[PyQt] Exception handling crash

[Phil Thompson]

On Friday 13 June 2008 10:54:47 am Arve Knudsen wrote:
> Hi
>
> I'm still having problems with SIP's exception handling support, using
> SIP snapshot 20080612). When an exception from my wrapped C++ code is
> caught, SIP triggers an error in the Python API (PyEval_SaveThread:
> NULL tstate), The relevant backtrace looks as follows:
>
> #3  0xb6854968 in Py_FatalError () from /usr/lib/libpython2.5.so.1.0
> #4  0xb682cad5 in PyEval_SaveThread () from /usr/lib/libpython2.5.so.1.0
> #5  0xb2627b69 in release_QString (sipCppV=0x825e3c0) at
> sipQtCorepart1.cpp:16116
> #6  0xb279c2db in release (addr=0x825e3c0, td=0xb27813a0, state=1) at
> siplib.c:5651
> #7  0xb279c299 in sip_api_release_instance (cpp=0x825e3c0,
> type=0x8495354, state=1) at siplib.c:5633
> #8  0xb2503186 in meth_BuildProject2_loadProject (sipArgs=0x854f9cc)
> at sip_wrapperBuildProject2.cpp:669
>
> AFAICT the problem is that outside the try block there is a call to
> Py_BEGIN_ALLOW_THREADS, which releases the GIL? Then, first thing
> after the exception is caught, an QString is released (a parameter to
> the wrapped method). As part of releasing the QString, in
> release_QString, Py_BEGIN_ALLOW_THREADS is called again, which is not
> allowed since the GIL is not being held, right?

Can you send me the generated code so that I can see the logic.

> >From what I understand there is a simpler alternative to
>
> Py_BEGIN_ALLOW_THREADS/Py_END_ALLOW_THREADS in the API:
> PyGILState_Ensure/PyGILState_Release. Could SIP make use of these
> functions instead?

They are not alternatives. One pair frees the GIL for a period, the other 
acquires the GIL for a period.

Phil