os.path.abspath regression #464

Activity

1. jeff5 commented on Oct 27, 2026

   

   jeff5

   on Oct 27, 2026

   Member

   More actions

   Sounds like we broke something there in how paths are being joined.

   Although the link is interesting, I don't think the problem stems from per-drive defaults, but is simpler. I recall struggling with (and giving up on) making os.chdir work properly in that way on Windows. This is interesting on where per-drive CWDs live according to your experience at the DOS prompt: https://stackoverflow.com/questions/53078528/how-to-get-current-working-directory-of-another-drive-in-batch
2. jeff5 commented on Oct 27, 2026

   

   jeff5

   on Oct 27, 2026

   Member

   More actions

   The root of this is that Jython does not provide nt._getfullpathname. Therefore when we import ntpath we take the "not running on Windows" definition of abspath. I think CPython 2.7 would behave the same way on Unix, but with a bit of effort one can explore the effect on Windows.

   Jython-2>python -S
   Python 2.7.18 (v2.7.18:8d21aa21f2, Apr 20 2020, 13:25:05) [MSC v.1500 64 bit (AMD64)] on win32
   >>> import nt
   >>> del nt._getfullpathname
   >>> import os.path
   >>> os.path.abspath('d:..')
   'd:..'
   

   In the absence of nt._getfullpathname, Python relies on ntpath.join(os.getcwd(), path), which cannot determine the CWD belonging to a drive and behaves like so:

   >>> os.path.join('c:\\a\\b\\c', 'd:..')
   'd:..'
   >>> os.path.join('d:\\x\\y', 'd:..')
   'd:\\x\\y\\..'
   

   in the first case treating d:... as absolute winner, and ignoring the CWD-per-drive issue, but in the second case spotting that the drive letters match.

   The version of ntpath we are using has older code than this that doesn't special case having the same drive. Changed in: python/cpython@31f5121