Should FilePathorBuffer use os.PathLike instead of pathlib.Path?

[willfrey]

The documentation for pandas.read_csv(...) states that it can support anything that adheres to the os.PathLike protocol, but the definition of FilePathOrBuffer is too strict.

pandas/pandas/_typing.py

Line 138 in 08e4baf

FilePathOrBuffer = Union[Path, FileOrBuffer[T]]

where Path is a pathlib.Path object.

Could this be changed to use os.PathLike[str] or just os.PathLike?

from os import PathLike

FilePathOrBuffer = Union["PathLike[str]", FileOrBuffer]

PathLike is in quotes because PathLike only supports __class_getitem__ for Python 3.9 and later and PathLike alone might be too broad, assuming you don't want to support byte-encoded paths.

Thanks!

[twoertwein]

Replacing pathlib.Path with os.PathLike[str] should definitely work! Most (all?) IO calls go through get_handle which internally calls stringify_path which converts all os.PathLike[str] using __fspath__ to a string. Unless you add a decode function to stringify_path, I wouldn't add os.PathLike[bytes].

I think we already have tests for non-pathlib.Path PathLike objects.

[Abhi-H]

From what I gathered through @twoertwein comment, the PathLike[str] change will bring the definition of FilePathorBuffer closer to being in-line with the documentation for pandas.read_csv(...). Should I go ahead and make that change?

[twoertwein]

yes, feel free to make the change.

[Abhi-H]

Okay, will make a PR soon. Setting up my dev environment currently.