Description

lubomir

opened on Aug 22, 2023

Issue body actions

Steps to reproduce:

$ mkdir foo
$ cd foo
$ git init
$ touch 'bokm'$'\345''l.alias'
$ python -c 'import git; git.Repo(".").untracked_files'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/usr/lib/python3.11/site-packages/git/repo/base.py", line 839, in untracked_files
    return self._get_untracked_files()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/site-packages/git/repo/base.py", line 856, in _get_untracked_files
    filename = filename.encode("ascii").decode("unicode_escape").encode("latin1").decode(defenc)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xe5 in position 4: invalid continuation byte

I can reproduce with 3.1.30 and 3.1.32.

A file with such name is present in ftp://ftp.gnu.org/gnu/aspell/dict/nb/aspell-nb-0.50.1-0.tar.bz2
I'm not sure what encoding it's supposed to be. Firefox detects it as windows-1252 (and displays it correctly).

Activity

1. Byron

   added

   acknowledged

   help wanted

   on Aug 22, 2023
2. Byron commented on Aug 22, 2023

   

   Byron

   on Aug 22, 2023

   Member

   More actions

   The handling of filepaths is completely broken in GitPython as it tries to handle it like it's strings even though these imply a known encoding.

   This could definitely use an overhaul, these issues are in many places.
3. pushfoo commented on Oct 16, 2023

   

   pushfoo

   on Oct 16, 2023

   More actions

   The handling of filepaths is completely broken

   1. How much of this is Python 2 legacy?
   2. Are there more detailed notes on how / why?
   3. Is there an external dependency policy of any sort restricting what can be used beyond license compatibility?
4. Byron commented on Oct 16, 2023

   

   Byron

   on Oct 16, 2023

   Member

   More actions

   Probably most of it is the way it is due to a lack of understanding of how to handle paths in general and in git specifically. Most problems occour merely by trying to decode bytes into something that can be used in python even though one is dealing with either OS paths or bundles of bytes produced by the git binary itself.

   I don't think there is a limit of what could be used in terms of external dependency if it has a compatible license and is truly required. I'd hope python can handle OS paths properly out of the box by now.
5. Byron

   added the

   Bug

   issue type on Jul 20, 2026
6. Byron

   closed this as not plannedon Jul 20, 2026

Issue actions