os.path.isabs change in behavior in Python 3.13 on MS-Windows

[ikappaki]

Bug report

Bug description:

I'm not sure if this is a fix or a regression issue?

os.path.isabs use to return True when a windows path started with /.

In python 3.13

Python 3.13.0 (tags/v3.13.0:60403a5, Oct  7 2024, 09:38:07) [MSC v.1941 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import os.path
>>> os.path.isabs("/c:")
False

In python 3.12

Python 3.12.7 (tags/v3.12.7:0b05ead, Oct  1 2024, 03:06:41) [MSC v.1941 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import os.path
>>> os.path.isabs("/c:")
True

CPython versions tested on:

3.13

Operating systems tested on:

Windows

[nineteendo]

Works as intended: #113829, maybe a new function could be considered to check if a path is rooted?

[ikappaki]

Works as intended: #113829, maybe a new function could be considered to check if a path is rooted?

This behavior changes how os.path.isabs interacts with urllib.parse.urlparse when dealing with absolute paths in Windows URIs. In Python 3.12, paths parsed from local file URIs (drive and network paths) were recognized as absolute, but in Python 3.13, they no longer are

import os.path
import pathlib
import urllib.parse

>>> urllib.parse.urlparse(pathlib.Path('c:\\xyz').as_uri())
ParseResult(scheme='file', netloc='', path='/c:/xyz', params='', query='', fragment='')
>>> urllib.parse.urlparse(pathlib.Path('\\\\machine\\xyz').as_uri())
ParseResult(scheme='file', netloc='machine', path='/xyz/', params='', query='', fragment='')

In Python 3.12, os.path.isabs would correctly identify these paths as absolute, including network paths:

>>> os.path.isabs(urllib.parse.urlparse(pathlib.Path('c:\\xyz').as_uri()).path)
True
>>> os.path.isabs(urllib.parse.urlparse(pathlib.Path('\\\\machine\\xyz').as_uri()).path)
True

but not any more in 3.13

>>> os.path.isabs(urllib.parse.urlparse(pathlib.Path('c:\\xyz').as_uri()).path)
False
>>> os.path.isabs(urllib.parse.urlparse(pathlib.Path('\\\\machine\\xyz').as_uri()).path)
False

It seems as if os.path.isabs was never intended to be used with ParseResult paths. I suppose a more appropriate solution would be to directly check if the path starts with '/', which would indicate that it's absolute.

[nineteendo]

You can use pathlib.Path.from_uri() starting from 3.13:

>>> import pathlib
>>> pathlib.Path.from_uri(pathlib.Path('c:/xyz').as_uri())
WindowsPath('c:/xyz')
>>> pathlib.Path.from_uri(pathlib.Path('//server/drive/xyz').as_uri())
WindowsPath('//server/drive/xyz')

This doesn't roundtrip:

>>> import pathlib
>>> import urllib.parse
>>> pathlib.Path(urllib.parse.urlparse(pathlib.Path('c:/xyz').as_uri()).path)
WindowsPath('/c:/xyz')
>>> pathlib.Path(urllib.parse.urlparse(pathlib.Path('//server/drive/xyz').as_uri()).path)
WindowsPath('/drive/xyz')

Also, aren't uri's always absolute?

[marc-hb]

Also, aren't uri's always absolute?

Relative URLs work perfectly fine.

EDIT: I meant in general in life, not related to Python.

[nineteendo]

Well, at least pathlib doesn't allow creating them.

[barneygale]

It seems as if os.path.isabs was never intended to be used with ParseResult paths. I suppose a more appropriate solution would be to directly check if the path starts with '/', which would indicate that it's absolute.

Checking whether it starts with a / makes most sense to me. Another option is posixpath.isabs().

[nineteendo]

Use p.scheme and p.netloc and p.path.startswith("/") to check if an uri is absolute, an absolute path doesn't change when joined with another path:

>>> import urllib.parse
>>> urllib.parse.urljoin("file://netloc/foo/", "bar")
'file://netloc/foo/bar'
>>> urllib.parse.urljoin("file://netloc/foo/", "/bar")
'file://netloc/bar'
>>> urllib.parse.urljoin("file://netloc/foo/", "//netloc2/bar")
'file://netloc2/bar'
>>> urllib.parse.urljoin("file://netloc/foo/", "file://netloc2/bar")
'file://netloc2/bar'

That's also why we changed the behaviour of os.path.isabs():

>>> import os.path
>>> os.path.join(r"c:\foo", "bar")
'c:\\foo\\bar'
>>> os.path.join(r"c:\foo", r"\bar")
'c:\\bar'
>>> os.path.join(r"c:\foo", r"d:\bar")
'd:\\bar'

[nineteendo]

I think this can be closed.

[barneygale]

Thanks @nineteendo! Closing.

[kayychou]

Works as intended: #113829, maybe a new function could be considered to check if a path is rooted?

hi @nineteendo, @barneygale,

the doc updated in #113829 still indicates the following behavior:

Return ``True`` if *path* is an absolute pathname.  On Unix, that means it
begins with a slash, on Windows that it begins with two (back)slashes, or a
drive letter, colon, and (back)slash together.

whereas the intended behavior, as illustrated in Lib/test/test_ntpath.py, appears to be now :

>>> os.path.isabs("/foo/bar") # path beginning with a slash on windows
False
>>> os.path.isabs("\\foo\\bar") # path beginning with two backslashes on windows
False

maybe I'm missing something under the hood

Thanks for your help!

[barneygale]

Hey @kayychou - "\\foo\\bar" expresses a string that begins with a single backslash. If you want two backslashes, you could use a raw string (r"\\foo\bar") or double the backslashes ("\\\\foo\\bar")

For more info: https://docs.python.org/3/reference/lexical_analysis.html#escape-sequences