The documentation for the stat module is inconsistent
(Doc/library/stat.rst, at least for Python 2.7.2 and 3.3.0):
It talks about a function stat.S_IFMT() and a bit mask stat.S_IFMT.
Only the former does exist.

Besides, it states: "For complete details about the stat(), fstat()
and lstat() calls, consult the documentation for your system."
I suggest to add some pointers on what systems one might consult
what documentation: "(e.g., on GNU/Linux invoke 'man 2 stat')"
I don't know about other systems, though.

So, doing "man 2 stat", which refers to the POSIX standard and which
seems to have served as blueprint for stat's interface, I expected
to find the bit mask S_IFMT.
However, that does not exist.  In contrast, in stat.py that bit mask
is hard-coded as 0o170000 in the definition of S_IFMT().

As long-term, backwards-incompatible fix, I suggest to export
S_IFMT = 0o170000
and to rename the function S_IFMT().  That way, stat would in fact
be aligned with the documentation for my system.

As short-term fix, I suggest to correct stat.rst.

Replace
> Use of the functions above is more portable than use of the first
> set of flags:
>
> stat.S_IFMT
>    Bit mask for the file type bit fields.

with
> Use of the functions above may be more portable than use of the
> first set of flags.
>
> Warning: Note that the stat module does not export a bit mask
> S_IFMT. (As stated incorrectly in previous versions of the
> documentation.)

Here, I replaced "is more portable" with "may be more portable" as
the constant 0o170000 is hard-coded into S_IFMT() in stat.py.  Maybe
somebody could add a hint in what sense portability is improved?
Alternatively that sentence could be deleted.

A diff for stat.rst (Python 3.3.0) is attached.