Fix for
http://mail.python.org/pipermail/python-dev/2006-April/063572.html

Convert all PyMem_* to PyObject_* that way we can't
have an inconsistent use of PyMem_ and PyObject_ APIs
in the Parser.  This could even be a perf win if more
of the allocations are smaller than 256 bytes.  I don't
know if that's more common though.

Logged In: YES 
user_id=6380

Works for me.  Thanks for the quick fix!

I suppose Tim should review this.

Logged In: YES 
user_id=31435

I noted some cautions on python-dev, but marked Accepted. 
In general I'm nervous about more-or-less mindless s/x/y/g
patches, but Guido's bug is proof that my time-consuming
selective approach may well not be the best approach to this
mess :-)

Logged In: YES 
user_id=33168

Heh, I'm going the opposite direction.  I'm going to
selectively move some of the known big calls back to
PyMem_*.  I also notice that there are a lot of allocs
without this patch are larger than 256 bytes.  I'm going to
research those too.  Thanks for the review.

Logged In: YES 
user_id=33168

The only inconsistency I could find was at line 818 of
tokenizer.c:

tok->buf = new;

new is PyMem_*, but tok->buf is PyObject_* everywhere except
this assignment.

I didn't see an easy fix other than reallocating and copying
instead of the assignment.

I undid a few of the changes in the patch.

Committed revision 45225.