The docs claim that os.DirEntry exists (https://docs.python.org/3/library/os.html#os.DirEntry) but it actually does not (it is internal to the posix module). Should probably actually make os.DirEntry exist (but continue to not document the constructor).

New changeset b841972ed0bd by Brett Cannon in branch 'default':
Issue bpo-27038: Expose DirEntry as os.DirEntry.
https://hg.python.org/cpython/rev/b841972ed0bd

Thanks for the patch, Jelle!

It would be nice if there was a supported way to create a DirEntry object from a path. If you don't want to document the constructor perhaps expose some helper function or class method (i.e. from_path)?

Exposing a class method to construct one is no different than exposing the constructor, so it's still not desired to have. What is the specific need you have for creating a DirEntry instance on its own?

It makes it much easier to write functions that can work with either a DirEntry object or a plain path.

Brendan Moloney added the comment:

It would be nice if there was a supported way to create a DirEntry object
from a path. If you don't want to document the constructor perhaps expose
some helper function or class method (i.e. from_path)?

It's a deliberate choice to not expose the constructor. You should not
built such objzct yourself. It requires low-level data coming from
readdir() or FindFirstFile(). If you don't pass such data, you loose the
whole purpose of the optimization.

Maybe the DirEntry doc must be more explicit on that point?

Please use pathlib instead. pathlib objects don't cache os.stat() result,
but that's also deliberate! It was proposed to add an optional cache, but
the idea was not accepted.

New changeset a95d98086621 by Ned Deily in branch 'default':
Issue bpo-27736: Prevent segfault after interpreter re-initialization due
https://hg.python.org/cpython/rev/a95d98086621

The pathlib Path class is different enough from the DirEntry class that it doesn't really help my goal of allowing a function to work with either a DirEntry or a plain (str) path.

These functions are going to be working with DirEntry objects generated by scandir 99% of the time so that we can leverage the improved performance, but I want to allow users to pass in plain paths as well. If the stdlib doesn't allow me to create my own DirEntry object from a plain path I will probably end up creating some sort of FakeDirEntry class that mimics a DirEntry, and that seems a bit silly and not a great use of time and effort.

If the constructor currently "requires low-level data coming from
readdir() or FindFirstFile()" then why not a class method that just takes a path and calls stat on it? You could document the fact that it calls stat and that it doesn't have any of the performance benefits of using scandir. Consenting adults and all that...

If all you want is to extract the path representation from an os.DirEntry instance and continue to work with strings/bytes then you can use os.fspath() which is new in Python 3.6 to get the representation: https://docs.python.org/3.6/library/os.html#os.fspath .

The functions were all written around DirEntry objects since that is what they will be processing 99% of the time and I want to do that as efficiently as possible. Converting a DirEntry object into a str representation doesn't help me with my use case at all. I want to be able to do it the other way around: str -> DirEntry

Brendan Moloney: Please open a thread on the python-ideas mailing list to discuss your idea. Please don't start discussing on a *closed* issue: since an issue is closed, it's hidden from the main bug tracker and so only very few people see it.