GeezerVison

Thanks a lot for contributing this fix!

It appears that CI is legitimately failing though.

Could you investigate and see what changed between the last known working version, and the first broken one? Was it really this that changed, or maybe something else?

I am wondering if there could be a different fix for this, that would then also pass CI naturally.

Thanks a lot for contributing this fix!

It appears that CI is legitimately failing though.

Could you investigate and see what changed between the last known working version, and the first broken one? Was it really this that changed, or maybe something else?

I am wondering if there could be a different fix for this, that would then also pass CI naturally.

Hi,
Indeed, it seems that a security fix for windows broke the progress bar.
At first sight, it seems that universal_newlines arguments does more than it said, clearly the fix will be more complex than this first try.
i'll have a look, until CI is green, i convert this PR to draft.

By the way, python 3.7 is supported by EOL and not package on ubuntu latest (i.e. ubuntu 24.04 ==> https://github.com/gitpython-developers/GitPython/actions/runs/11334079345/job/31534626048?pr=1971).

That's interesting, as it sounds like a python security update on Windows broke GitPython? Or only universal newlines?
In any case, I am looking forward to you tackling it, thanks again for your help.

By the way, python 3.7 is supported by EOL and not package on ubuntu latest (i.e. ubuntu 24.04 ==> https://github.com/gitpython-developers/GitPython/actions/runs/11334079345/job/31534626048?pr=1971).

That's true - thanks for pointing it out. Maybe you can remove testing it on CI while at it.

I didn't bisect yet, but the regression is introduced between 3.1.40 and 3.1.41.
As said in the related issue, i'm using python 3.10 and ubuntu 22.04.1 and i tried different git version with the same result (from 2.30, 2.34.1 to last release with the result). Whatever the git version, the last working progress bar for fetch (and likely pull) is w/ GitPython 3.1.40

Concerning git/remote.py file, the diff only concern documentation fix (rewording, style, typo fixes). So, in the last working release use stderr as bytes and not text, convert in handle_process_output.
But there is major security fix for windows that, somehow, changes the behavior in handle_process_output.pump_stream where

That's true - thanks for pointing it out. Maybe you can remove testing it on CI while at it.

As python 3.7 is officially supported by the package, IMHO, it must be covered by unit testing.
Maybe use matrix configuration to select suitable ubuntu version (ubuntu-22.04 is a valid label, see https://github.com/actions/runner-images?tab=readme-ov-file#available-images).

I didn't bisect yet, but the regression is introduced between 3.1.40 and 3.1.41.

I am definitely curious what it turns out to be in the end. As it seems, the issue is definitely within GitPython, and I'd hope the security-related change can also be made so that it doesn't break anything.

As python 3.7 is officially supported by the package, IMHO, it must be covered by unit testing.
Maybe use matrix configuration to select suitable ubuntu version (ubuntu-22.04 is a valid label, see https://github.com/actions/runner-images?tab=readme-ov-file#available-images).

That's a great point! I also think that using a specific version label should be the way to go, if it's still available.

It appears that CI is legitimately failing though.

It was an encoding issue, while using universal_newlines, one may specify the right charset.
This is fixed by 52cceaf