The idea of dynamic typing it seems quite heavy to me, but I'm not a
Python hacker, so I don't know what's the norm.

Notice that os.stat() does "PyInt_FromLong((long)st->st_uid)" on the
stat structure's st_uid field. On my platform (OS X 10.4), st_uid is
defined as a uid_t type.

So maybe os.stat() has the answer: ignore the signed vs. unsigned int
problem and just use a long. The actual chown() call in posix_chown()
casts the uid variable to a (uid_t) anyway. GCC doesn't seem to complain
when we cast a long to an unsigned int, even.