Fix review screen showing worktree branch instead of stored branch

[wesm]

Summary

• The TUI review screen resolved the branch name dynamically via git name-rev, which can return a worktree branch when the same SHA is reachable from multiple refs. The queue view already handled this correctly by preferring the stored job.Branch.
• fetchReview() now checks review.Job.Branch first and only falls back to git.GetBranchName() when the stored value is empty (for older jobs that predate the Branch column).
• Added a regression test for the stored-branch priority.

🤖 Generated with Claude Code

[roborev-ci]

roborev: Combined Review (39144a2)

Verdict: The PR successfully
updates review loading to prefer stored job branches over dynamic git resolution, but introduces a potential display regression for range reviews and lacks proper test coverage for the modified logic.

Medium

Potential behavior regression for range reviews
File: cmd/roborev/tui/fetch.go:508- 511
branchName = review.Job.Branch is now applied unconditionally before the range check. This means start..end range reviews (which previously stayed branchless in this code path) can now inappropriately display a branch if job.Branch is populated.
Suggested
fix: Gate the stored-branch usage with the same single-commit condition to preserve existing behavior for range reviews (and optionally exclude "dirty" if that should remain branchless too).

Ineffective regression test
File: cmd/roborev/tui/review_branch_test.go: 33-53 (TestTUIReviewMsgUsesStoredBranch)
The new regression test hardcodes reviewMsg.branchName = "main" instead of invoking the modified fetchReview() logic. It only verifies the Bubbletea model's message handling, completely bypassing and failing to validate the
new branch resolution logic.
Suggested fix: Update the test to call fetchReview() (or the relevant command generator), execute the returned tea.Cmd, and assert the computed branchName on the resulting reviewMsg prefers review.Job.Branch.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (f46a69d)

Verdict: The PR successfully refactors branch name resolution to prioritize stored job branches, but requires a
fix to prevent the "(none)" sentinel from rendering in the UI.

Medium

• cmd/roborev/tui/fetch.go:515-520, cmd/roborev/tui/render_review.go:52-53

  reviewBranchName now returns any non-empty stored job.Branch verbatim. In this codebase, "(none)" is already a client-side sentinel for “no branch” (cmd/roborev/tui/helpers.go:24-25), and
  the same commit already backfills that sentinel into jobs for dirty/task/remote/detached cases (cmd/roborev/tui/fetch.go:395-413). With this change, those reviews will render headers like ... on (none), which is a user-visible regression
  . Suggested fix: treat branchNone the same as empty in reviewBranchName and only display real branch names.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (ab31e87)

Verdict: The changes successfully update the review fetch logic to prefer the stored branch name, and no medium or higher severity issues were found.

All agents reviewed the commits and agree
the code is clean. No critical, high, or medium severity issues were identified.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (24879f5)

Verdict: The changes successfully wire the stored branch through the review-loading path to improve TUI display accuracy, but a medium-severity issue exists with the fallback behavior for the branch sentinel.

Medium

• cmd/roborev/tui/fetch. go:519-523 (Commit 24879f5)
  reviewBranchName() treats branchNone as "don't return the stored branch", but it still falls through to git.GetBranchName() for single-commit reviews with a RepoPath . That defeats the commit's stated behavior of omitting the branch for sentinel-backed jobs, and can reintroduce the original wrong-branch display if the SHA is reachable from another local branch/worktree ref.
  Suggested fix: Return "" immediately when job.Branch == branchNone, or
  only run the git fallback when job.Branch == "" (not when it is the sentinel). Add a test covering Branch: branchNone with a non-empty RepoPath.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (76cbabb)

Verdict: All review agents agree the
code is clean and found no issues.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)