Use gitoxide in get_diff

[cruessler]

This is a draft PR that is still somewhat rough around the edges. It replaces the libgit2 logic in get_diff with gitoxide. A few things don’t work yet, but the code is already capable of generating correct diffs in some cases. I hope to get this PR into a reviewable/mergable state by the end of next week, but I won’t make any promises. :-)

This will hopefully also address #2665.

Update 2026-01-16

I’ve updated to the latest changes in gix_diff’s API. Some keyboard shortcuts don’t yet work as expected in all scenarios which I will have a look at next.

Update 2025-08-08

I’m now down to just 4 tests not passing! Also, as I’d hoped, this PR addesses #2665, completely eliminating any CPU load because binary files are ignored by the diff algorithm.

There’s a few remaining issues/open questions:

• gitoxide does not provide structured diffs yet which is why this PR, as a workaround, resorts to string matching on a given hunk’s lines. This will hopefully be addressed upstream through Add HunkHeader and DiffLineType to ConsumeHunk::consume_hunk GitoxideLabs/gitoxide#2102. Once gitoxide has dedicated data structures, I should be able to simplify this PR considerably.
• gitoxide does not provide extended headers such as @@ -315,6 +315,12 @@ impl ConsumeHunk for FileDiff { the same way git2 does (unless I’m missing something). The impl ConsumeHunk for FileDiff { is not part of what we get from gitoxide. As far as I can see, this only affects the UI.
• gitoxide produces different diffs than git2 in certain cases which is known as the “slider problem”. I hope to be able to address this upstream as well, but it will take time and it will block this PR if we’re aiming for git2 compatibility (and I think there’s strong reasons to do that).

[extrawurst]

it looks like GitoxideLabs/gitoxide#2102 is merged, so we can move forward on this, right?

[cruessler]

it looks like GitoxideLabs/gitoxide#2102 is merged, so we can move forward on this, right?

Yes, we can, thanks for the reminder! I’ll have a look once I get to it.

[cruessler]

I’ve merged the latest master and updated to the new API. This still needs a bit of work as keyboard shortcuts don’t work in all cases yet. Also, I need to address the test failures. :-)