = 4; *** POTENTIAL INCOMPATIBILITY *** 7/29/2001: beazley Reorganization and simplification of type parsing. Types with 'const' should work correctly now. 7/29/2001: beazley Most swig directives related to the documentation system are now deprecated. 7/29/2001: beazley Removed support for Objective-C in order to simplify parser reconstruction. Will return if there is sufficient demand. *** POTENTIAL INCOMPATIBILITY *** 7/29/2001: beazley Code inclusion has been modified in the parser. A common directive %insert is now used for everything. This inserts a file into the output: %insert(header) "foo.swg" This inserts some inline code into the output %insert(header) %{ ... some code ... %} There are five predefined targets for the insert directive: "header" - Header section of wrapper file "runtime" - Runtime section of wrapper file "wrapper" - Wrapper section "init" - Initialization function "null" - Nothing. Discard. The following directives are still supported, but are now defined in terms of macros: %{ ... %} -> %insert(header) %{ ... %} %init %{ ... %} -> %insert(init) %{ ... %} %wrapper %{ ... %} -> %insert(wrapper) %{ ... %} %runtime %{ ... %} -> %insert(runtime) %{ ... %} Language modules can define new named targets by using the C API function Swig_register_filebyname() (see main.cxx). For example, if you wanted to expose a shadow class file, you could do this: Swig_register_filebyname("shadow", f_shadow); Then in the interface file: %insert(shadow) %{ ... %} Note: this change should not affect any old interfaces, but does open up new possibilities for enhancements. 7/29/2001: beazley SWIG now always includes a standard library file 'swig.swg'. This file defines a large number of macro definitions that define the behavior of various SWIG directives. Previously, all SWIG directives were handled as special cases in the parser. This made the parser a large bloated mess. Now, the parser is stripped down to a few simple directives and macros are used to handle everything else. 7/26/2001: cheetah (william fulton) * Fixes for Sourceforge bug #444748 - new testcase cpp_static: [TCL] Class with just static member variable/function fix [Java] Fixed static variables support [Ruby] Static variables workaround removed 7/27/2001: mkoeppe * stype.c (SwigType_default): Strip qualifiers first. The default type of "int * const" is now "SWIGPOINTER *". * main.cxx: Define "__cplusplus" in SWIG's preprocessor if in C++ mode. * [Guile]: Added some support for arrays and C++ references, fixing the "constant_pointers" test case. * Moved most tests from the old Guile-specific test-suite to the new test-suite. Also moved perl5/pointer-cxx example there. 7/26/2001: cheetah (william fulton) * Test-suite added. * Initial testcases: constant_pointers cpp_enum defines sizeof_pointers unions virtual_destructor * Make clean improvements. 7/24/2001: cheetah (william fulton) * [Java] Underscores in the package name and/or module name no longer give linking problems. 7/17/2001: cheetah (william fulton) * More parser bug fixes for constant pointers 7/19/2001: mkoeppe * [Guile] Aesthetic improvement in variable wrappers. 7/18/2001: beazley * Fixed core-dump problem in pointer library when freeing character arrays. SF Bug [ #415837 ] pointer lib core dump 7/18/2001: beazley * Fixed problem with default destructors and shadow classes. SF bug #221128. 7/18/2001: beazley * To provide better line-number tracking in interfaces with lots of macros, special locator comments are now generated by the SWIG preprocessor. For example: /*@foo.i,42,BLAH@*/expanded macro/*@@*/ The first /*@...@*/ sequence sets the context to point to the macro code. The /*@@*/ comment terminates the context. The SWIG parser should ignore all of the locator comments as should the C compiler (should such comments end up in generated wrapper code). 7/18/2001: mkoeppe * The parser now handles severely constified types in typemaps. This introduced a new shift/reduce conflict, but only with a heuristic function-pointer catch-all rule. * [Guile]: Added typemaps for severely constified types. * Fixed the "template-whitespace" problem by canonicalizing whitespace, especially around angle brackets and commas. 7/17/2001: mkoeppe * [Guile]: A Scheme file is emitted if the -scmstub FILE.SCM command-line option is used. The %scheme directive (implemented as a macro for a pragma) allows to insert arbitrary code here. In "simple" and "passive" linkage, the file gets filled with define-module and export declarations. 7/17/2001: cheetah (william fulton) * Parser bug fix to support constant pointers, eg int* const ptr. Fixed everywhere - variables, parameters, return types etc. Note that when wrapping a constant pointer variable only the getter is generated. 7/17/2001: mkoeppe * Fixed SF bug #441470 (#define X "//" would not be parsed, see test-suite entry "preproc-1"), reported by T. W. Burton . * Changed the type of character constants to "char", rather than "char *". Changed the individual language modules to keep the old behaviour, except for the Guile module, where it is desired to make them Scheme characters. This fixes SF bug #231409, test-suite entry "char-constant". * Applied patch for DOH/Doh/memory.c by Les Schaffer (avoid required side effects in assert). 7/17/2001: cheetah (william fulton) * Bug fix in parser for virtual destructor with void as parameter * Bug fix in parser #defines embedded within classes/structs/unions Consequently %constant can now also be placed within a struct/class/union. * Bug fix in parser to allow sizeof(*I_am_a_pointer) within a #define 7/16/2001: mkoeppe * Added changes for the Macintosh contributed by Luigi Ballabio . * Some "const" fixes in the code. * [Guile]: Made the constant-wrapper functions much shorter. 7/13/2001: mkoeppe * [Guile]: Some "const" fixes for Guile version 1.3.4. * Handle anonymous arguments with default values and static array members of classes. Both bugs reported by Annalisa Terracina ; see the files Examples/guile/test-suite/static-array-member.i and anonymous-arg.i. Version 1.3.6 (July 9, 2001) ============================= 7/09/2001: cheetah (william fulton) * GIFPlot examples: FOREGROUND and BACKGROUND definition missing after TRANSPARENT #define fix in GIFPlot 7/03/2001: beazley Fixed up the version numbers so that the release is known as 1.3.6. All future releases should have a similar version format. 7/02/2001: mkoeppe * [Python]: Prevent the problem of self.thisown not being defined if the C++ class constructor raised an exception. Thanks to Luigi Ballabio . 6/29/2001: mkoeppe * More portability fixes; fixed "gcc -Wall" warnings. 6/29/2001: cheetah (william fulton) * GIFPlot examples: TRANSPARENT #define multiple times on Solaris (clashes with stream.h). * Multiple definition bug fix for shadow classes. The perl and python modules had workarounds which have been replaced with fixes in the core. Many of the Language::cpp_xxxx functions now set a flag which the derived classes can access through is_multiple_definition() to see whether or not code should be generated. The code below would have produced varying degrees of incorrect shadow class code for the various modules: class TestClass { public: TestClass() {}; TestClass(int a) {}; ~TestClass() {}; unsigned long xyz(short k) {}; unsigned long xyz(int n) {}; static void static_func() {}; static void static_func(int a) {}; }; void delete_TestClass(int a); 6/27/2001: mkoeppe * [Perl] Another const-related portability fix. 6/26/2001: cheetah (william fulton) * [Java] Added in cpp_pragma() support with a host of new pragmas - see jswig.html. These are designed for better mixing of Java and c++. It enables the user to specify pure Java classes as bases and/or interfaces for the wrapped c/c++. * [Java] Old pragmas renamed. Warning given for the moment if used. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 6/25/2001: mkoeppe * Incorporated more build changes contributed by Wyss Clemens for swig/ruby on cygwin. 6/20/2001: cheetah (william fulton) * Makefile mods so that 'make check' uses the swig options in the makefiles * [Java] Removed Generating wrappers message * [Java] NULL pointer bug fix * [Java] Bug fix for Kaffe JVM 6/20/2001: mkoeppe * SWIG_TypeQuery from common.swg now returns a swig_type_info* rather than a void*. This fixes a problem when using pointer.i and C++, as illustrated by the new test-suite example perl5/pointer-cxx. * Portability fixes (const char *). * Incorporated build changes contributed by Wyss Clemens , which make swig runnable on cygwin. 6/19/2001: cheetah (william fulton) * [Java] Bug fix for SF bug #211144. This fix is a workaround until fixed in the core. 6/19/2001: mkoeppe * [Guile]: Portability fixes for use with the Sun Forte compilers. * [Tcl]: Portability fix (const char *). * [Tcl]: Configure now first tries to find a tclConfig.sh file in order to find the Tcl include directory, library location and library name. * [Python]: Added a few possible library locations. 6/18/2001: mkoeppe * [Guile]: Don't call scm_c_export if nothing is to be exported. Don't warn on %module if module has been set already (this frequently occurs when %import is used). 6/16/2001: mkoeppe * [Guile]: New "passive" linkage, which is appropriate for multi-module extensions without Guile module magic. 6/15/2001: mkoeppe * [Guile]: Fixed printing of smobs (space and angle were missing). * Properly generate type information for base classes imported with the %import directive. Thanks to Marcelo Matus for the report and the patch; this closes SF bug #231619; see also Examples/guile/test-suite/import*. * [Guile]: Fix casting between class and base class; the runtime type system had it the wrong way around; see Examples/guile/test-suite/casts.i * Make typemaps for SWIGPOINTER * with arg name take precedence over those without arg name, to match normal typemap precedence rules. * Fixed the random-line-numbers problem reported as SF bug #217310; thanks to Michael Scharf . * [Guile]: Handle the %name and %rename directives. * New syntax: %name and %rename now optionally take double quotes around the scripting name. This is to allow scripting names that aren't valid C identifiers. 6/14/2001: beazley Made a minor change to the way files are loaded in order to get file/line number reporting correct in the preprocessor. 6/14/2001: mkoeppe * The parser now understands the (non-standard) "long long" types. It is up to the individual language modules to provide typemaps if needed. Reported by Sam Steingold, SF bug #429176. * The parser now understands arguments like "const int * const i". This fixes SF bug #215649. * Fixed the Guile test-suite. 6/13/2001: mkoeppe Partial merge from the CVS trunk at tag "mkoeppe-merge-1". This covers the following changes: | 01/16/01: ttn | Wrote table of contents for Doc/engineering.html. Added section | on CVS tagging conventions. Added copyright to other docs. | 9/25/00 : beazley | Modified the preprocessor so that macro names can start with a '%'. | This may allow new SWIG "directives" to be defined as macros instead | of having to be hard-coded into the parser. | | *** Also a yet-to-be-documented quoting mechanism with backquotes | *** has been implemented? 6/13/2001: mkoeppe * When configure does not find a language, don't use default paths like /usr/local/include; this only causes build problems. * New directory: Examples/Guile/test-suite, where a few bugs in 1.3a5 are demonstrated. * Handle C++ methods that have both a "const" and a "throw" directive (see Examples/Guile/test-suite/cplusplus-throw.i); thanks to Scott B. Drummonds for the report and the fix. * Handle C++ pointer-reference arguments (like "int *& arg") (see Examples/Guile/test-suite/pointer-reference.i, reported as SF bug #432224). * [Ruby] Fixed typo in rubydec.swg; thanks to Lyle Johnson! * Don't stop testing when one test fails. * [Guile, MzScheme] Don't print "Generating wrappers...". 6/12/2001: mkoeppe [Guile] VECTORLENINPUT and LISTLENINPUT now have separate list length variables. TYPEMAP_POINTER_INPUT_OUTPUT attaches argument documentation involving SCM_TYPE to the standard pointer typemaps. INOUT is now an alias for BOTH. 6/12/2001: cheetah (william fulton) Some Java documentation added. [Java] Fixed bugs in import pragma and shadow pragma. 6/12/2001: mkoeppe Fix declarations of SWIG_define_class (Lib/ruby/rubydec.swg) and SWIG_TypeQuery (Lib/common.swg). Thanks to Lyle Johnson for the patches. 6/11/2001: mkoeppe [Guile] Use long instead of scm_bits_t; this makes the generated wrapper code compatible with Guile 1.3.4 again. Thanks to Masaki Fukushima for pointing this out. 6/11/2001: cheetah (william fulton) The generic INSTALL file from autoconf added. Few changes to README file. 6/11/2001: mkoeppe Fixed typo in Makefile.in; thanks to Greg Troxel . 6/08/2001: cheetah (william fulton) make check works again. Examples/GIFPlot configure generated by top level autoconf now. 6/08/2001: mkoeppe Another build change: The new script autogen.sh runs autoconf in the appropriate directories. The top-level configure also configures in Examples/GIFPlot. 6/07/2001: mkoeppe Made the Makefile work with non-GNU make again. 6/07/2001: cheetah (william fulton) [Java] Class/struct members that are arrays of pointers to classes/structs - Shadow class's get/set accessors now use Java classes instead of longs (pointers). [Java] Shadow classes will now clean up memory if function return type is a class/struct. [Java] New example called reference based on the same example from other modules. 6/06/2001: mkoeppe New configure option --with-release-suffix allows for attaching a suffix to the swig binary and the swig runtime libraries. Minor changes to the build system. "swig -swiglib" works again. If invoked with the new option "-ldflags", SWIG prints a line of linker flags needed to link with the runtime library of the selected language module. 6/06/2001: mkoeppe [Guile] gswig_list_p is an int, not a SCM. This typo caused warnings when compiling with a Guile configured with strict C type checking. In INPUT and BOTH typemaps generated by the SIMPLE_MAP macro, use the SCM_TO_C function to convert from Guile to C (rather than C_TO_SCM). Use scm_intprint to print pointers (rather than sprintf). Allow using "-linkage" instead of "-Linkage". 6/05/2001: cheetah (william fulton) [Java] Mods for using inherited c++ classes from Java [Java] New example called class based on the same example from other modules 6/05/2001: cheetah (william fulton) [Java] destructor (_delete()) was not aware of %name renaming [Java] extends baseclass did not know about %name renaming [Java] extends baseclass did extend even when the baseclass was not known to swig [Java] sometimes enum-declarations occured before the Java class declaration [Java] unrelated enum initialisations no longer appear in Java class [Java] if module ends in '_' correct JNI names are now produced 6/04/2001: cheetah (william fulton) [Java] Shadow class mods - Modified constructor replaces newInstance(). _delete() now thread safe. getCPtr() replaces _self. _selfClass() removed as now redundant. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** [Java] Not all output java files had SWIG banner. New banner. [Java] Shadow class finalizers are output by default: Command line option -finalize deprecated and replaced with -nofinalize. *** POTENTIAL INCOMPATIBILITY FOR JAVA MODULE *** 6/ 1/2001: mkoeppe [Guile] Cast SCM_CAR() to scm_bits_t before shifting it. This is required for compiling with a Guile configured with strict C type checking. 6/ 1/2001: mkoeppe Added configure option "--with-swiglibdir". 5/31/2001: mkoeppe [Guile] Support multiple parallel lists or vectors in the typemaps provided by list-vector.i. New typemaps file, pointer-in-out.i. 5/25/2001: cheetah (william fulton) [Java] HTML update for examples. 5/28/2001: mkoeppe Minor changes to the build system. Added subdirectory for Debian package control files. 5/28/2001: mkoeppe [Guile] Build a runtime library, libswigguile. 5/28/2001: mkoeppe [Guile] New typemap substitution $*descriptor. Use the {} syntax, rather than the "" syntax for the standard typemaps, in order to work around strange macro-expansion behavior of the SWIG preprocessor. This introduces some extra braces. 5/27/2001: mkoeppe [Guile] Handle pointer types with typemaps, rather than hard-coded. New typemap substitutions $descriptor, $basedescriptor; see documentation. Some clean-up in the variable/constants wrapper generator code. New convenience macro SWIG_Guile_MustGetPtr, which allows getting pointers from smobs in a functional style. New typemap file "list-vector.i", providing macros that define typemaps for converting between C arrays and Scheme lists and vectors. 5/25/2001: cheetah (william fulton) [Java] STL string moved into its own typemap as it is c++ code and it break any c code using the typemaps.i file. - Fixes for wrappers around global variables - applies to primitive types and user types (class/struct) and pointers to these. - Structure member variables and class public member variables getters and setters pass a pointer to the member as was in 1.3a3 and 1.1 (1.3a5 was passing by value) - Parameters that were arrays and return types were incorrectly being passed to create_function() as pointers. - Fix for arrays of enums. [Java] Updated java examples and added two more. [Java] Java module updated from SWIG1.3a3 including code cleanup etc. [Java] enum support added. [Java] Array support implemented [Java] Shadow classes improved - Java objects used rather than longs holding the c pointer to the wrapped structure/c++class 5/22/2001: mkoeppe [Guile] Fixed extern "C" declarations in C++ mode. Thanks to Greg Troxel . 5/21/2001: mkoeppe [Guile] New linkage "module" for creating Guile modules for Guile versions >= 1.5.0. 4/18/2001: mkoeppe [MzScheme] Added typemaps for passing through Scheme_Object pointers. 4/9/2001 : mkoeppe [MzScheme] Added typemaps for `bool'. Inclusion of headers and support routines is now data-driven via mzscheme.i. Headers come from the new file mzschemdec.swg. Don't abort immediately when a type-handling error is reported. When searching for typemaps for enums, fall back to using int, like the Guile backend does. Support char constants. Emit correct wrapper code for variables. 3/12/2001: mkoeppe [Guile] Fixed typemaps for char **OUTPUT, char **BOTH. 3/2/2001 : mkoeppe [Guile] Every wrapper function now gets a boolean variable gswig_list_p which indicates whether multiple values are present. The macros GUILE_APPEND_RESULT, GUILE_MAYBE_VALUES and GUILE_MAYBE_VECTOR use this variable, rather than checking whether the current return value is a list. This allows for typemaps returning a list as a single value (a list was erroneously converted into a vector or a multiple-value object in this case). 3/1/2001 : mkoeppe [Guile] Added support for returning multiple values as vectors, or passing them to a muliple-value continuation. By default, multiple values still get returned as a list. 3/1/2001 : mkoeppe [Guile] Added a "beforereturn" pragma. The value of this pragma is inserted just before every return statement. 3/1/2001 : mkoeppe [Guile] Added support for Guile 1.4.1 procedure documentation formats, see internals.html. 2/26/2001: mkoeppe [Guile] Made the wrapper code compile with C++ if the "-c++" command-line switch is given. Thanks to . 2/26/2001: mkoeppe [Guile] Now two type tables, swig_types and swig_types_initial, are used, as all other SWIG language modules do. This removes the need for the tricky construction used before that the broken Redhat 7.0 gcc doesn't parse. Reported by . 2/26/2001: mkoeppe [Guile] Fixed typemaps for char *OUTPUT, char *BOTH; a bad free() would be emitted. Added typemap for SCM. Version 1.3 Alpha 5 =================== 9/19/00 : beazley [Python] Python module generates more efficient code for creating the return value of a wrapper function. Modification suggested by Jon Travis. 9/19/00 : beazley Library files specified with the -l option are now included at the end of the interface file (reverting to the old behavior). 9/19/00 : beazley Fixed some problems with enum handling. enums are now manipulated as 'int', but cast into the enum type when values are passed to the corresponding C function. 9/19/00 : mkoeppe [Guile] Removed "-with-smobs" command-line option, as this is the default now. Added "-emit-setters" command-line option, which turns on generating procedures-with-setters; see internals.html. 9/18/00 : mkoeppe Incorporated patch #101430, fixing bugs in the Guile module: 1. Some arguments were erroneously taken as *optional* arguments when ignored arguments were present. 2. Guile 1.3.4 was not supported since functions introduced in Guile 1.4 were used. 3. Added handling of `const char *'. 9/17/00 : beazley Fixed problem with failed assertion and large files. 9/17/00 : beazley Fixed problem with the '%' character appearing in added methods and function bodies. Preprocessor bug. Version 1.3 Alpha 4 (September 4, 2000) ====================================== 9/3/00 : ttn Added instructions for maintainers in Examples/README on how to make examples also be useful in the testing framework. Also, "make check" now uses ./Lib by via env var `SWIG_LIB'. This is overridable like so: make chk-swiglib=/my/experimental/swig/Lib check 9/3/00 : beazley Added $typemap variable to typemaps. This gets replaced with a string indicating the typemap that is applied. Feature request from rsalz. 9/3/00 : beazley Experimental optimization to code generation for virtual member functions. If you have two classes like this: class A() { virtual void foo(); } class B() : public A { virtual void foo(); } Swig now will generate a single wrapper function for this A_foo(A *a) { a->foo(); } and use it as the implementation of both A_foo() and B_foo(). This optimization only takes place if both methods are declared as virtual and both take identical parameters. *** EXPERIMENTAL FEATURE *** 9/3/00 : beazley Restored the "memberin" typemap for setting structure members. Unlike the old version, the new version is expanded inline in the wrapper function allowing access to scripting language internals (a sometimes requested feature). The "memberout" typemap is gone. Use the "out" typemaps instead. *** POTENTIAL INCOMPATIBILITY *** 9/3/00 : beazley Attribute set methods no longer return the value of a member. For example: struct Foo { int x; ... } now gets set as follows: void Foo_x_set(Foo *f, int x) { f->x = x; } In SWIG1.1 it used to be this: int Foo_x_set(Foo *f, int x) { return (f->x = x); } This has been changed due to the complexity created by trying to do this with more exotic datatypes such as arrays. It also complicates inlining and handling of the "memberin" typemap. *** POTENTIAL INCOMPATIBILITY *** 9/2/00 : beazley Removed the ptrcast() and ptrmap() functions from the pointer.i library file. Old implementation is incompatible with new type system. *** POTENTIAL INCOMPATIBILITY *** 9/2/00 : beazley New runtime function SWIG_TypeQuery(const char *name) added. This function can be used to extract the type info structure that is used for type-checking. It works with either the nice C name or mangled version of a datatype. For example: swig_type_info *ty = Swig_TypeQuery("int *"); swig_type_info *ty = Swig_TypeQuery("_p_int"); This is an advanced feature that has been added to support some exotic extension modules that need to directly manipulate scripting language objects. *** NEW FEATURE *** 9/2/00 : beazley New directive %types() added. This is used to explicitly list datatypes that should be included in the runtime type-checking code. Normally it is never necessary to use this but sometimes advanced extensions (such as the pointer.i library) may need to manually add types to the type-checker. *** NEW FEATURE *** 8/31/00 : beazley Improved handling of string array variables. For example, a global variable of the form "char name[64]" is automatically managed as a 64 character string. Previously this didn't work at all or required the use of a special typemap. *** NEW FEATURE (Tcl, Perl, Python) *** 8/31/00 : ttn Added Makefile target `check-c++-examples', which uses new files under Examples/C++ contributed by Tal Shalif. Now "make check" also does "make check-c++-examples". Also, expanded actions in `check-gifplot-example' and `check-aliveness'. 8/30/00 : mkoeppe Major clean-up in the Guile module. Added typemap-driven documentation system. Changed to handle more than 10 args. Updated and extended examples. *** NEW FEATURE *** 8/29/00 : beazley Added new %insert directive that inserts the contents of a file into a portion of the output wrapper file. This is only intended for use by writers of language modules. Works as follows: %insert(headers) "file.swg"; %insert(runtime) "file.swg"; %insert(wrappers) "file.swg"; %insert(init) "file.swg"; *** NEW FEATURE *** 8/29/00 : beazley Added new %runtime directive which includes code into the runtime portion of the wrapper code. For example: %runtime %{ ... some internal runtime code ... %} There is no practical reason for ordinary users to use this feature (almost everything can be done using %{ ... %} instead). However, writers of language modules may want to use this in language configuration files. *** NEW FEATURE *** 8/28/00 : beazley Typemaps can now be specified using string literals like this: %typemap(in) int "$target = SvIV($source);"; When code is specified like this, it is *NOT* enclosed inside a local scope (as with older typemap declarations). Note: character escape sequences are interpreted in the code string so if you want to include a quote or some other special character, make sure you use a (\). *** NEW FEATURE *** 8/27/00 : beazley Typemaps have been modified to follow typedef declarations. For example, if you have this: typedef int Number; %typemap(in) int { ... get an integer ... } void foo(Number a); The typemap for 'int' will be applied to the argument 'Number a'. Of course, if you specify a typemap for 'Number' it will take precedence (nor will it ever be applied to an 'int'). *** POTENTIAL INCOMPATIBILITY *** 8/27/00 : beazley Default typemap specification has changed. In older versions of swig, you could do this: %typemap(in) int SWIG_DEFAULT_TYPE { ... } To specify the default handling of a datatype. Now that SWIG follows typedef declarations, this is unnecessary. Simply specifying a typemap for 'int' will work for all variations of integers that are typedef'd to 'int'. Caveat, specifying the default behavior for pointers, references, arrays, and user defined types is a little different. This must be done as follows: %typemap() SWIGPOINTER * { ... a pointer ... } %typemap() SWIGREFERENCE & { ... a reference ... } %typemap() SWIGARRAY [] { ... an array ... } %typemap() SWIGTYPE { ... a user-defined type (by value) ... } *** POTENTIAL INCOMPATIBILITY *** 8/15/00 : dustin The file swig-1.3a1-1.spec has been added to the Tools directory. It can be used to build a redhat package for SWIG, although it will need to be updated for the next public release. 8/15/00 : beazley Typemaps have been completely rewritten. Eventually they may be replaced with something better, but for now they stay. However, there are a number of a significant changes that may trip some people up: 1. Typemap scoping is currently broken. Because of this, the following code won't work. %typemap(in) blah * { ... } class Foo { ... int bar(blah *x); } %typemap(in) blah *; /* Clear typemap */ (this breaks because the code for the class Foo is actually generated after the entire interface file has been processed). This is only a temporary bug. 2. In SWIG1.1, the %apply directive worked by performing a very complex type-aliasing procedure. From this point on, %apply is simply a generalized typemap copy operation. For example, %apply double *OUTPUT { double *x, double *y }; Copies *ALL* currently defined typemaps for 'double *OUTPUT' and copies them to 'double *x' and 'double *y'. Most people probably won't even notice this change in %apply. However, where it will break things is in code like this: %apply double *OUTPUT { double *x }; %typemap(in) double *OUTPUT { ... whatever ... } void foo(double *x); In SWIG1.1, you will find that 'foo' uses the 'double *OUTPUT' rule even though it was defined after the %apply directive (this is the weird aliasing scheme at work). In SWIG1.3 and later, the 'double *OUTPUT' rule is ignored because it is defined after the %apply directive. 3. The %clear directive has been modified to erase all currently defined typemaps for a particular type. This differs from SWIG1.1 where %clear only removed rules that were added using the %apply directive. 4. Typemap matching is now performed using *exact* types. This means that things like this %typemap(in) char * { } %typemap(in) const char * { } are different typemaps. A similar rule applies for pointers, arrays, and references. For example: %typemap(in) double * { } used to apply to 'double &', 'double []', Now, it only applies to 'double *'. If you want a 'double &', you'll need to handle that separately. 5. Array matching has been simplfied. In SWIG1.1, array matching was performed by trying various combinations of dimensions. For example, 'double a[10][20]' was matched as follows: double [10][20] double [ANY][20] double [10][ANY] double [ANY][ANY] In SWIG1.3, only the following matches are attempted: double [10][20] double [ANY][ANY] On the positive side, typemap matching is now *significantly* faster than before. *** POTENTIAL INCOMPATIBILITY *** 8/15/00 : beazley Secret developer feature. Since datatypes are now represented as strings internally, you can bypass limitations of the parser and create a wild datatype by simply enclosing the raw string encoding in backticks (``) and sticking it in the interface file anywhere a type is expected. For example, `a(20).a(10).p.f(int,int)`. This feature is only intended for testing (i.e., you want to see what happens to your language module if it gets a reference to a pointer to an array of pointers to functions or something). *** SICK HACK *** 8/14/00 : beazley Completely new type-system added to the implementation. More details later. 8/11/00 : beazley Cleaned up some of the I/O handling. SWIG no longer generates any temporary files such as _wrap.wrap, _wrap.ii, _wrap.init. Instead, these "files" are kept around in memory as strings (although this is transparent to language modules). 8/4/00 : ttn Added Makefile target "check" and variants. This can be used like "make check" or, to explicitly skip a language LANG: "make skip-LANG=true check". LANG is skipped automatically if ./configure determines that LANG support is insufficient. Currently, the check is limited to doing the equivalent of "make all" in some of the Examples directories. This should be expanded both horizontally (different types of tests) and vertically (after "make all" in an Examples subdir succeeds, do some additional tests with the resulting interpreter, etc). 8/4/00 : ttn Added Makefile target "distclean", which deletes all the files ./configure creates, including config.status and friends. 8/3/00 : harcoh java changes??? [todo: document changes] 7/23/00 : beazley Typemaps have been modified to key off of the real datatypes used in the interface file. This means that typemaps for "const char *" and "char *" will be difference as will typemaps for "Vector" and "Vector *." *** POTENTIAL INCOMPATIBILITY *** This is likely to break interfaces that rely on the odd type handling behavior of typemaps in SWIG1.1--especially with respect to interfaces involving pass-by-value. 7/23/00 : beazley New %constant directive. This directive can be used to create true constants in the target scripting language. It's most simple form is something like this: %constant FOO 42; In this case, the type is inferred from the syntax of the value (in reality, all #define macros are translated into directives of this form). An expanded version is as follows: %constant(Foo *) FOO = &FooObj; In this case, an explicit type can be specified. This latter form may be useful for creating constants that used to be specified as const Foo *FOO = &FooObj; (which are now treated as variables). *** EXPERIMENTAL FEATURE *** The syntax may change in the final release. 7/23/00 : beazley Modified the parser so that variable declarations of the form "const type *a" are handled as variables, not constants. Note: SWIG1.1 handled this case erroneously because const char *a is a pointer variable that can be reassigned. *** POTENTIAL INCOMPATIBILITY *** Note: just because this is the "right" way to do things, doesn't mean it's the most appropriate interpretation. I suspect that many C programmers might use 'const char *' with the intent of creating a constant, without realizing that they've created a reassignable global variable. 7/23/00 : beazley The C/C++ wrapping layer has been completely redesigned and reimplemented. This change should iron out a few rough spots with the handling of datatypes. In addition, the wrapper code is somewhat cleaner. *** POTENTIAL INCOMPATIBILITY *** This change may break interfaces that involve subtle corner-cases with typemaps and the %addmethods directive since some of these features had somewhat type handling behavior in SWIG1.1. 7/23/00 : beazley The "memberin" and "memberout" typemaps are gone for the moment, but they might return as soon as I figure out how to integrate them with some of the streamlined C wrapper functions. *** POTENTIAL INCOMPATIBILITY *** 7/22/00 : beazley A variety of old type handling functions such as print_type(), print_full(), print_mangle(), etc... are gone and have been replaced with a smaller set of functions. See the file Doc/internals.html for details. This will break all third party language modules. *** POTENTIAL INCOMPATIBILITY *** 7/20/00 : beazley Deprecated the %val and %out directives. These directives shouldn't really be necessary since typemaps can be used to achieve similar results. This also cleans up the handling of types and parameters quite a bit. *** POTENTIAL INCOMPATIBILITY *** 7/20/00 : ttn Fixed unspecified-module bug in Guile support and removed more non-"with-smobs" functionality using patches submitted by Matthias Koeppe. Re-enable recognition of "-with-smobs" (with no effect since we use smobs by default now) for the time being. After the 1.3a4 release, this option will signal an error. 7/17/00 : ttn Fixed NULL-input bug in parameter list handling. Reported by Matthias Koeppe. 7/12/00 : beazley Fixed memory leak in Python type-checking code. Reported by Keith Davidson. Bug #109379. 7/10/00 : beazley Changed internal data structures related to function parameters. 7/10/00 : beazley Fixed some bugs related to the handling of the %name() directive and classes in the Tcl module. Problem reported by James Bailey. 7/10/00 : beazley Fixed parsing and enum handling problems with character constants. Reported by Greg Kochanski. 7/10/00 : beazley Removed WrapperFunction class from the core and updated the language module. This will break third party modules. *** POTENTIAL INCOMPATIBILITY *** 7/9/00 : beazley Implementation of SWIG no longer makes use of C++ operator overloading. This will almost certainly break *all* third party language modules that are not part of the main SWIG CVS tree. Sorry. *** POTENTIAL INCOMPATIBILITY *** 7/8/00 : beazley Removed the experimental and undocumented "build" typemap that was intended to work with multiple arguments. Simply too weird to keep around. Besides, a better replacement is in the works. 7/6/00 : ttn Removed non-"with-smobs" functionality (Guile support), i.e., "-with-smobs" is now the default and no longer needs to be specified on the command-line. 7/5/00 : ttn Incorporated Ruby support contributed by Masaki Fukushima. 6/28/00 : ttn Applied more-than-10-args bugfix patch contributed by Matthias Koeppe. 6/27/00 : beazley Rewrote some of the string handling and eliminated the C++ implementation (which is now just a wrapper). 6/27/00 : ttn Added Doc/index.html and Doc/internals.html. The target audience for the latter is new SWIG developers. Version 1.3 Alpha 3 (June 18, 2000) =================================== 6/18/00 : beazley Removed the naming.cxx, hash.cxx, and symbol.cxx files from the SWIG1.1 directory. Continued to migrate things away from the C++ base (although there's still a lot of work to do). 6/17/00 : beazley Added a few more examples to the Examples directory. Still need to do a lot of work on this. 6/16/00 : beazley Added -includeall to follow all #include statements in the preprocessor. 6/15/00 : beazley Tried to fix as many C++ warnings as possible when compiling with the Sun Workshop C++ compiler. Unfortunately, this means that there are a lot of statements that contain string literals of the form (char*)"Blah". 6/15/00: beazley A variety of cleanup and performance optimization in the low-level DOH library. This seems to result in a speedup of 50-100% for preprocessing and other related tasks. 5/10/00 : ttn Applied variable-wrapping bugfix patch contributed by Matthias Koeppe. 4/17/00 : ttn Updated MzScheme support contributed by Oleg Tolmatcev. We now use a `Scheme_Type'-based structure to wrap pointers. 4/11/00 : ttn Incorporated further Guile-support patch by Matthias Koeppe. Typemaps previously deleted have been re-added. There is now exception handling (see Doc/engineering.html). `SWIG_init' is now declared extern only for simple linkage. Some bugs were fixed. 4/06/00 : ttn Incorporated MzScheme support contributed by Oleg Tolmatcev. This includes new directories Lib/mzscheme and Examples/mzscheme. 4/03/00 : ttn Added Examples/guile and children. This is an adaptation of the same-named directory from the SWIG-1.1p5 distribution. Added Guile-specific section to Doc/engineering.html. 4/02/00 : ttn Incorporated new guilemain.i by Martin Froehlich. Incorporated Guile-support rewrite patch by Matthias Koeppe. The command line option "-with-smobs" enables implementation of pointer type handling using smobs, the canonical mechanism for defining new types in Guile. Previous implementation (using strings) is at the moment still supported but deprecated. At some point, "-with-smobs" will be the default and no longer required. 3/13/00 : beazley Added purify patches submitted by Ram Bhamidipaty. 3/02/00 : ttn Added support for different Guile "linkage" schemes. Currently, "-Linkage hobbit" works. Version 1.3 Alpha 2 (March 1, 2000) =================================== 2/29/00 : beazley Made SWIG ignore the 'mutable' keyword. 2/29/00 : beazley Incorporated some patches to the Perl5 module related to the -hide option and the destruction of objects. Patch submitted by Karl Forner. 2/27/00 : ttn Incorporated Guile support contributed by Matthias Koeppe. This includes a cpp macro in Lib/guile/guile.swg and the entire file Lib/guile/typemaps.i. 2/25/00 : ttn Modified configure.in and Makefile.in files to support non-local build (useful in multi-arch environments). 2/24/00 : ttn Incorporated Guile support contributed by Clark McGrew. This works with Guile 1.3, but since it depends heavily on the gh_ interface, it should work for all later versions. It has not been tested with versions before 1.3. WARNING: Code is unstable due to experimentation by ttn. 2/16/00 : beazley A variety of performance improvements to the Python shadow class code generation. Many of these result in substantial runtime performance gains. However, these have come at a cost of requiring the use of Python 1.5.2. For older versions, use 'swig -noopt -python' to turn off these optimization features. Version 1.3 Alpha 1 (February 11, 2000) ======================================= 2/11/00 : Added 'void' to prototype of Python module initializer. Reported by Mark Howson (1/20/00). 2/11/00 : beazley Modified the Python shadow class code to discard ownership of an object whenever it is assigned to a member of another object. This problem has been around for awhile, but was most recently reported by Burkhard Kloss (12/30/99). 2/11/00 : beazley Added braces around macros in the exception.i library. Reported by Buck Hodges (12/19/99) 2/11/00 : beazley Fixed bug in the constraints.i library. Reported by Buck Hodges (12/14/99) 2/11/00 : beazley The %native directive now generates Tcl8 object-style command calls. A full solution for Tcl7 and Tcl8 is still needed. Patch suggested by Mike Weiblen (11/29/99) 2/11/00 : beazley Modified the typemap code to include the $ndim variable for arrays. Patch provided by Michel Sanner (11/12/99). 2/11/00 : beazley Modified the Python module to raise a Runtime error if an attempt is made to set a read-only member of a shadow class. Reported by Michel Sanner (11/5/99). 2/10/00 : The documentation system has been removed. However, it is likely to return at some point in the future. 2/1/00 : Added a number of performance enhancements to the Python shadow classing and type-checking code. Contributed by Vadim Chugunov. 1. Remove _kwargs argument from the shadow wrappers when -keyword option is not specified. This saves us a construction of keyword dictionary on each method call. def method1(self, *_args, **_kwargs): val = apply(test2c.PyClass1_method1, (self,) + _args, _kwargs) return val becomes def method1(self, *_args): val = apply(test2c.PyClass1_method1, (self,) + _args) return val 2. Incorporate self into the _args tuple. This saves at least one tuple allocation per method call. def method1(self, *_args): val = apply(test2c.PyClass1_method1, (self,) + _args) return val becomes def method1(*_args): val = apply(test2c.PyClass1_method1, _args) return val 3. Remove *Ptr classes. Assume that we are SWIGging a c++ class CppClass. Currently SWIG will generate both CppClassPtr class that hosts all methods and also CppClass that is derived from the former and contains just the constructor. When CppClass method is called, the interpreter will try to find it in the CppClass's dictionary first, and only then check the base class. CppClassPtr functionality may be emulated with: import new _new_instance = new.instance def CppClassPtr(this): return _new_instance(CppClass, {"this":this,"thisown":0}) This saves us one dictionary lookup per call. The new module was first added in Python-1.5.2 so it won't work with older versions. I've implemented an alternative that achieves the same thing 4. Use CObjects instead of strings for pointers. Dave: This enhancements result in speedups of up to 50% in some of the preliminary tests I ran. 2/1/00 : Upgraded the Python module to use a new type-checking scheme that is more memory efficient, provides better performance, and is less error prone. Unfortunately, it will break all code that depends on the SWIG_GetPtr() function call in typemaps. These functions should be changed as follows: if (SWIG_GetPtr(string,&ptr,"_Foo_p")) { return NULL; } becomes if (SWIG_ConvertPtr(pyobj, &ptr, SWIG_TYPE_Foo_p) == -1) { return NULL; } Note: In the new implementation SWIG_TYPE_Foo_p is no longer a type-signature string, but rather an index into a type encoding table that contains type information. *** POTENTIAL INCOMPATIBILITY *** 1/30/00 : loic Conditionaly compile experimental code with --enable-experiment configure flag. Fix .cvsignore to ignore configrue & yacc generated files 1/28/00 : loic Apply automake everywhere Keep configure scripts so that people are not *forced* to autoconf Keep sources generated by yacc so that compilation without yacc is possible. Source/LParse/cscanner.c: change lyacc.h into parser.h to please default yacc generation rules. Use AC_CONFIG_SUBDIRS in configure.in instead of hand made script. Update all relevant .cvsignore to include .deps Fixed missing ; line 136 Source/Swig/swig.h 1/13/00 : beazley Fixed a number of minor end-of-file parsing problems in the preprocessor. 1/13/00 : beazley Added -freeze option that forces SWIG to freeze upon exit. This is only used as a debugging tool so that I can more easily examine SWIG's memory footprint. 1/13/00 : beazley Added patch to guile module for supporting optional arguments Patch contributed by Dieter Baron. 1/13/00 : loic Added .cvsignore, Examples/.cvsignore, Source/DOH/Doh/.cvsignore Source/SWIG1.1/main.cxx: Fixed -I handling bug Source/Modules1.1/java.cxx: fixed char* -> const char* warnings that are errors when compiling with gcc-2.95.2 Source/SWIG1.1/main.cxx: cast const char* to char* for String_replace token and rep should really be const. 1/12/00 : beazley Added Harco's Java modules. 1/12/00 : beazley Revoked the %ifdef, %ifndef, %endif, %if, %elif, and %else directives. These are no longer needed as SWIG now has a real preprocessor. *** POTENTIAL INCOMPATIBILITY *** 1/12/00 : beazley Moved the documentation modules from the SWIG directory to the Modules directory (where they really should have been to begin with). 1/12/00 : beazley Removed the -stat option for printing statistics. The statistics reporting was inadequate and mostly broken anyway. *** POTENTIAL INCOMPATIBILITY *** 1/12/00 : beazley Removed the -t option for reading a typemap file. More trouble than it's worth. Just include typemaps at the top of the interface file. *** POTENTIAL INCOMPATIBILITY *** 1/12/00 : beazley Removed the %checkout directive. *** POTENTIAL INCOMPATIBILITY *** 1/12/00 : beazley Removed the -ci option for file checkin. Too problematic to implement. Probably better to just put your SWIG library under CVS instead. *** POTENTIAL INCOMPATIBILITY ***. 1/11/00 : beazley Deleted the LATEX module. Sorry... Didn't know anyone who was using it. Besides, I'm looking to simplify the documentation system. *** POTENTIAL INCOMPATIBILITY *** 1/11/00 : beazley Modified the ASCII documentation module to use a .txt suffix for its output file instead of .doc. 1/11/00 : beazley Added the long-lost SWIG preprocessor back to the system. It should be enabled by default. Raw preprocessed output can be viewed using swig -E file.i. *** NEW FEATURE *** 1/11/00 : beazley and djmitche Completely reorganized the SWIG directory structure. The basic organization is now: Source/ SWIG source code Lib/ SWIG library files (swig_lib) Doc/ Documentation Examples/ Examples More directories will be added as needed. 12/08/99: Loic Dachary (loic@senga.org) Enhanced package handling for perl5 and c++. With new option -hide Foo::Bar, every perl5 object (Frob) is qualified by Foo::Bar::Frob. The package name is solely used to encapsulate C/C++ wrappers output in _wrap.c and the corresponding perl package in .pm. Note that a package name may contain :: (Frob::Nitz) and will be relative to the package name provided by -hide (Foo::Bar::Frob::Nitz). In *_wrap.c, SWIG_init macro is used. Was previously defined but not used and simplifies code. Added typemap(perl5,perl5in) and typemap(perl5,perl5out) that do the equivalent of typemap(perl5,in) and typemap(perl5,out) but contain perl code and applies to wrappers generated by -shadow. Lacking proper regression tests I used Examples/perl5/{c++,constraint,defarg,except, graph/graph[1234],multinherit,nested,shadow,simple,tree, typemaps/{argv,argv2,arraymember,database,file,ignore,integer, output,passref,reference,return}}/. I ran swig with and without the patches, diff the generatedsources, run the .pl files and checked that the results are identical. In all those examples I had no error. 11/21/99: Modified the Tcl module to provide full variable linking capabilities to all datatypes. In previous versions, a pair of accessor functions were created for datatypes incompatible with the Tcl_LinkVar() function. Now, we simply use variable traces to support everything. This may break scripts that rely upon the older behavior. *** POTENTIAL INCOMPATIBILITY *** 11/21/99: Added slight tweak to wrapper generator to collect local variables of similar type. Produces somewhat more compact wrapper code. 11/20/99: Modified the Tcl module to use SWIG_GetArgs() to parse arguments. This is a technique borrowed from Python in which arguments are converted using a format string convention similiar to fprintf(). This results in a *substantial* reduction in the size of the resulting wrapper code with only a modest runtime overhead in going through the extra conversion function. 11/13/99: Completely rewrote the class/structure generation code for the Tcl module. Now, a small set of runtime functions are used to implement the functionality for all classes (instead of a massive amount of runtime code being generated for each class). Class specific information is simply encoded in a series of static tables. This results in a *HUGE* reduction in wrapper code size--especially for C++. 11/13/99: Removed the -tcl (Tcl 7.x) module. Tcl 8.0 is now several years old and the defacto standard--no real reason to keep supporting the old version at this point. 11/13/99: Cleaned up -c option for Python module. The pyexp.swg file is now gone. 11/13/99: Fixed external declarations to work better with static linking on Windows. Static linking should now be possible by defining the -DSTATIC_LINK option on the command line. Patch contributed by Alberto Fonseca. 11/5/99 : Fixed an obscure code generation bug related to the generation of default constructors. Bug reported by Brad Clements. 11/5/99 : Fixed a few memory problems found by purify. 11/5/99 : Officially deprecated the -htcl, -htk, and -plugin options from the Tcl and Tcl8 modules. 10/26/99: Removed unused variable from python/typemaps.i. Patch contributed by Keith Davidson. 8/16/99 : Added _WIN32 symbol to libraries to better support Windows. 8/16/99 : Deprecated the Perl4 module. It is no longer included in the distribution and no longer supported. In the entire 3 years SWIG has been around I never received a single comment about it so I'm assuming no one will miss it... 8/16/99 : Modified the type-checking code to register type mappings using a table instead of repeated calls to SWIG_RegisterMapping(). This reduces the size of the module initialization function somewhat. 8/15/99 : Cleaned up the pointer type-checking code in the Tcl module. 8/15/99 : Many changes to the libraries to support runtime libraries. 8/13/99 : Eliminated C++ compiler warning messages about extern "C" linkage. 8/13/99 : Some cleanup of Python .swg files to better support runtime libraries on Windows. 8/13/99 : Modified the %pragma directive to attach pragmas declared inside a class definition to the class itself. For example: class foo { ... %pragma(python) addtomethod = "insert:print `hello world'" ... } Most people don't need to worry about how this works. For people writing backend modules, class-based pragmas work like this: lang->cpp_open_class() // Open a class lang->cpp_pragma() // Supply pragmas ... // Emit members lang->cpp_close_class() // Close the class All of the pragmas are passed first since they might be used to affect the code generation of other members. Please see the Python module for an example. Patches contributed by Robin Dunn. 8/13/99 : Patch to Python shadow classes to eliminate ignored exception errors in destructors. Patch contributed by Robin Dunn. 8/11/99 : Minor patch to swig_lib/python/swigptr.swg (added SWIGSTATIC declaration). Patch contributed by Lyle Johnson. 8/11/99 : Added FIRSTKEY/NEXTKEY methods to Perl5 shadow classes Patch contributed by Dennis Marsa. 8/11/99 : Modified Python module so that NULL pointers are returned and passed as 'None.' Patch contributed by Tal Shalif. 8/10/99 : Fixed missing 'int' specifiers in various places. 8/10/99 : Added Windows makefile for Runtime libraries. Contributed by Bob Techentin. 8/10/99 : Fixed minor problem in Python runtime makefile introduced by keyword arguments. 8/8/99 : Changed $target of perl5(out) typemap from ST(0) to ST(argvi). Patch contributed by Geoffrey Hort. 8/8/99 : Fixed bug in typemap checking related to the ANY keyword in arrays and ignored arguments. Error reported by Geoffrey Hort. 8/8/99 : %enabledoc and %disabledoc directives can now be used inside class/structure definitions. However, no check is made to see if they are balanced (i.e., a %disabledoc directive inside a class does not have to have a matching %enabledoc in the same class). 8/8/99 : Keyword argument handling is now supported in the Python module. For example: int foo(char *bar, int spam, double x); Can be called from Python as foo(x = 3.4, bar="hello", spam=42) To enable this feature, run SWIG with the '-keyword' command line option. Mixing keyword and default arguments should work as well. Unnamed arguments are assigned names such as "arg1", "arg2", etc... *** POTENTIAL INCOMPATIBILITY *** Functions with duplicate argument names such as bar(int *OUTPUT, int *OUTPUT) will likely cause problematic wrapper code to be generated. To fix this, use different names or use %apply to map typemaps to alternate names. 8/8/99 : Handling of the 'this' pointer has been changed in Python shadow classes. Previously, dereferencing of '.this' occured in the Python shadow class itself. Now, this step occurs in the C wrappers using the following function: SWIG_GetPtrObj(PyObject *, void **ptr, char *type) This function can accept either a string containing a pointer or a shadow class instance with a '.this' attribute of appropriate type. This change allows the following: 1. The real shadow class instance for an object is passed to the C wrappers where it can be examined/modified by typemaps. 2. Handling of default/keyword arguments is now greatly simplified. 3. The Python wrapper code is much more simple. Plus, it eliminated more than 300 lines of C++ code in the Python module. *** CAVEAT : This requires the abstract object interface. It should work with Python 1.4, but probably nothing older than that. 8/8/99 : Fixed handling of "const" and pointers in classes. In particular, declarations such as class foo { ... const char *msg; const int *iptr; } are handled as assignable variables as opposed to constant values (this is the correct behavior in C/C++). Note: declarations such as "char *const msg" are still unsupported. Constants declared at the global level using const are also broken (because I have a number of interfaces that rely upon this behavior). *** POTENTIAL INCOMPATIBILITY *** This may break interfaces that mistakenly treat 'const char *' types as constant values. 8/8/99 : Modified the parser to support bit-fields. For example: typedef struct { unsigned int is_keyword : 1; unsigned int is_extern : 1; unsigned int is_static : 1; } flags; Bit-fields can only be applied to integer types and their are other restrictions. SWIG performs no such type-checking (although the C compiler will catch problems when it tries to compile the wrapper code). 8/8/99 : Removed trailing space of $basetype substitution in typemaps. This is to allow things like this: %typemap(python, argout) spam** OUTPUT{ ... char* a = "$basetype_p"; ... } (Patch suggested by Nathan Dunfield). 6/22/99 : Made a very slight tweak to the Perl5 shadow class code that allows typemaps to alter the return type of objects (to support polymorphic types). Patch contributed by Drake Diedrich. 4/8/99 : Fixed null pointer handling bug in Perl module. Patch contributed by Junio Hamano. 3/17/99 : Fixed bug in perl5ptr.swg for ActiveState Perl. Patch contributed by Greg Anderson. 2/27/99 : Eliminated segmentation fault when Swig runs on empty files. 2/27/99 : Added patch to Guile module to eliminate unused variables. Contributed by Mike Simons. 2/27/99 : Fixed problem with %addmethods returning references. 2/27/99 : Fixed Runtime/Makefile. Patch contributed by Mike Romberg. 2/27/99 : Incorporated patches to the type-checker. 2/27/99 : Fixed problem with -exportall switch and shadow classes in Perl5 module. Patch contributed by Dennis Marsa. 2/27/99 : Modified Perl5 module to recognize 'undef' as a NULL char *. Patch contributed by Junio Hamano. 2/27/99 : Fixed the Perl5 module to support the newer versions of ActiveState Perl for Win32. 2/27/99 : Fixed the include order of files specified with the -I option. 2/5/98- : Dave finishes his dissertation, goes job hunting, moves to 2/5/99 Chicago and generally thrashes about. Version 1.1 Patch 5 (February 5, 1998) ====================================== 2/4/98 : Fixed a bug in the configure script when different package locations are specified (--with-tclincl, etc...). 2/2/98 : Fixed name-clash bug related to the switch to C macros for accessor functions. The new scheme did not work correctly for objects with members such as 'obj', 'val', etc... Fixed the bug by appending the word 'swig' to macro argument names. Patch contributed by Rudy Albachten. 2/2/98 : Slight fix to the Perl5 module to eliminate warning messages about 'varname used only once : possible typo'. Fix contributed by Rudy Albachten. 1/9/98 : Fixed a bug in the Perl 5 module related to the creation of constants and shadow classes. 1/9/98 : Fixed linking bug with Python 1.5 embed.i library file. Version 1.1 Patch 4 (January 4, 1998) ===================================== 1/4/98 : Changed structured of the Examples directory to be more friendly to Borland C++. 1/4/98 : Added the function Makefile.win.bc for compiling the examples under Borland 5.2. 1/4/98 : Slight change to the perl5 module and C++ compilation. The library is now included before any Perl headers because Perl the extern "C" linkage of math.h screws alot of things up (especially on Windows). 1/2/98 : Change to the Python module that reduces the number of constants created by C++ classes, inheritance, and shadow classes. This modification may introduce a few slight incompatibilities if you attempt to use the non-shadow class interface with shadow classes enabled. Patch contributed by Mike Romberg. 1/2/98 : Support for Tcl 8.0 namespaces has been added. This *replaces* the original SWIG mechanism that assumed [incr Tcl] namespaces. To use namespaces, simply run SWIG with the following options swig -tcl -namespace foo.i This places everything in a namespace that matches the module name swig -tcl -namespace -prefix bar foo.i This places everything in the namespace 'bar' The use of namespaces is new in Tcl 8.0. However, the wrapper code generated by SWIG will still work with all versions of Tcl newer than and including Tcl 7.3/Tk3.6 even if the -namespace option is used. *** POTENTIAL INCOMPATIBILITY *** This change may break existing applications that relied on the -prefix and -namespace options. 1/2/98 : Added the following constants to the Tcl wrapper code SWIG_name - Name of the SWIG module SWIG_prefix - Prefix/namespace appended to command names SWIG_namespace - Name of the namespace SWIG library writers can use these to their advantages. 1/2/98 : Fixed a bug in the Tcl8 module related to the creation of pointer constants (the function SWIG_MakePtr was missing from the wrapper code). 1/2/98 : Added the consthash.i library file to the Tcl and Tcl8 modules. 1/1/98 : Changed and cleaned up the Python typemaps.i file. The following significant changes were made : 1. The OUTPUT typemap now returns Python tuples instead of lists. Lists can be returned as before by using the L_OUTPUT type. If compatibility with older versions is needed, run SWIG with the -DOUTPUT_LIST option. 2. The BOTH typemap has been renamed to INOUT. For backwards compatibility, the "BOTH" method still exists however. 3. Output typemaps now generate less code than before. Changes to typemaps.i may break existing Python scripts that assume output in the form of a list. *** POTENTIAL INCOMPATIBILITY *** 12/31/97: Fixed long overdue problems with the testing scripts and certain makefiles that required the use of the bash shell. Everything should work properly with the standard Bourne shell (sh) now. 12/31/97: Modified typemaps to allow $basetype as a valid local variable. This allows for all sorts of bizarre hackish typemaps that do cool things. Patch contributed by Dominique Dumont. 12/31/97: Switched accessor functions generated for member data to C preprocessor macros (except in cases involving typemaps or char *). 12/31/97: Fixed a bug related to C++ member data involving references. 12/31/97: Changed accessor functions for C++ member functions to preprocessor macros. This cleans up the wrapper code and results in fewer function definitions. 12/31/97: Changed the default C constructor to use calloc() instead of malloc() 12/30/97: Changed the creation of constants in the Perl5 module. For all practical purposes, they should work in exactly the same way as before except that they now require much less wrapper code. Modules containing large numbers of constants may see greater than a 50% reduction in wrapper code size. 12/30/97: Modified the Python module to be more intelligent about the creation of constants. SWIG no longer generates redundant global variables and the size of the module initialization function should be reduced. (Many thanks to Jim Fulton). 12/29/97: Fixed a bug in C++ code generation related to member functions, default arguments, and references. 12/29/97: Fixed configure script and a few makefiles to support Python 1.5 12/29/97: Added 'embed15.i' library file. This file should be used to staticly link versions of Python 1.5. To make it the default, simply copy 'swig_lib/python/embed15.i' to 'swig_lib/python/embed.i' Version 1.1 Patch 3 (November 24, 1997) ======================================== 11/23/97: Fixed a bug in the Perl5 module with shadow classes and static class functions that return class instances. Note : The fix for this bug requires a slight restructuring of of the .pm files created by SWIG. 11/23/97: Fixed a bug in the Tcl/Tcl8 modules related to variable linking of character arrays. If you declared a global variable 'char foo[10]', the generated wrapper code would either cause a segmentation fault immediately upon loading or weird memory corruption elsewhere. This should now be fixed although character arrays can only be read-only. 11/23/97: Fixed a bug with the %import directive that caused it to fail if files were imported from directories other than the current working directory. 11/23/97: Fixed incorrect diagnostic message in the ASCII documentation module. 11/23/97: Changed the behavior of the -o option when used with shadow classes. If -o was used to specify both the pathname and filename of SWIG's output such as swig -o /home/swig/wrapper.c -shadow -perl5 foo.i The wrapper code would be placed the file specified with -o, but the .pm file and documentation would be placed in the directory where SWIG was run. Now, these files are placed in the same directory as the file specified with the -o option. This change is also needed for proper operation on the Macintosh. 11/23/97: Added a 'this()' method to Perl5 shadow classes. This can be used to return the normal pointer value from a shadow class that is represented as a tied hash. To use just invoke as a method like this : $l = new List; # Create an object $ptr = $l->this(); # Get the normal pointer value *** NEW FEATURE *** 11/23/97: Fixed the Tcl 8 pointer.i library file (which was completely broken in 1.1p2). 11/23/97: Modified the Perl5 type-checker to fix a few problems with global variables of pointer types and to allow tied hashes to be used interchangably with normal pointer values. 11/23/97: Modified the typemap mechanism to allow output typemaps of type 'void'. These were ignored previously, but now if you specify, %typemap(lang,out) void { ... return a void ... } You can change or assign a return value to the function. 11/23/97: Fixed processing of 'bool' datatypes in the Python module. 11/23/97: Fixed minor parsing error with C++ initializers. For example, class B : public A { public: B() : A() { ... }; ... } 11/23/97: Fixed the Tcl8 module so that C functions that call back into Tcl don't corrupt the return result object (SWIG was gathering the result object too early which leads to problems if subsequent Tcl calls are made). 11/23/97: Fixed a code generation bug in the Python module when two or more output parameters were used as the first arguments of a function. For example : %include typemaps.i void foo(double *OUTPUT, double *OUTPUT, double a); Previously, doing this resulted in the creation of an extraneous comma in the output, resulting in a C syntax error. 11/22/97: Fixed a bug when template handling that was stripping whitespace around nested templates. For example : Foo > was getting munged into Foo> which is a syntax error in in the C++ compiler. 11/22/97: Fixed bugs in the Borland C++ makefiles. 11/22/97: Fixed memory corruption bug when processing integer arguments in Tcl8 module. 11/21/97: Fixed a bug in the Runtime/Makefile related to Tcl 8. 11/21/97: Fixed a bug with the %new directive and Perl5 shadow classes. No longer generates a perl syntax error. 11/9/97 : Changed a strncpy() to strcpy() in the pointer type-checker. This results in a substantial performance improvement in type-checking. 10/29/97: Fixed a bug in the code generation of default arguments and user-defined types. For example : void foo(Vector a, Vector b = d); should now work properly. Version 1.1 Patch 2 (September 4, 1997) ======================================= 9/4/97 : Fixed problem with handling of virtual functions that was introduced by some changes in the C++ module. Version 1.1 Patch 1 (August 27, 1997) ===================================== 8/26/97 : Fixed compilation and run-time bugs with Tcl 8.0 final. 8/21/97 : Fixed code generation bug with arrays appearing as arguments to C++ member functions. For example : class Foo { public: void Bar(int a[20][20]); }; There is still a bug using arrays with added methods however. 8/20/97 : Fixed a bug with generating the code for added methods involving pass-by-value. 8/19/97 : Modified the typemapper to substitute the '$arg' value when declaring local variables. For example : %typemap(in) double * (double temp_$arg) { ... do something ... } When applied to a real function such as the following : void foo(double *a, double *b, double *result); three local variables will be created as follows : double temp_a; double temp_b; double temp_result; This can be used when writing multiple typemaps that need to access the same local variables. 7/27/97 : Fixed a variety of problems with the %apply directive and arrays. The following types of declarations should now work : %apply double [ANY] { Real [ANY] }; %apply double [4] { double [10] }; A generic version of apply like this : %apply double { Real }; should now work--even if arrays involving doubles and Reals are used later. 7/27/97 : Changed warning message about "Array X has been converted to Y" to only appear if running SWIG in verbose mode. 7/27/97 : Added the variables $parmname and $basemangle to the typemap generator. $parmname is the name of the parameter used when the typemap was matched. It may be "" if no parameter was used. $basemangle is a mangled version of the base datatype. Sometimes used for array handling. 7/27/97 : Changed the behavior of output arguments with Python shadow classes. Originally, if a function returned an object 'Foo', the shadow class mechanism would create code like this : def return_foo(): val = FooPtr(shadowc.return_foo()) val.this = 1 return val The problem with this is that typemaps allow a user to redefine the output behavior of a function--as a result, we can no longer make any assumptions about the return type being a pointer or even being a single value for that matter (it could be a list, tuple, etc...). If SWIG detects the use of output typemaps (either "out" or "argout") it returns the result unmodified like this : def return_foo(): val = shadowc.return_foo() return val In this case, it is up to the user to figure out what to do with the return value (including the possibility of converting it into a Python class). 7/26/97 : Fixed a parsing problem with types like 'unsigned long int', 'unsigned short int', etc... 7/24/97 : Minor bug fix to Tcl 8 module to parse enums properly. Also fixed a memory corruption problem in the type-checker. (patch contributed by Henry Rowley. 7/24/97 : Added Python-tuple typemaps contributed by Robin Dunn. 7/24/97 : Incorporated some changes to the Python module in support of Mark Hammond's COM support. I'm not entirely sure they work yet however. Needs documentation and testing. 7/24/97 : Fixed code generation bugs when structures had array members and typemaps were used. For example : %typemap(memberin) double [20][20] { ... get a double [20][20] ... } struct Foo { double a[20][20]; } Originally, this would generate a compiler-type error when the wrapper code was compiled. Now, a helper function like this is generated : double *Foo_a_set(Foo *a, double val[20][20]) { ... memberin typemap here ... return (double *) val; } When writing typemaps, one can assume that the source variable is an array of the *same* type as the structure member. This may break some codes that managed to work around the array bug. *** POTENTIAL INCOMPATIBILITY *** 7/13/97 : Fixed bug in Perl5 module when using C global variables that are pointers. When used in function calls and other operations, the value of the pointer would be invalid---causing core dumps and other problems. SWIG implements global variables using Perl magic variables. As it turns out, the error was caused by the fact that the pointer-extraction code was somehow bypassing the procedure used to resolve magical variables (hence, leaving the value undefined). To fix the problem, SWIG now explicitly resolves magic before extracting pointer values. 7/12/97 : Eliminated the last remnants of free() and malloc() from the SWIG compiler. 7/12/97 : Fixed parsing problems with typemaps involving arrays and temporary variables of arrays. Also made it possible for SWIG to handle typemaps like this : %typemap(in) double [ANY] (double temp[$dim0]) { ... store data in temp[$dim0] ... } Not only does this typemap match any double [] array, it creates a local variable with precisely the right dimensions. (ie. $dim0 gets filled in with the real number of dimensions). Of course, off the record, this will be a way to add more functionality to the typemaps.i libraries. 7/9/97 : Fixed some problems with Perl5, static linking, and shadow classes. When statically linking multiple modules together, write a top-level interface file like this when shadow classes are not used : %module swig, foo, bar, glob; %include perlmain.i When shadow classes are used, the module names have an extra 'c' appended so it should read as : %module swig, fooc, barc, globc; %include perlmain.i When linking multiple modules, consider using the SWIG runtime library. 7/8/97 : Incorporated fixed versions of the Borland C++ Makefiles. 7/8/97 : First cut at trying to eliminate excessive compiler warnings. As it turns out, alot of warnings go away if you just make declarations like this clientData = clientData; in the resulting wrapper code. Most compilers should just ignore this code (at least would can hope). 7/8/97 : Fixed bizarre code generation bug with typemaps and C++ classes. In some cases, typemaps containing printf formatting strings such as %typemap(memberout) int * { printf("%d",42); } Would generate completely bogus code with garbage replacing the '%d'. Caused by one faulty use of printf (wasn't able to find any other occurences). 7/7/97 : Fixed bug in Python shadow class generation with non-member functions that are returning more than one value. 7/7/97 : Incorporated modifications to make SWIG work with Guile 1.2. Still need to test it out, but it is rumored to work. 7/2/97 : Fixed some bugs related to output arguments and Python shadow classes. If an output argument is detected, SWIG assumes that the result is a list and handles it appropriately. If the normal return type of an function is an object, it will be converted into a shadow class as before, but with the assumption that it is the first element of a list. *** NOTE : This behavior has been subsequently changed *** 6/29/97 : Changed EXPORT to SWIGEXPORT in all of the language modules. Should provide better compatibility with Windows. 6/29/97 : Modified Python shadow classes so that output arguments work correctly (when typemaps are used). Version 1.1 (June 24, 1997) =========================== 6/24/97 : Fixed Objective-C constructor bug when working with Perl5 shadow classes. 6/23/97 : Fixed some parsing problems with Objective-C. Declarations such as the following should work now : - foo : (int) a with: (int) b; 6/22/97 : Added SWIG Runtime library. This library contains the SWIG pointer type-checker and support functions that are normally included in every module. By using the library, it is easier to work with multiple SWIG generated modules. 6/22/97 : Fixed minor bug in Perl5 module related to static linking of multiple modules. 6/22/97 : Fixed some bugs with the %import directive. When used with Perl5 shadow classes, this generates a 'require' statement to load in external modules. 6/22/97 : Added -swiglib option. This prints out the location of the SWIG library and exits. This option is only really useful to configuration tools that are looking for SWIG and its library location (e.g. autoconf, configure, etc...). 6/21/97 : Fixed export bug with Perl5.004 on Windows-NT. 6/20/97 : Minor change to code generation of class/structure members in order to work better with typemaps. Should have no noticable impact on existing SWIG modules. 6/19/97 : Added -t option. This allows SWIG to load a typemap file before processing any declarations. For example : swig -t typemaps.i -python example.i At most, only one typemap file can be specified in this manner. *** NEW FEATURE *** 6/18/97 : Need a Makefile fast? Type swig [-tcl, -perl5, -python] -co Makefile and you will get a Makefile specific to that target language. You just need to modify it for your application and you're ready to run. 6/18/97 : Completed the -ci option. This option checks a file into the SWIG library. It should be used in conjunction with a language option. For example : swig -tcl -ci foobar.i Checks the file foobar.i into the Tcl part of the library. In order to check a file into the general library (accessible to all languages modules), do the following swig -ci -o ../foobar.i foobar.i (Admittedly this looks a little strange but is unavoidable). The check-in option is primarily designed for SWIG maintenance and library development. The command will fail if the user does not have write permission to the SWIG library. Third party library extensions can easily install themselves by simply providing a shell script that uses 'swig -ci' to install the appropriate library files. It is not necessary to know where the SWIG library is located if you use this mechanism. *** NEW FEATURE *** 6/16/97 : Fixed a bug in shadow class generation when %name() was applied to a class definition. Unfortunately, fixing the bug required a change in the Language C API by adding an extra argument to the Language::cpp_class_decl() function. This may break SWIG C++ extensions. *** POTENTIAL INCOMPATIBILITY *** 6/15/97 : Added a warning message if no module name is specified with the %module directive or -module option. 6/15/97 : Fixed line number bug when reporting errors for undefined base classes. 6/15/97 : Added new %rename directive. This allows the forward declaration of a renaming. For example : %rename OldName NewName; .... later ... int OldName(int); Unlike %name, %rename will rename any occurence of the old name. This applies to functions, variables, class members and so forth. There is no way to disable %rename once set, but you can change the name by redeclaring it to something else. *** NEW FEATURE *** 6/15/97 : Improved the implementation of the %name directive so that it could be used with conditional compilation : #ifdef SWIG %name(NewName) #endif int OldName(int); 6/15/97 : Added support for functions with no return datatype. In this case, SWIG assumes a return type of 'int'. 6/11/97 : Improved error reporting in the parser. It should be a little less sensitive to errors that occur inside class definitions now. Also reports errors for function pointers. 6/11/97 : Made '$' a legal symbol in identifiers. This is to support some Objective-C libraries. Some compilers (such as gcc) may also allow identifiers to contain a $ in C/C++ code as well (this is an obscure feature of C). When '$' appears in identifier, SWIG remaps it to the string '_S_' when creating the scripting language function. Thus a function 'foo$bar' would be called 'foo_S_bar'. 6/11/97 : Fixed bug in Python shadow classes with __repr__ method. If supplied by the user, it was ignored, but now it should work. 6/9/97 : Fixed the Tcl 8.0 module to work with Tcl 8.0b1. SWIG is no longer compatible with *any* alpha release of Tcl 8.0. *** POTENTIAL INCOMPATIBILITY *** 6/7/97 : Put a maximal error count in (currently set to 20). SWIG will bail out if it generates more errors than this (useful for preventing SWIG from printing 4000 syntax errors when it gets confused). 6/7/97 : Fixed segmentation fault when parsing variable length arguments. 6/7/97 : Minor change to Perl5 module. C++ static functions are now put in the same package as their class when using shadow classes. 6/7/97 : Centralized the naming of functions, members, wrappers etc... By centralizing the naming scheme, it should be possible to make some multi-file optimizations. Also, it should be possible to change SWIG's naming scheme (perhaps a new feature to be added later). 6/2/97 : Added 'arginit' typemap. This can be used to assign initial values to function arguments. Doing so makes it somewhat easier to detect improper argument passing when working with other typemaps. 6/2/97 : Fixed code generation bug when read-only variables were inherited into other classes. Under inheritance, the variables would become writable, but this has now been corrected. 5/30/97 : An empty %name() directive is no longer allowed or supported. This directive was originally used to strip the prefix off of a class or structure. Unfortunately, this never really seemed to work right and it complicated the C++ code generator significantly. As far as I can tell no one uses it, so it is now history. *** POTENTIAL INCOMPATIBILITY *** 5/28/97 : Fixed a parsing bug with #define and C++ comments. Declarations such as the following now work properly : #define CONST 4 // A Comment 5/28/97 : Made some performance improvements to the SWIG String class. (only affects the SWIG compiler itself). 5/28/97 : Modified the parser to skip template definitions and issue a warning message. 5/28/97 : Preliminary support for parameterized types added (ie. templates). Types such as the following should pass through the SWIG compiler void foo(vector *a, vector *b); When used, the entire name 'vector' becomes the name of the datatype. Due to space limitations in datatype representations, the name should not exceed 96 characters. Note : This is only part of what is needed for template support. Template class definitions are not yet supported by SWIG. The template notation above may also be used when specifying Objective-C protocol lists. *** NEW FEATURE *** 5/24/97 : First cut at Objective-C support added. As it turns out, almost everything can be handled with only a few minor modifications to the C++ module. *** NEW FEATURE *** 5/23/97 : Fixed repeated definition bug in multiple inheritance handling when multiple base classes share a common base class (ie. the evil diamond). 5/21/97 : Fixed rather embarrassing typo that worked its way into the Tests/Build directory. 5/19/97 : Fixed code generation bug when using native methods and shadow classes with Python and Perl5 modules. 5/19/97 : Modified the %apply directive slightly so that it would work with pointers a little better. For example : %apply unsigned long { DWORD }; Applies *all* typemaps associated with "unsigned long" to "DWORD". This now includes pointers to the two datatypes. For example, a typemap applied to "unsigned long **" would also be applied to any occurrence of "DWORD **" as well. 5/19/97 : Fixed an ownership assignment bug in the Perl5 module when class members were returning new objects belonging to different classes. 5/17/97 : Added a few more typemap variables. $name - Name of function/variable/member $basetype - Base datatype (type without pointers) $argnum - Argument number 5/16/97 : Fixed embarrassing underscore error in local variable allocator. 5/16/97 : Fixed namespace clash bug in parameterized typemaps when creating arrays as new local variables. 5/15/97 : Fixed some bugs with inheritance of added methods across multiple files. SWIG now uses names of base classes when generating such functions. 5/14/97 : Finished support for default typemaps. Primarily used internally, they can be used to match the basic built-in datatypes used inside of SWIG. You can specify them in interface files as well like this : %typemap(tcl,in) int SWIG_DEFAULT_TYPE { $target = atoi($target); } Unlike normal typemaps, this default map will get applied to *all* integer datatypes encountered, including those renamed with typedef, etc... 5/13/97 : Fixed substring bug in type checker. 5/12/97 : Fixed bug in parameterized typemaps when declaring local variables of structures. Version 1.1 Beta6 (May 9, 1997) =============================== 5/9/97 : Fixed bizarre NULL pointer handling bug in Perl5 module. 5/8/97 : Fixed mysterious segmentation fault when running SWIG on an empty file. 5/7/97 : The code generator will now replace the special symbol "$cleanup" with the cleanup code specified with the "freearg" typemap. This change needed to properly manage memory and exceptions. 5/5/97 : Added the 'typemaps.i' library file. This contains a variety of common typemaps for input values, pointers, and so on. 5/5/97 : Changed behavior of "argout" typemap in Python module. Old versions automatically turned the result into a Python list. The new version does nothing, leaving the implementation up to the user. This provides more flexibility but may break older codes that rely on typemaps. *** POTENTIAL INCOMPATIBILITY *** 5/5/97 : Fixed bug in Python module related to the interaction of "argout" and "ignore" typemaps. 5/5/97 : Fixed bug in Python module that would generate incorrect code if all function arguments are "ignored". 5/4/97 : Added %apply and %clear directives. These form a higher level interface to the typemap mechanism. In a nutshell, they can be used to change the processing of various datatypes without ever having to write a typemap. See the SWIG documentation for more details. ** NEW FEATURE ** 5/4/97 : Added a local variable extension to the typemap handler. For example : %typemap(tcl,in) double *(double temp) { temp = atof($source); $target = &temp; } In this case, 'temp' is a local variable that exists in the entire wrapper function (not just the typemap code). This mechanism provides better support for certain types of argument handling and also makes it possible to write thread-safe typemaps. Any number local variables can be declared by supplying a comma separated list. Local variables are guaranteed to be unique, even if the same typemap is applied many times in a given function. ** Not currently supported in Perl4 or Guile modules. 5/2/97 : Fixed processing of %ifdef, %endif, %if, etc... (These are SWIG equivalents of the C preprocessor directives that can pass through the C preprocessor without modification). 5/2/97 : Fixed major (but subtle) bug in the run-time type checker related to searching and type-checking for C++ inheritance. To make a long story short, if you had two classes named "Foo" and "FooObject" the type checker would sometimes get confused and be unable to locate "Foo" in an internal table. 5/2/97 : Fixed some bugs in the -co option. 4/24/97 : Pointer library added to the SWIG library. 4/19/97 : Added the %new directive. This is a "hint" that can be used to tell SWIG that a function is returning a new object. For example : %new Foo *create_foo(); This tells SWIG that create_foo() is creating a new object and returning a pointer to it. Many language modules may choose to ignore the hint, but when working with shadow classes, the %new is used to handle proper ownership of objects. %new can also be used with dynamically allocated strings. For example : %new char *create_string(); When used, all of the language modules will automatically cleanup the returned string--eliminating memory leaks. ** NEW FEATURE ** 4/19/97 : Added a new typemap "newfree". This is used in conjunction with the %new directive and can be used to change the method by which a new object returned by a function is deleted. 4/19/97 : The symbol "__cplusplus" is now defined in the SWIG interpreter when running with the -c++ option. 4/17/97 : Added support for static member functions when used inside the %addmethods directive. 4/15/97 : Added a special typemap symbol PREVIOUS that can be used to restore a previous typemap. For example : %typemap(tcl,in) int * = PREVIOUS; This is primarily used in library files. 4/13/97 : Added %pragma directive for Perl5 module. Two new pragmas are available : %pragma(perl5) code = "string" %pragma(perl5) include = "file.pl" Both insert code into the .pm file created by SWIG. This can be used to automatically customize the .pm file created by SWIG. 4/13/97 : Scanner modified to only recognize C++ keywords when the -c++ option has been specified. This provides support for C programs that make use of these keywords for identifiers. SWIG may need to be explicitly run with the -c++ option when compiling C++ code (this was allowed, but not recommended in previous versions). **POTENTIAL INCOMPATIBILITY** 4/11/97 : Fixed a rather nasty bug in the Perl5 module related to using variable linking with complex datatypes and pointers. On Unix, code would work (somehow), but would cause an access violation under Windows-NT. The fix should correct the problem, but there may still be a problem using global variables of complex datatypes in conjunction with shadow classes. Fortunately, this sort of thing seems to be relatively rare (considering that the bug has been around for more than a year - yikes!). 4/11/97 : Fixed bizarre constant evaluation bug in Perl5 code generation when running under Windows-NT. 4/8/97 : Bug when using default arguments and C++ references fixed. 4/8/97 : Fixed code generation bugs in Python and Perl5 modules related to using class renaming (applying the %name directive to a class definition) and shadow classes. 4/7/97 : Fixed minor bugs in swigptr.swg, tcl8ptr.swg, and perl5ptr.swg to prevent infinite loops when weird datatypes are passed. 3/29/97 : 'Makefile.win' added. This is used to build most of the examples in the Examples directory under Windows NT/95. 3/27/97 : Fixes to SWIG's error return codes. SWIG now returns non-zero exit codes for certain kinds of errors (which makes it more friendly to makefiles). An overhaul of the error handling is on the to-do list and will probably show up in a later release. 3/25/97 : Bug fix. "freearg" and "argout" typemaps have been fixed in the Perl5 module. In previous versions, function input parameters and function output parameters shared the same memory space--causing all sorts of nasty problems when trying to pass perl values by reference. SWIG now internally makes a "copy" (which is really just a pointer) of affected parameters and uses that. This is done transparently so there is no noticable impact on any SWIG generated modules. This change is probably only noticable to expert users. 3/25/97 : Added type-check to verbose and stat mode. SWIG will now generate a list of all datatypes that were used but undefined (useful for tracking down weird bugs). This is enabled with the -v option (which is now officially known as "overly verbose" mode) or the -stat option. 3/25/97 : Slight change to the parser to make include guards work correctly. For example : #ifndef INTERFACE_I #define INTERFACE_I %module foobar.i ... declarations ... #endif 3/24/97 : %checkout directive added. This allows an interface file to extract files from the SWIG library and place them in the current directory. This can be used to extract scripts and other helper code that might be associated with library files. For example : %checkout array.tcl Will look for a file "array.tcl" in the library and copy it to the current directory. If the file already exists in the directory, this directive does nothing (it will not overwrite an existing file). This only an experimental feature for now. 3/24/97 : SWIG will now look in the SWIG Library for a file if it can't find it in the current directory. As a result, it is easy to make modules from SWIG library files. For example, if you want to make a Python module from the SWIG timers library, just type this in any directory : swig -python timers.i You will get the files timers_wrap.c and timers_wrap.doc in the current directory that you can now compile. The file remains in the SWIG library (although you can check it out using the -co option). *** New Feature *** 3/24/97 : -co option added to SWIG to allow easy access to the SWIG library. When used, this instructs SWIG to check out a library file and place it in the current directory. For example : unix > swig -co array.i array.i checked out from the SWIG library unix > Once in your directory you can customize the file to suit your particular purposes. The checkout option makes it easy to grab library files without knowing anything about the SWIG installation, but it also makes it possible to start including scripts, C code, and other miscellaneous files in the library. For example, you could put a cool script in the library and check it out whenever you wanted to use it. *** New Feature *** 3/24/97 : #pragma export directives added to Tcl output for compiling shared libraries on the Mac. 3/24/97 : Minor changes to wish.i and tclsh.i library files to provide support for the Macintosh. 3/19/97 : SWIG's policy towards NULL pointers has been relaxed. The policy of requiring a special compiler directive -DALLOW_NULL to use NULL pointers is no longer supported. While this may seem "unsafe", it turns out that you can use a "check" typemap to achieve some safety. For example : %typemap(perl5,check) Node * { if (!$target) croak("NULL Pointers not allowed."); } This prevents any NULL value of a "Node *" pointer to be passed to a function. (I think this is much cleaner than the old -DALLOW_NULL hack anyways). 3/19/97 : Fixed pointer handling errors in Perl5 module. Modules no longer core dump when a Perl reference is inadvertently passed in as a C pointer. 3/18/97 : Added a "check" typemap. This can be used to check the validity of function input values. For example : %typemap(perl5,check) int posint { if ($target < 0) croak("Argument is not a positive integer"); } 3/18/97 : Added an $arg variable to Tcl typemaps. This makes it easier to return argument values by "reference". 3/18/97 : Fixed a code generation bug when using C++ references and the %addmethods directive. 3/18/97 : Fixed a few glitches in the typemap module with respect to chaining. For example : %typemap(tcl,in) int { $in // Inserts prexisting typemap printf("Received a %d\n", $target); } This has been allowed for quite some time, but didn't work if no existing typemap was defined. Now, it still doesn't work if no existing typemap is defined, but it issues a warning message. There is some support using default typemaps, but none of the language modules take advantage of it. This should be considered experimental at this time. Version 1.1b5 Patch 1 (March 16, 1997) ====================================== 3/16/97 : Fixed references bug with C++ code generation. 3/16/97 : Fixed initialization bug in the documentation system that was causing weird problems. 3/16/97 : Fixed fatal bug with -c option in the Python module. 3/13/97 : Fixed bug in the documentation system involving the %text directive and sorting. In the old system, %text entries would float to the top of a section because they were "nameless". Now they are attached to the previous declaration and will stay in the proper location relative to the previous entry. Version 1.1b5 (March 12, 1997) ============================== 3/11/97 : Fixed compilation problems introduced by Tcl/Tk 8.0a2. *** INCOMPATIBILITY *** SWIG no longer works with Tcl/Tk 8.0a1. 3/10/97 : Fixed bug with ignored arguments and C++ member functions in the Python module. 3/9/97 : Parsing bugs with nested class definitions and privately declared nested class definitions fixed. 3/9/97 : Fixed a few minor code generation bugs with C++ classes and constructors. In some cases, the resulting wrapper code would not compile properly. SWIG now attempts to use the default copy constructor instead. 3/8/97 : Added a -l option to SWIG that allows additional SWIG library files to be grabbed without having them specified in the interface file. This makes it easier to keep the interface file clean and move certain options into a Makefile. For example : swig -tcl example.i # Build a normal Tcl extension swig -tcl -lwish.i example.i # Build it as a wish extension # by including the 'wish.i' file. swig -python example.i # Build a dynamically loaded extension swig -python -lembed.i example.i # Build a static extension These kinds of options could previously be accomplished with conditional compilation such as : %module example ... #ifdef STATIC %include embed.i #endif 3/8/97 : Incorporated changes to Guile module to use the new gh interface in FSF Guile 1.0. The older gscm interface used in Cygnus Guile releases is no longer supported by SWIG. 3/8/97 : Cleaned up the Tcl Netscape plugin example. It should work with version 1.1 of the plugin now. 3/8/97 : Added better array support to the typemap module. The keyword ANY can now be used to match any array dimension. For example : %typemap(tcl,in) double [ANY] { ... get an array ... } This will match any single-dimensional double array. The array dimension is passed in the variables $dim0, $dim1, ... $dim9. For example : %typemap(tcl,in) double [ANY][ANY][ANY] { printf("Received a double[%d][%d][%d]\n",$dim0,$dim1,$dim2); } Any typemap involving a specific array dimension will override any specified with the ANY tag. Thus, a %typemap(tcl,in) double [5][4][ANY] {} would override a double [ANY][ANY][ANY]. However, overuse of the ANY tag in arrays of high-dimensions may not work as you expect due to the pattern matching rule used. For example, which of the following typemaps has precedence? %typemap(in) double [ANY][5] {} // Avoid this! %typemap(in) double [5][ANY] {} 3/7/97 : Fixed a number of bugs related to multi-dimensional array handling. Typedefs involving multi-dimensional arrays now works correctly. For example : typedef double MATRIX[4][4]; ... extern double foo(MATRIX a); Typecasting of pointers into multi-dimensional arrays is now implemented properly when making C/C++ function calls. 3/6/97 : Fixed potentially dangerous bug in the Tcl Object-oriented interface. Well, actually, didn't fix it but issued a Tcl error instead. The bug would manifest itself as follows: % set l [List] # Create an object ... % set m [List -this $l] # Make $m into an object assuming $l # contains a pointer. # Since $m == $l, $l gets destroyed # (since its the same command name) % $m insert Foo Segmentation fault # Note : the list no longer exists! Now, an error will be generated instead of redefining the command. As in : % set l [List] ... % set m [List -this $l] Object name already exists! Use catch{} to ignore the error. 3/3/97 : Better support for enums added. Datatypes of 'enum MyEnum' and typedefs such as 'typedef enum MyEnum Foo;' now work. 3/3/97 : Parser modified to ignore constructor initializers such as : class Foo : public Bar { int a,b; public: Foo(int i) : a(0), b(i), Bar(i,0) { }; }; 3/3/97 : Modified parser to ignore C++ exception specifications such as : int foo(double) throw(X,Y); 3/3/97 : Added %import directive. This works exactly like %extern except it tells the language module that the declarations are coming from a separate module. This is usually only needed when working with shadow classes. 3/2/97 : Changed pointer type-checker to be significantly more efficient when working with derived datatypes. This has been accomplished by storing type-mappings in sorted order, using binary search schemes, and caching recently used datatypes. For SWIG generated C++ modules that make a large number of C function calls with derived types, this could result in speedups of between 100 and 50000 percent. However, due to the required sorting operation, module loading time may increased slightly when there are lots of datatypes. 3/2/97 : Fixed some C++ compilation problems with Python embed.i library files. 2/27/97 : Slight change to C++ code generation to use copy constructors when returning complex data type by value. 2/26/97 : Fixed bug in Python module with -c option. 2/26/97 : Slight tweak of parser to allow trailing comma in enumerations such as enum Value (ALE, STOUT, LAGER, }; 2/25/97 : Fixed code generation bug in Tcl module when using the %name() directive on a classname. 2/25/97 : Finished code-size optimization of C++ code generation with inheritance of attributes. Inherited attributes now only generate one set of wrapper functions that are re-used in any derived classes. This could provide big code size improvements in some scripting language interfaces. 2/25/97 : Perl5 module modified to support both the Unix and Windows versions. The windows version has been tested with the Activeware port of Perl 5.003 running under Windows 95. The C source generated by SWIG should compile without modification under both versions of Perl, but is now even more hideous than before. 2/25/97 : Modified parser to allow scope resolution operation to appear in expressions and default arguments as in : void foo(int a = Bar::defvalue); 2/25/97 : Fixed bug when resolving symbols inside C++ classes. For example : class Foo { public: enum Value {ALE, STOUT, LAGER}; ... void defarg(Value v = STOUT); }; 2/24/97 : Fixed bug with member functions returning void *. 2/23/97 : Modified Python module to be better behaved under Windows - Module initialization function is now properly exported. It should not be neccessary to explicitly export this function yourself. - Bizarre compilation problems when compiling the SWIG wrapper code as ANSI C under Visual C++ 4.x fixed. - Tested with both the stock Python-1.4 distribution and Pythonwin running under Win95. 2/19/97 : Fixed typedef handling bug in Perl5 shadow classes. 2/19/97 : Added exception support. To use it, do the following : %except(lang) { ... try part of the exception ... $function ... catch part of exception ... } $function is a SWIG variable that will be replaced by the actual C/C++ function call in a wrapper function. Thus, a real exception specification might look like this : %except(perl5) { try { $function } catch (char *& sz) { ... process an exception ... } catch(...) { croak("Unknown exception. Bailing out..."); } } 2/19/97 : Added support for managing generic code fragments (needed for exceptions). 2/19/97 : Fixed some really obscure typemap scoping bugs in the C++ handler. 2/18/97 : Cleaned up perlmain.i file by removing some problematic, but seemingly unnecessary declarations. 2/18/97 : Optimized handling of member functions under inheritance. SWIG can now use wrapper functions generated for a base class instead of regenerating wrappers for the same functions in a derived class. This could make a drastic reduction in wrapper code size for C++ applications with deep inheritance hierarchies and lots of functions. 2/18/97 : Additional methods specified with %addmethods can now be inherited along with normal C++ member functions. 2/18/97 : Minor internal fixes to make SWIG's string handling a little safer. 2/16/97 : Moved some code generation of Tcl shadow classes to library files. 2/16/97 : Fixed documentation error of '-configure' method in Tcl modules. 2/16/97 : Modified Perl5 module slightly to allow typemaps to use Perl references. 2/12/97 : Fixed argument checking bug that was introduced by default arguments (function calls with too many arguments would still be executed). Functions now must have the same number of arguments as C version (with possibility of default/optional arguments still supported). 2/12/97 : Fixed default argument bug in Perl5 module when generating wrapper functions involving default arguments of complex datatypes. 2/12/97 : Fixed typemap scoping problems. For example : %typemap(tcl,in) double { .. get a double .. } class Foo { public: double bar(double); } %typemap(tcl,in) double { .. new get double .. } Would apply the second typemap to all functions in Foo due to delayed generation of C++ wrapper code (clearly this is not the desired effect). Problem has been fixed by assigning unique numerical identifiers to every datatype in an interface file and recording the "range of effect" of each typemap. 2/11/97 : Added support for "ignore" and "default" typemaps. Only use if you absolutely know what you're doing. 2/9/97 : Added automatic creation of constructors and destructors for C structs and C++ classes that do not specify any sort of constructor or destructor. This feature can be enabled by running SWIG with the '-make_default' option or by inserting the following pragma into an interface file : %pragma make_default The following pragma disables automatic constructor generation %pragma no_default 2/9/97 : Added -make_default option for producing default constructors and destructors for classes without them. 2/9/97 : Changed the syntax of the SWIG %pragma directive to %pragma option=value or %pragma(lang) option=value. This change makes the syntax a little more consistent between general pragmas and language-specific pragmas. The old syntax still works, but will probably be phased out (a warning message is currently printed). 2/9/97 : Improved Tcl support of global variables that are of structures, classes, and unions. 2/9/97 : Fixed C++ compilation problem in Python 'embed.i' library file. 2/9/97 : Fixed missing return value in perlmain.i library file. 2/9/97 : Fixed Python shadow classes to return an AttributeError when undefined attributes are accessed (older versions returned a NameError). 2/9/97 : Fixed bug when %addmethods is used after a class definition whose last section is protected or private. 2/8/97 : Made slight changes in include file processing to support the Macintosh. 2/8/97 : Extended swigmain.cxx to provide a rudimentary Macintosh interface. It's a really bad interface, but works until something better is written. 1/29/97 : Fixed type-casting bug introduced by 1.1b4 when setting/getting the value of global variables involving complex data types. 1/29/97 : Removed erroneous white space before an #endif in the code generated by the Python module (was causing errors on DEC Alpha compilers). 1/26/97 : Fixed errors when using default/optional arguments in Python shadow shadow classes. 1/23/97 : Fixed bug with nested %extern declarations. 1/21/97 : Fixed problem with typedef involving const datatypes. 1/21/97 : Somewhat obscure, but serious bug with having multiple levels of typedefs fixed. For example : typedef char *String; typedef String Name; Version 1.1 Beta4 (January 16, 1997) ==================================== Note : SWIG 1.1b3 crashed and burned shortly after take off due to a few major run-time problems that surfaced after release. This release should fix most, if not all, of those problems. 1/16/97 : Fixed major memory management bug on Linux 1/14/97 : Fixed bug in functions returning constant C++ references. 1/14/97 : Modified C++ module to handle datatypes better. 1/14/97 : Modified parser to allow a *single* scope resolution operator in datatypes. Ie : Foo::bar. SWIG doesn't yet handle nested classes, so this should be sufficient for now. 1/14/97 : Modified parser to allow typedef inside a C++ class. 1/14/97 : Fixed some problems related to datatypes defined inside a C++ class. SWIG was not generating correct code, but a new scoping mechanism and method for handling datatypes inside a C++ class have been added. 1/14/97 : Changed enumerations to use the value name instead of any values that might have appeared in the interface file. This makes the code a little more friendly to C++ compilers. 1/14/97 : Removed typedef bug that made all enumerations equivalent to each other in the type checker (since it generated alot of unnecessary code). Version 1.1 Beta3 (January 9, 1997) ==================================== Note : A *huge* number of changes related to ongoing modifications. 1. Support for C++ multiple inheritance added. 2. Typemaps added. 3. Some support for nested structure definitions added. 4. Default argument handling added. 5. -c option added for building bare wrapper code modules. 6. Rewrote Pointer type-checking to support multiple inheritance correctly. 7. Tcl 8.0 module added. 8. Perl4 and Guile modules resurrected from the dead (well, they at least work again). 9. New Object Oriented Tcl interface added. 10. Bug fixes to Perl5 shadow classes. 11. Cleaned up many of the internal modules of the parser. 12. Tons of examples and testing modules added. 13. Fixed bugs related to use of "const" return values. 14. Fixed bug with C++ member functions returning void *. 15. Changed SWIG configuration script. Version 1.1 Beta2 (December 3, 1996) ==================================== 1. Completely rewrote the SWIG documentation system. The changes involved are too numerous to mention. Basically, take everything you knew about the old system, throw them out, and read the file Doc/doc.ps. 2. Limited support for #if defined() added. 3. Type casts are now allowed in constant expressions. ie #define A (int) 3 4. Added support for typedef lists. For example : typedef struct { double x,y,z; } Vector, *VectorPtr; 5. New SWIG directives (related to documentation system) %style %localstyle %subsection %subsubsection 6. Reorganized the C++ handling and made it a little easier to work with internally. 7. Fixed problem with inheriting data members in Python shadow classes. 8. Fixed symbol table problems with shadow classes in both Python and Perl. 9. Fixed annoying segmentation fault bug in wrapper code generated for Perl5. 10. Fixed bug with %addmethods directive. Now it can be placed anywhere in a class. 11. More test cases added to the SWIG self-test. Documentation tests are now performed along with other things. 12. Reorganized the SWIG library a little bit and set it up to self-document itself using SWIG. 13. Lots and lots of minor bug fixes (mostly obscure, but bugs nonetheless). Version 1.1 Beta1 (October 30, 1996) ==================================== 1. Added new %extern directive for handling multiple files 2. Perl5 shadow classes added 3. Rewrote conditional compilation to work better 4. Added 'bool' datatype 5. %{,%} block is now optional. 6. Fixed some bugs in the Python shadow class module 7. Rewrote all of the SWIG tests to be more informative (and less scary). 8. Rewrote parameter list handling to be more memory efficient and flexible. 9. Changed parser to ignore 'static' declarations. 10. Initializers are now ignored. For example : struct FooBar a = {3,4,5}; 11. Somewhat better parsing of arrays (although it's usually just a better error message now). 12. Lot's of minor bug fixes. Version 1.0 Final (August 31, 1996) =================================== 1. Fixed minor bug in C++ module 2. Fixed minor bug in pointer type-checker when using -DALLOW_NULL. 3. Fixed configure script to work with Python 1.4beta3 4. Changed configure script to allow compilation without yacc or bison. Version 1.0 Final (August 28, 1996) =================================== 1. Changed parser to support more C/C++ datatypes (well, more variants). Types like "unsigned", "short int", "long int", etc... now work. 2. "unions" added to parser. 3. Use of "typedef" as in : typedef struct { double x,y,z; } Vector; Now works correctly. The name of the typedef is used as the structure name. 4. Conditional compilation with #ifdef, #else, #endif, etc... added. 5. New %disabledoc, %enabledoc directives allow documentation to selectively be disabled for certain parts of a wrapper file. 6. New Python module supports better variable linking, constants, and shadow classes. 7. Perl5 module improved with better compatibility with XS and xsubpp. SWIG pointers and now created so that they are compatible with xsubpp pointers. 8. Support for [incr Tcl] namespaces added to Tcl module. 9. %pragma directive added. 10. %addmethods directive added. 11. %native directive added to allow pre-existing wrapper functions to be used. 12. Wrote configure script for SWIG installation. 13. Function pointers now allowed with typedef statements. 14. %typedef modified to insert a corresponding C typedef into the output file. 15. Fixed some problems related to C++ references. 16. New String and WrapperFunction classes add to make generating wrapper code easier. 17. Fixed command line option processing to eliminate core dumps and to allow help messages. 18. Lot's of minor bug fixes to almost all code modules Version 1.0 Beta 3 (Patch 1) July 17, 1996 ========================================== 1.0 Final is not quite ready yet, but this release fixes a number of immediate problems : 1. Compiler errors when using -strict 1 type checking have been fixed. 2. Pointer type checker now recognizes pointers of the form _0_Type correctly. 3. A few minor fixes were made in the Makefile Version 1.0 Beta 3 (June 14, 1996) =================================== There are lots of changes in this release : 1. SWIG is now invoked using the "swig" command instead of "wrap". Hey, swig sounds cooler. 2. The SWIG_LIB environment variable can be set to change the location where SWIG looks for library files. 3. C++ support has been added. You should use the -c++ option to enable it. 4. The %init directive has been replaced by the %module directive. %module constructs a valid name for the initialization function for whatever target language you're using (actually this makes SWIG files a little cleaner). The old %init directive still works. 5. The syntax of the %name directive has been changed. Use of the old one should generate a warning message, but may still work. 6. To support Tcl/Tk on non-unix platforms, SWIG imports a file called swigtcl.cfg from the $(SWIG_LIB)/tcl directory. I don't have access to an NT machine, but this file is supposedly allows SWIG to produce wrapper code that compiles on both UNIX and non UNIX machines. If this doesn't work, you'll have to edit the file swigtcl.cfg. Please let me know if this doesn't work so I can update the file as necessary. 7. The SWIG run-time typechecker has been improved. You can also now redefine how it works by supplying a file called "swigptr.cfg" in the same directory as your SWIG interface files. By default, SWIG reads this file from $(SWIG_LIB)/config. 8. The documentation system has been changed to support the following : - Documentation order is printed in interface file order by default. This can be overridden by putting an %alpha directive in the beginning of the interface file. - You can supply additional documentation text using %text %{ put your text here %} - A few minor bugs were fixed. 9. A few improvements have been made to the handling of command line options (but it's still not finished). 10. Lots of minor bug fixes in most of the language modules have been made. 11. Filenames have been changed to 8.3 for compatibility with a SWIG port to non-unix platforms (work in progress). 12. C++ file suffix is now .cxx (for same reason). 13. The documentation has been upgraded significantly and is now around 100 pages. I added new examples and a section on C++. The documentation now includes a Table of Contents. 14. The SWIG Examples directory is still woefully sparse, but is getting better. Special notice about C++ ------------------------ This is the first version of SWIG to support C++ parsing. Currently the C++ is far from complete, but seems to work for simple cases. No work has been done to add special C++ processing to any of the target languages. See the user manual for details about how C++ is handled. If you find problems with the C++ implementation, please let me know. Expect major improvements in this area. Note : I have only successfully used SWIG and C++ with Tcl and Python. Notice about Version 1.0Final ----------------------------- Version 1.0B3 is the last Beta release before version 1.0 Final is released. I have frozen the list of features supported in version 1.0 and will only fix bugs as they show up. Work on SWIG version 2.0 is already in progress, but is going to result in rather significant changes to SWIG's internal structure (hopefully for the better). No anticipated date for version 2.0 is set, but if you've got an idea, let me know. Version 1.0 Beta 2 (April 26, 1996) =================================== This release is identical to Beta1 except a few minor bugs are fixed and the SWIG library has been updated to work with Tcl 7.5/Tk 4.1. A tcl7.5 examples directory is now included. - Fixed a bug in the Makefile that didn't install the libraries correctly. - SWIG Library files are now updated to work with Tcl 7.5 and Tk 4.1. - Minor bug fixes in other modules. Version 1.0 Beta 1 (April 10, 1996). ===================================== This is the first "semi-official" release of SWIG. It has a number of substantial improvements over the Alpha release. These notes are in no particular order--hope I remembered everything.... 1. Tcl/Tk SWIG is known to work with Tcl7.3, Tk3.6 and later versions. I've also tested SWIG with expect-5.19. Normally SWIG expects to use the header files "tcl.h" and "tk.h". Newer versions of Tcl/Tk use version numbers. You can specify these in SWIG as follows : % wrap -htcl tcl7.4.h -htk tk4.0.h example.i Of course, I prefer to simply set up symbolic links between "tcl.h" and the most recent stable version on the machine. 2. Perl4 This implementation has been based on Perl-4.035. SWIG's interface to Perl4 is based on the documentation provided in the "Programming Perl" book by Larry Wall, and files located in the "usub" directory of the Perl4 distribution. In order to compile with Perl4, you'll need to link with the uperl.o file found in the Perl4 source directory. You may want to move this file to a more convenient location. 3. Perl5 This is a somewhat experimental implementation, but is alot less buggy than the alpha release. SWIG operates independently of the XS language and xsubpp supplied with Perl5. Currently SWIG produces the necessary C code and .pm file needed to dynamically load a module into Perl5. To support Perl5's notion of modules and packages (as with xsubpp), you can use the following command line options : % wrap -perl5 -module MyModule -package MyPackage example.i Note : In order for dynamic loading to be effective, you need to be careful about naming. For a module named "MyModule", you'll need to create a shared object file called "MyModule.so" using something like % ld -shared my_obj.o -o MyModule.so The use of the %init directive must match the module name since Perl5 calls a function "boot_ModuleName" in order to initialize things. See the Examples directory for some examples of how to get things to work. 4. Python1.3 This is the first release supporting Python. The Python port is experimental and may be rewritten. Variable linkage is done through functions which is sort of a kludge. I also think it would be nice to import SWIG pointers into Python as a new object (instead of strings). Of course, this needs a little more work. 5. Guile3 If you really want to live on the edge, pick up a copy of Guile-iii and play around with this. This is highly experimental---especially since I'm not sure what the official state of Guile is these days. This implementation may change at any time should I suddenly figure out better ways to do things. 6. Extending SWIG SWIG is written in C++ although I tend to think of the code as mostly being ANSI C with a little inheritance thrown in. Each target language is implemented as a C++ class that can be plugged into the system. If you want to add your own modifications, see Appendix C of the user manual. Then take a look at the "user" directory which contains some code for building your own extenions. 7. The SWIG library The SWIG library is still incomplete. Some of the files mentioned in the user manual are unavailable. These files will be made available when they are ready. Subscribe to the SWIG mailing list for announcements and updates. 8. SWIG Documentation I have sometimes experienced problems viewing the SWIG documentation in some postscript viewers. However, the documentation seems to print normally. I'm working on making much of the documentation online, but this takes time. Version 0.1 Alpha (February 9, 1996) ==================================== 1. Run-time type-checking of SWIG pointers. Pointers are now represented as strings with both numeric and encoded type information. This makes it a little harder to shoot yourself in the foot (and it eliminates some segmentation faults and other oddities). 2. Python 1.3 now supported. 3. #define and enum can be used to install constants. 4. Completely rewrote the %include directive and made it alot more powerful. 5. Restructured the SWIG library to make it work better. 6. Various bug fixes to Tcl, Perl4, Perl5, and Guile implementations. 7. Better implementation of %typedef directive. 8. Made some changes to SWIG's class structure to make it easier to expand. SWIG is now built into a library file that you can use to make your own extenions. 9. Made extensive changes to the documentation. 10. Minor changes to the SWIG parser to make it use less memory. Also took out some extraneous rules that were undocumented and didn't work in the first place. 11. The SWIG library files "tclsh", "wish", "expect", etc... in the first release have been restructured and renamed to "tclsh.i", "wish.i", and so on.