Issue 34384: os.readlink does not accept pathlib.Path on Windows

Documentation for os.readlink says "Changed in version 3.6: Accepts a path-like object.". This works fine on macOS, but blows up on Windows:

Python 3.7.0 (v3.7.0:1bf9cc5093, Jun 27 2018, 04:06:47) [MSC v.1914 32 bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import os
>>> import pathlib
>>> os.readlink(pathlib.Path('foo'))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: readlink() argument 1 must be str, not WindowsPath

Steve and/or Eryk, I was adding some tests for os.readlink() in PR 8740 and I noticed that os.readlink() always returns str on Windows. However, the documentation for os.readlink() says:

    If the path is a bytes object (direct or indirectly), the result will
    be a bytes object.

The question is, should I add support for bytes as part of PR 8740? And what is the best way to implement it? (e.g. use CreateFileA() if PyUnicode_Check(path.object) returns false?)

In msg235615 (Issue 9949), Zachary said using bytes paths on Windows is deprecated, but I can't see the actual conversation because Rietveld seems to be down: https://bugs.python.org/review/9949/#ps5077 I think the os.readlink() documentation needs to be updated if we decide to keep the status quo.

Currently the behavior doesn't match the documentation. Initially I thought that this can be solved by adding the support of path-like objects and backporting this up to 3.6. But os.readlink() on Windows doesn't not support also bytes paths, and never supported. This looks to me now more like a new feature. In 3.6 and 3.7 we can resolve this issue by just updating the documentation.

Bytes paths on Windows were deprecated before 3.6. Since implementing PEP 529, their should be supported on Windows if they are supported on other platforms. We shouldn't use an 8-bit API like CreateFileA, but instead decode bytes paths with UTF-8 and use a Unicode API (this should be implemented in path_converter()).

Serhiy is exactly right, but to emphasise, the best way to do paths now is to use argument clinic with its path_t type. On Windows, .narrow is a flag that indicates whether you should PyUnicode_FSEncode() any results before returning them, and .wide is always initialized correctly.

Do not use any *A APIs - only *W from now on :)

Thanks for the suggestions! I've updated PR 8740. I will submit separate PRs to fix os.readlink() documentation in 3.6 and 3.7 branches.

New changeset e0b5b2096ead4cc094a129ce7602ac5c0e998086 by Berker Peksag in branch 'master':
bpo-34384: Fix os.readlink() on Windows (GH-8740)
https://github.com/python/cpython/commit/e0b5b2096ead4cc094a129ce7602ac5c0e998086

can this be closed?

Yep

stage: patch review -> resolved