[Cython] PATH: force threads initialization for Py2.3, please review

[Stefan Behnel]

Hi again,

Lisandro Dalcin wrote:
> The problem: If at the point PyGILState_Release() is called, the GIL
> whas not yet created, then the Python (2.3) interpreter segfautls.
> 
> The solution: Call PyEval_InitThread() in the module init function.
> But this could have performance impacts for non threaded applications
> using a Cython-generated extension module, so this thread
> initialization should be done only if needed. A simple approach is to
> call PyEval_InitThread() only if the Cython module ever
> released/aquired the GIL
> 
> The attached patch implement all the ideas discussed above. Any chance
> to this being accepted? Any better idea? Note that PyEval_InitThread()
> is a no-op if the GIL is already created.

I'm fine with calling InitThread() when "with nogil" or "with gil" is used
anywhere in the source. And I think it should be left to the users to handle
all other cases themselves.

Also note that UtilityCode() accepts a keyword "init", so you can just put the
call to InitThreads() in there and it will do the right thing. I think the
"proto" can be dropped completely in this case.

Stefan