Handle leading `//` for `posixpath.realpath`

[nineteendo]

Bug report

Bug description:

>>> import posixpath
>>> posixpath.realpath("//foo")
'/foo'

Expected: //foo with precisely two leading slashes, see: https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap04.html#tag_04_13

This can be quite easily solved using posixpath.splitroot, see the PR.

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-117338: Handle leading // for posixpath.realpath #117340

[barneygale]

os.path.realpath() resolves the path using the current OS's rules, which might be more specific than POSIX. Both my Linux and macOS machines resolve //foo to /foo. Are there any supported Python platforms where // isn't resolved to /?

[nineteendo]

I'm currently on my Windows laptop, so I can't check.
In any case, the old implementation can never return //foo because path is set to os.sep.
Is there a "needs investigation" label?

[barneygale]

There doesn't seem to be anything like a "needs investigation" label.

Happy to leave this issue open for a bit while you further research a repro case, but if we can't find one, then I think this issue should be closed as "can't repro".

[nineteendo]

Also, the current behaviour is undocumented:

Join two paths, normalizing and eliminating any symbolic links encountered in the second path.

Normalizing doesn't touch the double slash & // is not a symlink.

[barneygale]

You're quoting a comment above the internal _joinrealpath() function.

Both the posixpath.realpath() docstring, and the online docs, describe it as returning a canonical path. The online docs add:

This function emulates the operating system’s procedure for making a path canonical

[nineteendo]

OK, but that comment doesn't document this feature, which it probably should, in case this works correctly.

[bluss]

Why do you want to fix this, and could you describe in more detail how the changes relate to the specification you are pointing to? I don't see what in the specification requires this.

[nineteendo]

Yep, the old behaviour seems to be correct (cd doesn't resolve symlinks), in that case the feature should just be documented and tested.

• macOS:
  
• Pydroid 3:
  

[barneygale]

We don't need to document every detail of what realpath() does. Tests would be welcome, though

[nineteendo]

I've added just one sentence in this pull request: #117350.

Join two paths, normalizing and eliminating any symbolic links encountered in the second path. Two leading slashes are replaced by a single slash.

Where should I add the test though? It doesn't really fit in with the rest of the changes.