While this might do for a temporary fix, I don't like this solution because while it filters most things out, it could still result in a false positive.

The real problem here is this line:

git_cmd = "#{Git::Base.config.binary_path} #{global_opts} #{cmd} #{opts} #{command_opts[:redirect]} 2>&1"

Which forces STDERR to be captured along with STDOUT. This causes problems in a numerous places.

Needless to say, fixing this would take a lot more effort than your quick hack.

Let me consider this a bit more. As always, I am open to additional feedback.

Thank you for following up! I had figured that stderr was getting conflated with the rest of the output but hadn't found where the command was actually run, and wasn't certain to what extent the library relies on stderr in other scenarios.

I definitely agree that my solution is crude, and risks silently swallowing other types of problems; but at minimum, process_full_diff should probably offer a more graceful failure mode, as the current exception for unexpected output is very difficult to reason about.

Thanks again, both for the extra context and for your contributions to open source.

@aalong-tr I am closing this in favor of #617 which will fix the underlying problem reported in #574. Happy to have your feedback there if you would like.