Sorry, I'm lazy and this is a single patch to 2 files
to fix reference leaks.  

The int patch can be demonstrated by doing:

  abs(-2147483648)

The second happens when:

  import pyexpat
  p = pyexpat.ParserCreate()
  p.__members__

Logged In: YES 
user_id=6656

Looks fine to me.

I haven't tested it, but I assume it compiles and the tests
pass.  I can't see how you could write a test for this.

Logged In: YES 
user_id=80475

There is an error in the patch to intobject.c.  When 
o==NULL, then result is returned without having been set.  
A revised patch is attached.  Otherwise, it's okay:  I verified 
the leak fix and ran regression tests without exception.

The second patch checks out okay.  It might be cleaner 
and faster to alter the PyList_New for the correct number 
of entries and then just use PyList_SET_ITEM.  That saves 
the double INCDEF followed by a DECREF and it shortens 
the code a bit.

Logged In: YES 
user_id=80475

When the patches are fixed-up, they should also be 
backported.

Logged In: YES 
user_id=33168

Raymond, you are correct about the problem in intobject.  I
agree about that it could be cleaner to use PyList_New, but
I'm not making the change now.  Just trying to make things
correct (and I'm too lazy :-).

Michael, these could be tested with pydebug builds, if
sys.getrefcount() was captured before and verified it was
correct after calling the code to test/exercise the
problems.  I'm not sure it's worth it, though.

Checked in as:
  intobject.c 2.99 and 2.79.6.6
  pyexpat.c  2.77 and 2.57.6.5