I created in the python module gsl_fft a __del__ method to release 
workspace when the data size changes.  It is called upon exiting 
python, and gave errors without traceback.  So I registered it with 
atexit.  This caused core dump.  The attached tape archive has 
necessary modules, and file gdb.output.  The latter contains, I hope, 
enough information to reconstruct the command line and even to find 
the cause.  Additionally you'll need the gsl and gslcblas shared 
object libraries.

I installed gsl using "sudo apt-get install libgsl0ldbl". Then I 
added ".0" to the library filenames (eg. CDLL('libgsl.so.0')).

$ python3.0 dumpdriver.py
Exception AttributeError: "'NoneType' object has no attribute 'debug'" 
in <bound method Real_FFT.__del__ of <gsl_fft.Real_FFT object at 
0xb7bb45ac>> ignored

Real_FFT.__del__() uses logging.debug, logging.info and logging.warn 
whereas logging is None when the destructor is called.

About your gdb output:
 - You have a 64 bits CPU (I have a 32 bits CPU)
 - You are using Red Hat 6.3.0.0-1.132.EL4rh (I'm using Ubuntu Hardy)
 - The crash occurs in PyErr_Fetch() on "*p_type = 
tstate->curexc_type;": I guess that p_type and/or tstate are NULL. The 
most probable cause is that PyThreadState_GET() returns a NULL pointer.

I can reproduce the problem, and it seems that it is already corrected
by r66562 (issue3666), one week after 3.0rc1.

Can you try with a newer version?

I rebuilt python3k from "today's snapshot".  No more core dump.  Problem
solved, close case. IMH'edO.