On Windows, the subprocess module makes use of functions publicly exposed by PC/_subprocess.c to interact with Win32 API functions. However, no documentation exists for these functions, neither in the online docs nor in docstrings.

Attached is a patch which adds docstrings to the functions in PC/_subprocess.c, documents the functions in Doc/library/subprocess.rst, and removes a chunk of unneeded import code from Lib/subprocess.py

This is likely because these functions are meant to be undocumented implementation details.

True. Is there value in eventually privatizing these functions?

It feels weird having them exposed but not documented at all...maybe just keep the docstrings around?

Actually, I consider them private now, given that they aren't in __all__ or documented. I'd prefer if the code was changed to qualify their use with _subprocess.X actually. Docstrings are fine, I suppose.

I agree.  Docstrings never hurt.

Here's a patch which keeps docstrings, and explicitly qualifies the Windows specific functions from _subprocess rather than using import * (which causes a couple of lines just over 79 chars). Now the functions are more hidden than before.

Ran the tests to be sure I got each change, nothing seems to have slipped by.

I know nothing about Windows, but the patch looks sensible.

Typos in docstrings: 'associatedwith' missing space;
'a duplicate a handle object' (remove first 'a')?; 
'return value is tuple' ('is a tuple').

Here's a patch which makes those changes. Thanks for taking a look and catching that stuff.

Fixed with r80439-r80442.

New changeset 80971f71b0d9 by Brian Curtin in branch '3.1':
Further fix #7838. CREATE_NEW_CONSOLE was exposed, but none of the
http://hg.python.org/cpython/rev/80971f71b0d9