Repository navigation
Conversation
|
👋 Hi! This PR is currently in draft status. Workflow runs won't be auto-approved until it's marked as ready for review. When you're ready, click "Ready for review" and the workflows will be approved on the next cycle automatically. |
UX Review (Opus 5.5, fork) — 🔴 BLOCKUX-level review of UX-Verdict: BLOCK cannot evaluate: missing a screenshot of the recents-failure Retry on Browse, in both its idle and its "Retrying…" state; no image reached this review. Evidence gaps
Watch
[UX-REVIEWED] e2d9750 Description read[DESCRIPTION-READ] f6e43083636f9e24261d5e0e5da2de6c82ab47a97e8920a779e08ea55c85f32f That sha256 names everything this verdict read: the PR description, plus the 3 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Design Review (Opus 5.5, fork) — 🔴 BLOCK (blocking)Design-level review of Design-Verdict: BLOCK cannot evaluate: missing a screenshot of the recents-failure Retry on the Browse notice at this revision. The description also describes a different change from the diff. BlockersNo rendered evidence for a new control WatchThe description does not match the diff. The problem it states is already fixed on the base:
The
[DESIGN-REVIEWED] e2d9750 Description read[DESCRIPTION-READ] e80b27a7bf4bf40ab3456a5d870aa9f9e4a075b45fc2db353b7b25c0e6a39d4b That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5, fork) — 🔴 BLOCKPremise-level review of First-Principles-Verdict: BLOCK The description describes a different change: base already names the recents failure, and the Not justified as shipped
What this change shipsInventory (3 items) — 1 justifiedIntent: let a user re-read a failed recents list without closing and reopening the picker. The title calls it a FIX; it is an ADDITION.
Blockers
[FIRST-PRINCIPLES-REVIEWED] e2d9750 Description read[DESCRIPTION-READ] 5f38a62965a5062fa7306abc0f73537b5f382a1775980734bfb23404084fbfee That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review (fork) — ✅ no blocking findingsReviewed Review detailsNo findings. |
Opus 5.5 Review (fork) — ✅ no blocking findingsReviewed Review detailsPressing Enter to open a folder leaves the user unprotected from the late tab switch that a mouse click on the same folder is now protected from. FINDING — website/src/components/ProjectPicker.tsx:917 — When the user presses Enter in the path field to open a folder, the code runs [OPUS-REVIEWED] e2d9750 |
f8fba90 to
c75a8b9
Compare
c75a8b9 to
cb4df84
Compare
3b40de0 to
53966bb
Compare
81a57ac to
e2fb6f3
Compare
e2fb6f3 to
e902d04
Compare
e902d04 to
85ba53f
Compare
85ba53f to
81553ef
Compare
|
🤖 Kiro Crew [operator: dodgeblaster#b5f0d847]: This PR has been inactive for 7+ days. I reviewed the blockers but they require your input:
When you've addressed these, the pipeline will re-assess on its next cycle. |
|
Triage (PR-overlap scan vs Already on main via #7068 (
Not on main, so this is still unique to this PR: a Retry that re-reads recents, the Rebase needed: the branch is 2266 commits behind main, and Merge discussion: #16611 (Favorites tab) rewrites the same recents landing-tab and No breaking change or P0 found, so no change request. |
The picker now says the recents list could not be read instead of silently falling back to Browse, and pairs that notice with a Retry that re-reads it.
|
Thank you @rnoack1 for this fix and for the careful write-up and tests. The naming half of this PR had already landed on main via #7068, so we carried the remaining part, the recents Retry, onto current main in #21148. Your commit is cherry-picked there with your authorship kept, and the follow-up commit credits you as co-author. One change from your version: the Closing this in favour of #21148. Your branch is left in place. |
Problem / Motivation
A failed recents read in the project picker was swallowed.
api.recentProjects()ended in
.catch(() => setTab('browse')), which acts on the failure but throws thecause away — so the Recent tab silently held nothing, and reopening the popover
looked like it had lost the list. Nothing named the failure and nothing offered to
re-read it.
Why it matters
A failed read is indistinguishable from an empty one, so the user is told "No recent
projects" about a list that was never fetched. The only recovery was to close and
reopen the popover, which re-runs the same read with no indication that a read is
what failed.
What changed (motivation → approach → change)
The read now reports whether it landed and how many rows it produced, so a single
rule decides the landing tab for both the initial open and a retry, rather than that
rule living twice. When it failed, the picker says so and pairs the notice with a
Retry that re-reads the list and, on recovery, returns focus to the first control
that still exists.
The notice follows the user rather than sitting on one tab. The failure itself is
what selects Browse, so a Recent-only notice would never be seen: it shows on either
tab until she chooses one deliberately, and only on Recent thereafter.
RetryControlis its own component rather than markup at the call site because theparts that are easy to get wrong are the invisible ones — the
aria-busypair, the"<label>: <cause>"accessible name, and returning focus to the button after afailed attempt so a keyboard user does not spend a Tab per retry.
Deliberately not included, both narrowing this unit:
useQuerymigration. An earlier revision of this PR converted recents fromuseStatetouseQueryto obtainrefetch. Extracting the existing fetch into acallback yields the same handle, so the retry does not require that architectural
change or the tab-landing bookkeeping that came with it.
controlRefonRetryControl. Nothing in this unit reads it, so it wouldship as an unused optional prop. The surface that needs it (keyboard nav handing
Tab and Enter to the control) can add it with its own caller.
This unit does not touch the listing failure notice. That is already on mainline
via #11424, and is left exactly as it is.
Tests
website/src/test/ProjectPicker.recentsRetry.test.tsx, 7 cases: the failure isnamed; nothing is said on success; Retry clears the notice and shows the rows; a
failed retry keeps the notice and leaves the control enabled; the notice reaches
Browse, which the failure selected; it stops following once Browse is chosen
deliberately; and it is still there on the Recent tab she opens to look for it.
Break-armed: restoring the old
.catch(() => setTab('browse'))swallow fails 6 ofthe 7, and the case that survives is "says nothing when the read succeeds" — the one
that should pass without this change. So the suite discriminates rather than
failing wholesale.
Manual verification
Verified locally at this revision:
tsc -p tsconfig.app.jsonclean,eslint srcclean,
i18n:check/check-i18n-keys/lint:i18nall pass, and 185 tests greenacross the seven affected suites (
ProjectPicker.recentsRetry,ProjectPicker,ProjectPicker.keyboardIsolation,ProjectPicker.pathDelimiter,FolderConfigModal,WorkspacePickerCoverage,ImeEnterGuardSites).Two catalog keys across all 13 locales. The English pair is in
en.jsonalongsideevery other
projectPickerkey — noten.manual.json, which carries no suchnamespace — and
en-XAis regenerated byi18n:pseudorather than hand-written.Screenshots
One frame applies, captured by the harness at an earlier revision of this branch
(
d55a0105) and showing precisely what this revision renders: the notice on theBrowse tab, which the failed read selected, with Retry beside it.
The notice row's markup is unchanged between that revision and this one — same
wrapper classes, same
ErrorNotice, same gate shape; only the state variable wasrenamed — so the frame is faithful to what ships here.
The harness's other five frames are not cited, because they show behaviour this
unit no longer ships: a recents loading state, and the listing-failure retry.
loading_recent,listCauseandlistRecoverableall appear 0 times in thisrevision.
Pattern harvest
Rule candidate: eslint
Pattern: a rejection from an
api.*call discarded by a handler that neithersurfaces nor stores the cause, so the caller renders its empty state as though the
read had succeeded. At this revision's base (
upstream/main) the instance this PRfixes was
website/src/components/ProjectPicker.tsx:146—.catch(() => setTab('browse')), which acts on the failure but keeps none of it.The retirable rule is narrow and syntactic: flag a
.catchon anapi.*promisewhose handler binds no rejection parameter, or binds one and never reads it. A
handler that assigns the cause to state or rethrows is not caught. This belongs in
lint rather than a review prompt because the defect is invisible at the call site by
construction — the code reads as though failure were handled.
One instance remains unfixed and would be caught by that rule:
website/src/components/WorkspacePicker.tsx:35, a body-less.catch(() => {})onapi.browseDirs, which leaves its list claiming no subdirectories for a directorythat was never read. It is left for the unit that owns that surface.