Merge commit diffs show all incoming changes instead of merge-only changes

[sdrshn-nmbr]

When a branch contains merge commits, the diff view shows all changes introduced by the merged branch rather than just the changes unique to the merge itself (conflict resolutions, manual edits). For example, merging origin/dev into a feature branch shows 65 changed files (+973/-3928) in GitButler, while GitHub shows 24 files for the same PR — because GitHub diffs against the merge-base.

Root cause

CommitDetails::from_commit_id() in crates/but-core/src/diff/commit_details.rs and commit_changes_with_line_stats_by_worktree_dir() in crates/but-core/src/diff/ui.rs both take only the first parent for diffing:

let first_parent_commit_id = commit.parent_ids().map(|id| id.detach()).next();

For regular commits this is correct. For merge commits, the first parent is the branch being merged into, so the diff includes everything from the other branch.

Proposed fix

For commits with >1 parent, diff against the merge-base of the parents instead of the first parent. This shows only what the merge introduced (conflict resolutions + manual changes), matching GitHub's behavior. The merge-base pattern already exists in the codebase (but-rebase/src/graph_rebase/cherry_pick.rs).

The change would be ~20 lines across the two functions. No frontend changes needed since the JSON serialization already maps diff_with_first_parent to changes.

Happy to put up a PR if this approach makes sense — or if a toggle between first-parent and merge-base diff would be preferred.

[Byron]

Thanks a lot for reporting and for researching the issue!

Yes, please do post a PR, ideally with a test. If a test seems harder to achieve, I'd try to add one for you, as this seems important to not break ever again :).

[Byron]

I don't think this is entirely correct. For instance here we have a merge-commit, and GitHub clearly shows a diff with the first parent of the merge commit.

Here a comparison with a locally changed version that takes the merge-base, and nightly which is unchanged, and GitHub.

❯ git diff  4b521213144eb2a3e2bc7db9c69c622ed1e7e4bc~1  4b521213144eb2a3e2bc7db9c69c622ed1e7e4bc --stat
[..]
 47 files changed, 5015 insertions(+), 4342 deletions(-)

We have a couple of different ways of diffing here, what GItHub and Git do is git show effectively, which diffs merges with the first parent.

❯ git show 4b521213144eb2a3e2bc7db9c69c622ed1e7e4b  --stat
[..]
 47 files changed, 5015 insertions(+), 4342 deletions(-)

Then there is git diff, which shows something very different, and it's different to -c and --cc as well. I didn't get the documentation on this topic.

❯ git diff 4b521213144eb2a3e2bc7db9c69c622ed1e7e4b  --stat
[..]
 10 files changed, 91 insertions(+), 194 deletions(-)

And finally, --cc (or -c), it's the same in this case:

❯ git diff 4b521213144eb2a3e2bc7db9c69c622ed1e7e4b  --stat --cc
 REPORT.md | 85 -------------------------------------------------------------------------------------
 1 file changed, 85 deletions(-)

So I think there is more here, and GitButler just does a git show equivalent with the first parent. Maybe it can be smarter and at least can show a combined diff, something which doesn't sound super trivial.