`Path("").exists()` always returns True, which is a bug

[yingshaoxo]

Bug report

Bug description:

from pathlib import Path
print(Path("").exists())

import os
print(os.path.exists(""))

For empty string "", pathlib returns True, but os returns False.

Clearly, os is right, because you can't read anything from "".

CPython versions tested on:

3.10

Operating systems tested on:

Linux

[terryjreedy]

Behavior verified in 3.13.0a2.

[barneygale]

This is longstanding behaviour in pathlib - Path(), Path('') and Path('.') all have an __fspath__() value of '.'

(Not meaning to imply this bug is invalid, just adding context)

[ronaldoussoren]

This is longstanding behaviour in pathlib - Path(), Path('') and Path('.') all have an __fspath__() value of '.'

(Not meaning to imply this bug is invalid, just adding context)

That's a side effect of pathlib.Path ignoring empty segments:

>>> import pathlib
>>> pathlib.Path("a", "", "", "b")
PosixPath('a/b')
>>> pathlib.Path("a////b")
PosixPath('a/b')

This is not documented, but is somewhat implied by "Spurious slashes and single dots are collapsed, " in the documentation. That said, it would IMHO be better to explicitly document this behaviour.

[barneygale]

pathlib models a path as three parts: (drive, root, tail), where the first two elements are strings, and tail is a list of strings. Each element in tail corresponds to a real path component (no "" or "."), and this is leveraged by relative_to() to generate the right number of .. segments.

Right now, a path with no segments is represented as ("", "", []). Its __str__() and __fspath__() would be "" according to normal joining rules, but pathlib special-cases it to "." instead.

I'm not sure how we'd represent both "" and "." in this model. Any suggestions? Also, what would Path("", "") and Path("", ".") produce?

[ronaldoussoren]

Is there any reason to change the current behaviour?

Treating the empty segment in Path("") differently from empty segments in multi-segment names complicates the semantics of the API with (IMHO) little to no advantage. The current behaviour is easy to explain: all empty segments are ignored, and Path() represents the current directory (e.g. Path(".").

[barneygale]

Closing as "won't fix" - this is deliberate and longstanding behaviour, and pathlib's model of paths doesn't leave room for a distinction between "" and ".". Thanks for the bug report with the clear repro steps all the same!