[Martin]
> The patch that Meador committed is incorrect: METH_NOARGS functions
> still take a PyObject* args argument, which will be NULL.

Hmm.  I see this usage in a lot of places---e.g. see unicode_capitalize, unicode_casefold, unicode_title etc. in Objects/unicodeobject.c.  So it looks like we're relying on the (PyCFunction) cast to convert from a one-argument function pointer to a two-argument function pointer, which sounds a bit worrisome---I guess it just happens to work with common ABI calling conventions. I'm a bit surprised that we're not seeing compiler warnings about this sort of thing.

[Meador]
> I will happily fix it, but if it is wrong one place, then it is wrong
> everywhere.

It sounds like 'wrong everywhere' is accurate, unfortunately.