Unassigning for the moment - I'll take a look at adding the Py3k warning
back in at some point, but the main thing I would like now is a sanity
check from Guido or Barry (or anyone else happy to dig into the guts of
typeobject.c) that the basic idea behind my fix is sound.

It would be nice to avoid the explicit check for Py_None in
PyObject_Compare by getting update_one_slot() to replace the Py_None it
encounters in __hash__ with the default slot_tp_hash implementation, but
my attempts at doing that have all led to segfaults.

(Also, glyph, I'd love to know if my patch helps to clear up Django's
issues with 2.6b1)