I think this is also needed to fix #2 because right now a line like -- will be counted as a "line to be removed" while it's actually the signature marker.

What is the state of this changed? is it intended to be complete?

FYI: test failures on regression patches:

This is failing on the chunk header - but see below.

https://github.com/gitpatch-rs/gitpatch/actions/runs/18760185548/job/53522404788?pr=3#step:5:146

failed to parse "tests/samples/sample1.diff", error: Line 13: Error while parsing:  @@ -5,16 +11,13 @@
...

This patch seems to be being mis-parsed possibly around the end-of-chunk detection?

https://github.com/gitpatch-rs/gitpatch/actions/runs/18760185548/job/53522404788?pr=3#step:5:192

failed to parse "tests/wild-samples/sample2.patch", error: Line 27: Error while parsing: diff --git a/js/src/jsfriendapi.cpp b/js/src/jsfriendapi.cpp
....

It seems like conceptually a good idea to make use of the counts in the headers, but I'm not really sure from the PR or a brief read of the diff which one is meant to have priority if the number of lines expected disagrees with the lines that are apparently present?

Sorry for dropping the ball here. We've switched to our own fork of diffy and made it handle all edge cases we found in the wild: https://github.com/prefix-dev/diffy

IIRC there were some test cases here that did not contain valid diffs which made the line counting fail :) But I might misremember!