Repository navigation
Code Review: show review comments with truthful file and line locations - #520
Conversation
Decision 5 (C65): a file can be marked viewed. The mark is kept on this Mac per account and belongs to one revision; a new head or base shows none. Marking folds the file; the tree and a count show progress. Nothing is read from or sent to GitHub. Decision 6 (C07): the unavailable new-tab button is removed. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Measured on real pull requests, every control was reachable with Tab but only through 34 to 124 presses between areas, and 147 arrows to the last of 151 files. F6 and Shift+F6 now move between the inbox, the pane tabs, the pane and the file list; [ and ] move to the previous and next file. Keys typed in a text field stay text. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Measured on 10 real pull requests, opening one took 2 to 5 s to its title and 6 to 7 s to its checks. Three causes are addressed: - Reads for an opened pull request go ahead of inbox lookups that have not started; a running read is never interrupted. - The overview uses resolve's read of the pull request once, within 2 s, under the same account, and only if nothing newer read it. - The merge base and the file page are read together between the two guarding reads. Checks and activity keep their order: a failed first read still costs GitHub nothing further. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ox rows GitHub comment bodies often carry HTML (details, sup, img, links). The overview showed the tags as raw text. Known tags outside code are now dropped, and an image becomes "Image: alt". Inbox rows show repo #number. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… first benchmark fixes (#319) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
CodeQL flagged the tag pattern (polynomial backtracking on an unclosed tag) and the comment strip (one replace can leave a new opener). A character scanner replaces both; hostile inputs are timed in the test. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The real-use benchmark found that comments on lines of the diff were counted but shown nowhere (F3).
The first activity page now also reads pulls/{n}/comments (a GET, first 100) and lists each with its
file, line and whether it is a reply. A failed read leaves the timeline usable and marks the page
partial. Whether a thread is resolved is not read; GitHub gives that only over GraphQL.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The overview always receives the write service, so the plain Comment,
Edit pull request and Request reviewers fallbacks ("This action isn't
connected yet.") were unreachable. The service is now required.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The bracket key focuses its file on the next animation frame. The keyboard test went on to focus the filter and type at once, so on a slow runner that focus arrived mid-typing and the filter kept only the first character. This failed the merge queue run for #506. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
# Conflicts: # apps/desktop/src/main/engine/code-review/read-service.ts # apps/desktop/src/renderer/code-review/PrOverview.tsx # apps/desktop/test/code-review-read-service.test.ts # docs/status.md
Preserve PR #513 source history and authors, current-main status facts, and the delivered CR-8 adversarial acceptance receipt. Add held-comment account and whole-PR revocation controls.
Distinguish file subjects and unavailable line/side metadata; reject malformed locations as partial coverage and update the connected write-boundary description. Address all three Minor review findings with reproduced service/Electron regressions.
A live public schema check showed file subjects may include line/side metadata. Preserve valid supplied fields instead of dropping the comment, allow omitted file locations, and retain strict malformed-data coverage. Reproduced with service and Electron regressions.
Record exact frozen-source full local and installed-CLI results, skipped/expected-fault boundaries, verified recovery and review/schema corrections. Keep source-hosted and protected delivery pending.
Preserve the cancelled first run and raise only the whole-job budget to 45 minutes. Native tests and assertions stay unchanged.
|
The first source iPhone job at Setup/build/native units consumed about 14.5 minutes before UI execution. A signed local correction at The original required macOS job is still running and will finish before the correction is pushed. No workflow was restarted. All 852 qualified runtime/test/dependency producer inputs remain byte-identical, but fresh corrected-head hosted qualification is still required before protected queue delivery. The cancelled first iPhone run will remain recorded; no parent issue, live/human acceptance, activation or release is accepted from this checkpoint. |
Record the successful original macOS job separately from the cancelled iPhone suite and the scoped budget review.
Code Review now shows the first 100 REST review comments in Activity and its Comments filter, with truthful file, line, before-side, reply and unavailable-location labels. Partial or capped reads remain explicit, and comment Markdown stays inert. This integrates #513 source
c599dd7a9fbaeb4d6946666d8859e892e6847aca, preserving its source history/authorship in the integration branch and recovery bundle, and retaining unrelated current-main status entries; held-comment tests also cover whole-PR denial and account replacement.Local runtime/test source
57d2773passed build/typecheck, 2,587 units with 36 skips, 530 desktop cases with 17 opt-in skips, eight site cases and four web cases, without retries. All 852 producer hashes remain unchanged.The original source
249154dmacOS job passed 2,587 units, 533 desktop cases with 14 opt-in skips, eight site cases and four web cases, without retry annotations. Its actual preview tree was independently reproduced and all four run artifacts passed API-digest and ZIP-integrity checks.Separate installed Claude Code 2.1.323 qualification passed four ordinary cases and two intentional fault controls against a scripted provider. These results do not establish a real-provider review, product live write, installed human acceptance or the skipped Local Models gates.
The first iPhone job hit its confirmed 30-minute job maximum after 65 native-unit passes and 39 UI-case pass lines, including one expected-failure control. Another case was still running, and its paired result bundle is unfinalized: the full iPhone suite did not pass. The separately reviewed correction raises only that whole-job limit to 45 minutes; tests, individual waits, assertions and retries are unchanged. Two workflow regression tests pass. Local actionlint has the same pre-existing code-quality scope limitation on baseline/current; the remaining lint passes with only that exact diagnostic excluded.
Corrected source
f9bef700390eb300d40424699193b2731d00e36cpassed 2,587 units with 36 skips, 533 desktop cases with fourteen opt-in skips in 25.5 minutes, eight site cases in 18.5 seconds and four web cases in 3.7 seconds, without retry annotations. Its iPhone job passed 65 native units, 43 ordinary UI cases and one intentional expected-failure control, with zero unexpected failures/skips and complete result bundles. Both jobs checked out preview8aaa11eb9468a2b23ff3529e86eb61382fb4d561, parents main927b443and the final source head, tree6a241c26d089ac4d7daf3e4b30b23a28257010c5; that tree was independently reproduced. All four source artifacts were verified against API SHA-256 and ZIP integrity, including the complete 87,207,851-byte iPhone artifact. These are hosted fixture/simulator results; physical phone, older Companion, two-LAN and human acceptance remain separate.The original run was allowed to finish and its first results remain retained. Final signed head
f9bef700390eb300d40424699193b2731d00e36cadds the reviewed CI budget correction and receipt documentation; the final commit changes documentation only. Corrected-head source CI run43366524777passed all twelve checks. Protected queue run43370489247passed on signed squash967b703768d040d64456c7d118548b97a3bae35b, which merged on 2026-11-13 at 05:53:08 UTC. Its parent is927b443414befe5f4300bb3f3d39af4666032609, and its tree exactly matches the qualified source tree6a241c26d089ac4d7daf3e4b30b23a28257010c5. The actual protected checkout passed 2,587 units with 36 skips, 533 desktop cases with 14 opt-in skips in 32.4 minutes, eight site cases in 23.0 seconds and four web cases in 4.2 seconds, with no retry annotations. Both protected artifacts passed API-digest and ZIP-integrity checks; the merge-group iPhone and coverage-upload jobs were skipped as configured. This is a squash delivery, not a claim that the integration branch’s original commit IDs became main ancestors. The feature review corrected all three Minor findings; the separate CI correction review found no severity findings, with none deferred.Resolved-thread state, inline placement and later comment pages remain open requirements. No parent issue, activation, release, live-provider/write or human/device gate is accepted by this slice. See the complete integration receipt and the retained first cancellation. Related: #323, #324 and #327.
The source-identical build also completed ten real-PR comparisons against GitHub’s pages, plus real instructions, divider and narrow-layout controls. The benchmark receipt and ranked remaining friction retain both diagnosed probe failures, scope restrictions and acceptance limits. No original-reference or human box is ticked by that benchmark.