pathlib: inconsistent handling of rootless Windows paths

>>> from pathlib import PureWindowsPath
>>> a = 'C:/a/b'
>>> b = 'C:x/y'
>>> PureWindowsPath(a, b)
PureWindowsPath('C:x/y')
>>> PureWindowsPath(a) / b
PureWindowsPath('C:/a/b/x/y')

Seems to be the only case where the PureWindowsPath constuctor and joinpath() disagree when given the same arguments. This has implications for future work, so I'd love to make them consistent.

Affects all supported versions of Python.

Related to #64107, which improved the joinpath() behaviour.

Here's a patch that will be applicable when #31691 lands. It uses ntpath.join() to join the path parts, and consequently scans left-to-right without an early exit from the loop. Optimizations are possible, I just went for minimal diff:

diff --git a/Lib/pathlib.py b/Lib/pathlib.py
index bd86403c8e..ad3c03f1bd 100644
--- a/Lib/pathlib.py
+++ b/Lib/pathlib.py
@@ -266,10 +266,8 @@ def _parse_parts(cls, parts):
         sep = cls._flavour.sep
         altsep = cls._flavour.altsep
         drv = root = ''
-        it = reversed(parts)
-        for part in it:
-            if not part:
-                continue
+        if parts:
+            part = cls._flavour.join(*parts)
             if altsep:
                 part = part.replace(altsep, sep)
             drv, rel = cls._flavour.splitdrive(part)
@@ -289,20 +287,6 @@ def _parse_parts(cls, parts):
             else:
                 if rel and rel != '.':
                     parsed.append(sys.intern(rel))
-            if drv or root:
-                if not drv:
-                    # If no drive is present, try to find one in the previous
-                    # parts. This makes the result of parsing e.g.
-                    # ("C:", "/", "a") reasonably intuitive.
-                    for part in it:
-                        if not part:
-                            continue
-                        if altsep:
-                            part = part.replace(altsep, sep)
-                        drv = cls._flavour.splitdrive(part)[0]
-                        if drv:
-                            break
-                break
         if drv or root:
             parsed.append(drv + root)
         parsed.reverse()

I definitely agree that this is sufficiently complex that pathlib should use ntpath wherever possible.

pythongh-94909: fix joining of absolute and relative Windows paths in…

In fact, this would be good to fix before #31691 lands, as it makes backporting easier. PR available: #95450.

gh-94909: fix joining of absolute and relative Windows paths in pathl…

The fix only went into main to minimize risk of breaking and out of simplicity since this fixes an edge case only.

Navigation Menu

Repository navigation

pathlib: inconsistent handling of rootless Windows paths #94909

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

pathlib: inconsistent handling of rootless Windows paths #94909

Description

Activity

barneygale commented on Jul 16, 2022

barneygale commented on Jul 16, 2022

ericvsmith commented on Jul 19, 2022

barneygale commented on Jul 29, 2022

brettcannon commented on Aug 12, 2022

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions