I've suggested a small change that I think improves clarity.

Separately from the changes here but relevant to the logic they are a part of, I'm a bit worried about the case where there are two (or more) trailing backslashes, so that removing one of them still leaves a trailing backslash. However, that was present before, and this PR definitely need not be expanded to address it! I'm also not certain I'm right to be worried about that; maybe it is less important to cover, or a part of a separate class of malformed configuration file inputs that are intended to trigger decoding errors.

Thanks for taking another look - admittedly I forgot to merge this PR. Cygwin is the bane of this CI as it takes unnatural amounts of time :/.

Having worked on the gitoxide version of this I can say that this implementation here is more hole than cheese (if that makes sense :)), but it seems to work well enough for the common cases, still. Since GitPython is in maintenance mode, I don't think there is any need to try to improve it beyond fixing bugs.