feat: comment on an issue whose fix reaches beyond the pull request diff

[bugale]

What

An issue was commented on only when the pull request diff showed every line it spans, since GitHub rejects a comment anchored outside the diff. That made the action fail a step without any comment explaining why.

Now the comment is anchored on the run of lines the diff does show around an added line of the range — the widest anchor GitHub accepts — and a fix reaching beyond that anchor is rendered as a plain code block with a note naming the lines it covers, instead of a suggestion that would rewrite lines the pull request does not show.

Why

Found on a real clang-format run (winagent run 11519055):

• the PR changed one line (877) of a 16-line IoCreateFileEx(...) call;
• git-clang-format correctly passed --lines=877:877, but clang-format's smallest unit is the whole statement, so it re-laid-out lines 868–883;
• the PR diff has a single hunk covering 874–880.

isNewIssue (some line added) kept the issue → the step failed. isCommentableIssue (every line in the diff) rejected it → no comment at all. This closes that gap: since isNewIssue guarantees at least one added line, an anchor always exists.

Implementation

• isCommentableIssue is replaced by getCommentAnchor, which returns the normalized path, the anchored range and whether it is partial, or undefined when the diff adds none of the issue's lines. It grows the anchor from the first added line outwards while the neighbouring lines are in the diff, staying inside both the issue range and a single hunk.
• buildCommentBody takes a partial flag and renders the fix as a plain fence plus the note instead of a suggestion.
• addComments uses the anchor for path/line/start_line, so the comment lands on lines GitHub accepts.

Testing

• Unit tests for getCommentAnchor: full range, truncated range, path relativization, and the cases that anchor nothing (no added line, unknown file, missing path/line, end line before start line).
• A fixBeyondDiff suite mirroring the winagent case exactly — issue 868–883 against a @@ -874,7 +874,7 @@ hunk — asserting the 874–880 anchor, the plain-fence body with the note, and that a fully shown fix still renders as a suggestion.
• Verified end-to-end against the real data: the actual clang-format diff (reproduced locally byte-identical to the run's SARIF) plus the real PR diff now yield an anchor of 874–880 and a comment carrying the full 15-line replacement, where the run produced nothing.

[github-actions]

🎉 This PR is included in version 5.1.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀