review: keep completed patch reviews when one patch fails

[arighi]

A local review of a series collects the per-patch results and returns early on the first patch error. When the LLM fails on one patch, for example because a response was truncated by an output token limit, the completed reviews of every other patch are discarded and the command prints only the error.

Keep the results of every other patch instead. Mark a failed patch as incomplete in the patches array and include its error. The combined result is marked as partial and includes a top-level error listing the failed patches.

The text report prints the error followed by the findings from the completed patches. It does not report "no issues found" when any patch failed, and the command still exits with an error status. The progress display also marks failed patches as incomplete.

The daemon reviews one patch per worker, so a failed patch still produces a failed review that is retried. The worker now reports the patch status and any error as part of its normal result, allowing the daemon to record the application status and take the tool-error path instead of the missing-status path.

Assisted-by: LLM

[sashiko-bot]

Sashiko review

Commit 1/1 — 224fe273 review: keep completed patch reviews when one patch fails

• [LOW] In src/local_review.rs (run_worker_in_worktree), the review_errors
  vector is populated with strings formatted as patch {patch_index}: {err}
  and then sorted. Because sorting strings applies a lexicographical sort,
  patches with indices 10 or higher will incorrectly sort before patch 2.
  This is a cosmetic misordering issue that affects the top-level combined
  error string for failed patches.

Full review and stage logs on sashiko.sashiko.dev

[sashiko-bot]

Sashiko review — v2

✓ No issues found across 1 commit.

Full review log on sashiko.sashiko.dev