Logged In: YES 
user_id=6656

FWIW, I can (a) reproduce this on alpha/OSF1 (recompiling
stringobject.c without optimizations makes no odds,
recompiling all of Python takes too long on this particular
alpha) (b) make it go away by replacing the call to
PyMem_MALLOC in mymemreplace with one to PyMem_Malloc.

The difference is that PyMem_Malloc(0) can't be NULL;
PyMem_MALLOC can.

So either do this, or check if new_len is zero and return
PyString_FromString("") or something (in mymemreplace).