Add `follow_symlinks` to `os.path.exists()`

[nineteendo]

Feature or enhancement

Proposal:

Because we have os.path.exists() & os.path.lexists(), it might make sense to add a follow_symlinks argument to the former:

-def exists(path):
+def exists(path, *, follow_symlinks=True):
     """Test whether a path exists.  Returns False for broken symbolic links"""
     try:
-        os.stat(path)
+        os.stat(path, follow_symlinks=follow_symlinks)
     except (OSError, ValueError):
         return False
     return True

This matches pathlib.Path.exists() (as well as os.stat() & os.chmod()). Also, it looks like pathlib.Path doesn't have an lexists() function. (but that's less explicit and so unnecessary, see #21157 (comment))

Note: this argument also needs to be added to nt._path_exists(). Done.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• gh-117705: Add follow_symlinks to os.path.exists() #117729

[erlend-aasland]

[...] Also, it looks like pathlib.Path doesn't have an lexists() function.

See previous discussion:

• Add Path.lexist() to pathlib #78318

[nineteendo]

OK, thanks for the information. I won't be pushing for that then.

[erlend-aasland]

OK, thanks for the information. I won't be pushing for that then.

It is useful to search the tracker, Discourse, and the old python-ideas/python-dev mailing list archives before submitting feature requests.

BTW, I would encourage you to take a good look at the devguide; it is a good read. Suggesting to start with the lifecycle of a pull request, and especially the subsection about making good PRs.

[nineteendo]

It is useful to search the tracker, Discourse, and the old python-ideas/python-dev mailing list archives before submitting feature requests.

It's indeed useful, it looks like this functionality was rejected for os.path.isdir(): #57180, but this was later added to pathlib without mention of it: #105793. I guess we could then also reopen the discussing about adding follow_symlinks to os.path.isdir() & os.path.isfile().

[erlend-aasland]

I guess we could then also reopen the discussing about adding follow_symlinks to os.path.isdir() & os.path.isfile().

The place for such a discussion would be Discourse. Iff there is consensus, follow up with an issue on the tracker.

[nineteendo]

BTW, I would encourage you to take a good look at the devguide; it is a good read. Suggesting to start with the lifecycle of a pull request, and especially the subsection about making good PRs.

I have printed the latter, but I have 2 questions:

1. How can I run tests on Windows? I don't have the make command.
2. How am I supposed to configure my linter in VS Code? Autopep8 & isort are way too strict and I had to turn them off. It seems like only a trailing newline is required.

[erlend-aasland]

1. How can I run tests on Windows? I don't have the make command.

You'll find instructions both for building and running tests in the devguide.

2. How am I supposed to configure my linter in VS Code? Autopep8 & isort are way too strict and I had to turn them off. It seems like only a trailing newline is required.

I have no idea; I don't use VS Code :) I think you are better off asking that question in a VS Code forum.

[eryksun]

On Windows, the os.path functions exists(), isfile(), isdir(), and islink() are implemented in "Modules/posixmodule.c" as the corresponding os__path_*_impl() builtin functions. We never got around to implementing isjunction() as a builtin. I don't recall why lexists() wasn't implemented as a builtin.

In the context of stat queries, a name-surrogate reparse point is considered a 'symlink'. The GetFileInformationByName() fast path will have to be modified to use the result when not following symlinks if it's a name-surrogate reparse point. The slow path will have to call STAT() when not following symlinks, or win32_xstat_slow_impl() could be refactored into reusable helper functions that encapsulate the design decisions and logic.

[nineteendo]

Is this something you would be willing to make a pull request for?

[nineteendo]

Alternatively, we could make ntpath.exists() a wrapper for nt._path_exists() & genericpath.lexists(). Then we wouldn't need to change the C code.

[eryksun]

That works, if you're averse to programming in C, assuming the wrapper doesn't significantly hurt the performance of exists(). Otherwise, I'd rather have a fast path for lexists(). It's currently a slower path based on a full os.lstat(). If _path_exists() supports follow_symlinks, we can trivially implement builtin _path_lexists() via os__path_exists_impl().

[nineteendo]

There's around 1% hit on performance. We have better things to do than optimise that using C code.
Let's leave implementing _path_lexists() for another time.