[C++-sig] A very strange behaviour of Boost (Mingw, gcc 4.5)

[Jérôme Laheurte]

Hello. I already asked this on StackOverflow but it doesn't seem to 
inspire many people. I managed to reduce my problem to a trivial 
extension module:

#include <Python.h>
#include <boost/python.hpp>

using namespace boost::python ;
using namespace boost;

class PyTest
{
public:
     PyTest();
};

PyTest::PyTest()
{
     PyErr_SetString(PyExc_RuntimeError, "FOO");
     throw_error_already_set();
}

BOOST_PYTHON_MODULE(testmod)
{
     class_<PyTest>("Test", init<>())
         ;
}

On Windows XP SP3, if I build Boost 1.46.1 and then this extension with 
GCC 3.4.5 from an older version of Mingw, everything behaves as expected:

 >>> import testmod
 >>> testmod.Test()
Traceback (most recent call last):
   File "<stdin>", line 1, in ?
RuntimeError: FOO

Now if I rebuild both Boost and the module using a newer Mingw, with GCC 
4.5.2, this happens:

 >>> import testmod
 >>> testmod.Test()

This application has requested the Runtime to terminate it in an unusual 
way.
Please contact the application's support team for more information.

Putting a break on abort in GDB and printing the stack gives this:

(gdb) bt
#0  0x77c36bb3 in msvcrt!abort () from C:\WINDOWS\system32\msvcrt.dll
#1  0x69acfdb2 in testmod!_Unwind_SetGR () from C:\Python22\DLLs\testmod.pyd
#2  0x69b2a45d in 
testmod!_ZNK5boost6python7objects23caller_py_function_implINS0
_6detail6callerIPFvP7_objectENS0_21default_call_policiesENS_3mpl7vector2IvS6_EEE
EE9signatureEv () from C:\Python22\DLLs\testmod.pyd
#3  0x6e9544a3 in libgcc_s_dw2-1!__trunctfxf2 ()
    from C:\WINDOWS\system32\libgcc_s_dw2-1.dll
#4  0x6e954877 in libgcc_s_dw2-1!_Unwind_RaiseException ()
    from C:\WINDOWS\system32\libgcc_s_dw2-1.dll
#5  0x6fcbc6e2 in libstdc++-6!__cxa_throw ()
    from C:\WINDOWS\system32\libstdc++-6.dll
#6  0x63455d4c in boost::python::throw_error_already_set() ()
    from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#7  0x69ac12de in testmod!_ZN6PyTestC2Ev () from 
C:\Python22\DLLs\testmod.pyd
#8  0x69adbbf1 in 
testmod!_ZN5boost6python7objects11make_holderILi0EE5applyINS1_
12value_holderI6PyTestEENS_3mpl7vector0IN4mpl_2naEEEE7executeEP7_object ()
    from C:\Python22\DLLs\testmod.pyd
#9  0x69adbde9 in 
testmod!_ZN5boost6python7objects23caller_py_function_implINS0_
6detail6callerIPFvP7_objectENS0_21default_call_policiesENS_3mpl7vector2IvS6_EEEE
EclES6_S6_ () from C:\Python22\DLLs\testmod.pyd
#10 0x63450031 in boost::python::objects::function::call(_object*, 
_object*) con
st () from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#11 0x634501fb in 
boost::detail::function::void_function_ref_invoker0<boost::pyt
hon::objects::(anonymous namespace)::bind_return, 
void>::invoke(boost::detail::f
unction::function_buffer&) ()
    from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#12 0x63455e5a in 
boost::python::handle_exception_impl(boost::function0<void>)
     () from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#13 0x634512be in function_call ()
    from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#14 0x1e006ccc in python22!PyObject_Call ()
    from C:\WINDOWS\system32\python22.dll
#15 0x0089d138 in ?? ()
#16 0x63469420 in boost::python::docstring_options::show_py_signatures_ ()
    from C:\WINDOWS\system32\libboost_python-mgw45-mt-1_46_1.dll
#17 0x0084d0f0 in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

It looks like the error_already_set exception is somewhat not catched as 
it should be. In my actual use case, the class has several methods. This 
only happens with some of them; I couldn't find any difference between 
the methods for which the mechanism works and the other ones (same 
number and type of parameters, same exception type)...

Any hint would be greatly appreciated.