Update os.path module to use file descriptor where possible

[mkurek]

Fix #1653
Use _FdOrPathType when possible in os.path module.

[JelleZijlstra]

(The Travis failure is due to a mypy change. I'll submit a typeshed PR to fix it, but you may end up having to rebase this PR.)

[mkurek]

@JelleZijlstra sure, thanks!

[JelleZijlstra]

Fixed now (just needed to restart the build).

[JelleZijlstra]

Is it documented that these accept fds? They're not in os.supports_fd and I don't see anything in https://docs.python.org/3/library/os.path.html#os.scandir.

Looking at the code for os.path.exists (https://github.com/python/cpython/blob/master/Lib/genericpath.py#L16), it seems like it accidentally supports fds because it calls os.stat directly.

[mkurek]

@JelleZijlstra I wasn't aware of os.supports_fd. So the question is if support for fds for os.path functions is accidentally (and that's why it's undocumented and it could change in the future) or it should be documented. On the other hand https://docs.python.org/3/library/os.path.html#os.path.exists and https://docs.python.org/3/library/os.path.html#os.path.samefile mention using os.stat which is allowing fds.

So what do you suggest? If you like, I could prepare some pull requests to update cpython, either adding these functions to os.supports_fd or documenting that these functions support fds.

[gvanrossum]

My position is that if it isn't documented it's not intentional.

[JelleZijlstra]

I agree, so filing a ticket with CPython to clarify whether this behavior is intentional sounds like the right approach.

[mkurek]

Ok, I've created ticket to CPython to clarify this: https://bugs.python.org/issue31871

[JelleZijlstra]

Serhiy's response on bugs.python.org suggests that the fact that these functions support fd arguments in some platforms is unintentional. Therefore, I'm inclined to not support fd arguments in typeshed.

[gvanrossum]

Agreed.