We can remove this:

Right ! Removed the reminder for the whole codebase in 7ded8d8

yikes, we had skip instead of skipif and the tests where broken

965e942 introduced the issue replacing a pylib basestring with just str

wrong ref used

Thank you for looking into it, I was stuck. I'm not sure I understand though, should I replace ["*dir", "*dir", b"*dir"], back to ["*dir", "*dir", pytest.mark.skipif("sys.version_info < (3,6)")(b"*dir")], and then fix the tests ? Maybe in another merge request at this point ?

in the Visitor class of the local path clone, there is a condition we need to replace
https://github.com/pytest-dev/pytest/blob/main/src/_pytest/_py/path.py#L140 - it should also check for bytes

when asottile took out a lot of cruft a basestring usage was misinterpreted and replaced with just str

unfortunately the skips where wrong and thus the error was never noted as the tests skipped instead of passing

I'm not sure that's where things went wrong.

I checked the py code, and in Python 3.0+ py.basestring was indeed set equal to str. The problem is that using str and bytes interchangeably isn't valid in Python 3, and py.path.local never fully accounted for that.

Even in the upstream, this happens in Python 3.6 (or higher, I'm sure):

>>> import py
>>> dir = py.path.local(b'/tmp')
>>> dir
local(b'/tmp')
>>> dir.strpath
b'/tmp'
>>> dir.parts()
[local(b'/'), local(b'/tmp')]
>>> dir.basename
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File ".../venv3.6/lib64/python3.6/site-packages/py/_path/common.py", line 141, in basename
    return self._getbyspec('basename')[0]
  File ".../venv3.6/lib64/python3.6/site-packages/py/_path/local.py", line 293, in _getbyspec
    parts = self.strpath.split(self.sep)
TypeError: a bytes-like object is required, not 'str'
>>> 
>>> # Part of the problem being...
>>> dir.sep
'/'
>>> dir.bestrelpath(b'/')
"b'/'"
>>> # ^^^ Note that that's a _string_ containing "b'/'",
>>> # IOW the result of calling str() on a bytes object

...That the tests were being unconditionally skipped, and never caught that, is the bigger problem. But fixing bytes support will not be easy as _pytest/_py/path.py (like py.path.local before it) is very sloppy with how it handles inputs. When passed bytes, things will sorta work... until they very suddenly really don't.

Oh, and just for fun, I found two more instances of tests being unconditionally skipped by pytest.mark.skip(), instead of conditionally with .skipif(). Just in that one test file.

thanks for the report; I've opened a new issue to warn when that kind of thing happens 🙏