Boost C++ Libraries: Ticket #2776: [fix in git] Failed to wrap classes with virtual inheritance

attachment set

Wed, 18 Feb 2009 12:03:07 GMT

attachment → test.hxx

attachment set

Wed, 18 Feb 2009 12:03:28 GMT

attachment → test.cxx

attachment set

anonymous — Wed, 18 Feb 2009 20:31:40 GMT

attachment → virtual_inheritance.cpp

attachment set

anonymous — Wed, 18 Feb 2009 20:32:00 GMT

attachment → virtual_inheritance_to_be_exported.hpp

roman.yakovenko@… — Wed, 18 Feb 2009 20:32:34 GMT

The bug description:

The user has two classes:

struct base{

virtual void do_smth() {}

};

struct derived : virtual public base {};

and exposes them using the following code:

namespace bp = boost::python;

struct base_wrapper : base, bp::wrapper< base > {

<<< constructors >>> virtual void do_smth() {...} void default_do_smth( ) {...}

};

struct derived_wrapper : derived, bp::wrapper< derived > {

<<< constructors >>>

virtual void do_smth() {...} void default_do_smth() {...}

};

BOOST_PYTHON_MODULE(virtual_inheritance){

bp::class_< base_wrapper >( "base" )

.def(

"do_smth" , (void ( ::base::* )( ) )(&::base::do_smth) , (void ( base_wrapper::* )( ) )(&base_wrapper::default_do_smth) );

bp::class_< derived_wrapper, bp::bases< base > >( "derived" )

.def(

"do_smth" , (void ( ::base::* )( ) )(&::base::do_smth) , (void ( derived_wrapper::* )( )

)(&derived_wrapper::default_do_smth) ); compiler complains to this line }

GCC 4.1.3 and MSVC 7.1 complain on the last line. If the user removes "virtual" from inheritance everything works fine.

In the past, a bug similar to this one was already fixed: http://mail.python.org/pipermail/cplusplus-sig/2003-September/005197.html

I also attach source code and generated code. It is pretty close to "minimum".

summary changed

troy d. straszheim — Fri, 25 Sep 2009 04:46:45 GMT

summary Failed to wrap classes with virtual inheritance → [fix in git] Failed to wrap classes with virtual inheritance

Fix here:

http://gitorious.org/~straszheim/boost/straszheim/commit/ce46ee7984344e32983330d5bc4b910002d1cdcc

Instead of checking the virtual function by constructing the function we're passed (which isn't possible), we just check if it will be possible out in type_traits and function_types land.

This fix will make its way back up to svn trunk eventually.