fix(types): get the os.PathLike type as correctly as possible

[muggenhor]

This should make our internal PathLike type compatible with Python < 3.6 and < 3.9.

Fixes#1219

[muggenhor]

Switched away from try-except to explicit version checks because that seems to be the recommended style python/mypy#698

[Byron]

Thanks, very elegant!

I find it interesting that all PathLike types seem to use str in some way even though AFAIK these are to be decoded with some encoding from what the OS provides. As this encoding is unspecified or at least very dependent on knowledge about the OS and its settings, using a path based on str can fail with some sort of decoding error.

[muggenhor]

FYI: this is the type annotation definition of os.PathLike. So what os.PathLike[str] means is: it has to be a type with an __fscache__(self) -> str method. So that's the method where any conversions should be performed.

[muggenhor]

I find it interesting that all PathLike types seem to use str in some way even though AFAIK these are to be decoded with some encoding from what the OS provides. As this encoding is unspecified or at least very dependent on knowledge about the OS and its settings, using a path based on str can fail with some sort of decoding error.

Hmm. Does that mean you think it should accept bytes in some locations too?

[Byron]

Hmm. Does that mean you think it should accept bytes in some locations too?

I would refrain from doing anything that reduces convenience or being idiomatic or that is breaking just to gain correctness.
But I think that as it stands, GitPython will work fine until it doesn't because one tries to handle a path with characters that don't decode with the pre-configured default encoding. Right now this is UTF-8, certainly a decent choice. But it won't always work either.

To me the whole encoding topic in GitPython is a lost cause as correctness could (to my mind) only be achieved if the API doesn't use strings (with encoding) to handle them.

And by the looks of it, it's idiomatic to do that, too:

# py 3.8>>> type(os.listdir('.')[0])
<class'str'>

Edit: On the bright side, I believe just today I had an epiphany which helps me to see a future for GitPython, leveraging it as a strength and not seeing it as a burden. So I hope this will change my mindset too when helping to maintain it in the midterm until such a future can be realized.