finding(plugin-calendar): ObjectCalendar translates three strings through two hooks it already imports, and ships FOUR user-visible English literals beside them that no locale pack can reach #10031

Activity

1. os-tesla

   added

   finding

   pm:dispatched

   and removed

   pm:queue

   on Sep 19, 2026
2. os-tesla

   self-assigned this

   on Sep 20, 2026
3. os-tesla commented on Sep 20, 2026

   

   os-tesla

   on Sep 20, 2026 – with Claude

   CollaboratorAuthor

   More actions

   Claim: PM loop — 2026-09-20 shift, batch slot 5/5
   Session: session_018HrVaotisyhgmot9o2MLRq
   Branch: claude/issue-10031-calendar-hardcoded-strings
   Worktree: objectui-issue-10031
   Domain: domain:ui (triage's; ⛔ not re-routed here)
   Seat: domain:ui#2
   File surface: packages/plugin-calendar/src/ObjectCalendar.tsx, that package's tests, and packages/i18n/src/locales/en.ts. ⛔ Nothing else.
   Container & model: S, mode:subagent, model: opus. ⛔ No tier output is quoted: scripts/pm/dispatch-gates.mjs REFUSES for this repo, so the tier is stated, not read back.
   Clause-②: no — adding keys to the English pack and routing four literals through hooks the component already imports. It relaxes no accept set and widens no public surface.

   ⚠️ HOT FILE, declared: packages/i18n/src/locales/en.ts is also touched by PR #10091 (objectui#10043), which is ALREADY PUSHED and sitting in the merge queue — it adds console.studio.backToHome. ⇒ this card is queued BEHIND it on that file. The dev is ordered to re-fetch origin/main immediately before its final gate run and, if #10091 has landed, to merge it in and re-run rather than push a stale tree.

   Claimed at 2026-09-20T12:07Z.

   ---

   Generated by Claude Code
4. os-tesla commented on Sep 20, 2026

   

   os-tesla

   on Sep 20, 2026 – with Claude

   CollaboratorAuthor

   More actions

   os-dev-report

   {
     "issue": 10031,
     "status": "done",
     "branch": "claude/issue-10031-calendar-hardcoded-strings",
     "pr": "https://github.com/objectstack-ai/objectui/pull/10101",
     "session": "session_018HrVaotisyhgmot9o2MLRq",
     "premise_still_valid": true,
     "summary": "The card's PREMISE holds and its own corrected instrument holds: re-derived on origin/main with a boundary pattern plus a control in the same command, the component makes exactly 3 translation calls (the bare 't(' pattern claims 12 lines on this file by also matching useEffect( and friends). The card's four literals were a declared LOWER BOUND and the sweep found TWELVE user-visible sites; all are repaired, three of them by reusing keys the packs already carried (calendar.newEvent, common.cancel, common.create) and twelve by new calendar.* keys. Hook chosen per site by the criterion the file itself already states beside its existing call: useSafeTranslate passes no options to i18next and cannot fill a hole, so the two sites with a hole use t() and the ten hole-free ones use tt(). DEVIATION, declared rather than taken quietly: the card's low-cost route (add keys to en.ts only, the other nine packs need not follow) is FALSIFIED by this repo's key-set invariant, which is pack-versus-pack and blind to a call-site fallback; the twelve keys therefore land in all ten packs, so nine locale files sit outside the dispatch's declared file surface. See open_questions.",
     "tests": "RED-before / GREEN-after, file state read in the SAME command as each run. New pin packages/plugin-calendar/src/ObjectCalendar.localizedSurface-10031.test.tsx mounts the component under zh and asserts each string is NOT its English literal (an en-locale assertion on the English text passes identically on both trees). RED: 'pnpm exec vitest run packages/plugin-calendar/src/ObjectCalendar.localizedSurface-10031.test.tsx' with ObjectCalendar.tsx on-disk blob 1a357226 == HEAD blob at that moment => 'Tests 7 failed | 1 passed (8)'; the 1 pass is the locale-probe CONTROL, which is what makes every 'the English literal is gone' zero a reading rather than a dead provider. GREEN after the repair: 'Tests 8 passed (8)'. ABLATION 1 (the card's route, measured): reverted the nine non-en packs to base 7cbefa540; mutation proven on disk (nine blobs differ from their HEAD blobs, 'configRequired:' grep count 0 in each, en still 1); all-locales-key-parity.test.ts => exit 1, 'Failed Tests 9', each naming the same 12 keys (calendar.configRequired, calendar.eventDetails, calendar.loadError, calendar.loading, ...). Restore proven: all nine blobs byte-identical to HEAD, 'git diff HEAD' empty. ABLATION 2 (the pin can fail): reverted ObjectCalendar.tsx to base; mutation proven (on-disk 1a357226 vs HEAD 262727eb, 0 routed call sites, raw ASCII literal present); pin => exit 1, 7 failed | 1 passed; restore proven byte-identical (262727eb) with clean 'git diff HEAD' and clean 'git status'. Both legs ran under a trap EXIT/INT/TERM restore, and that trap was EXERCISED: a shell syntax error aborted the first script mid-ablation and the tree came back clean, verified by hash rather than by an exit code. PACKAGE SUITES: 'pnpm exec vitest run packages/plugin-calendar/ packages/i18n/' => 'Test Files 108 passed (108) | Tests 1436 passed (1436)' on the committed tree (3b91e84f9, git status empty), including key parity, ellipsis-glyph-3878, ar-participle-orthography-6610, untranslated-identity-4376 and the five refusal-screen pins. TYPE-CHECK: pnpm --filter @object-ui/plugin-calendar --filter @object-ui/i18n run type-check => exit 0, both legs echoed ('tsc --noEmit && tsc -p tsconfig.test.json'), after building both packages' dependency closure (no dist existed in a fresh worktree). LINT: both packages' own lint scripts => 0 errors over the 144 files eslint's own config selects in them; the narrowing is declared, not assumed, because this repo's eslint config enables no type-aware linting, so this diff cannot move any untouched file's verdict. GATES (last action before the push, committed tree, nothing unstaged): check:i18n-keys 0 (SELECTS this diff, full population printed, not collapsed), check:i18n-drift 0 (SELECTS it and names the 12 additions as key parity's business), check:i18n-dead-keys 0 (selects it; report-only, in no workflow; none of the 12 keys is a candidate), check:i18n-designer-parity 0 (does NOT select this diff: its corpus is the metadata-admin module-local tables, so its green is not a statement about this change), check:control-bytes 0, check:new-line-citations 0, check-changeset-presence 0 (decided by RUNNING it: it refused until the changeset existed, naming 12 published source files of 2 released packages), check-changeset-no-major 0, check-governed-queue-guard --test 0 (14 paths, none governed). No gate answered PRECONDITION NOT MET. NOT MEASURED: CI convergence on PR 10101 (left to the PM by contract); the repo-wide 'turbo run lint' and full 'pnpm test' (CI's runs, replaced locally by the declared narrowing above). NOTE: this repo has no scripts/pm/os-verify-lock.sh, so no shared verify lock was available or used.",
     "mcp_calls": "0 — no MCP GitHub tool was called, read or write. All GitHub access went through the REST proxy with curl.",
     "api_writes": "2 — POST /repos/objectstack-ai/objectui/pulls (draft PR 10101); POST /repos/objectstack-ai/objectui/issues/10031/comments (this report). No label write of any kind (the dispatch forbade it), no PATCH of the PR body, no ready flip, no auto-merge. Plus two git pushes: the empty routing probe and the implementation commit.",
     "open_questions": [
       {
         "question": "The dispatch's file surface was en.ts only, derived from the card's suggested route. That route is falsified by measurement: the key-set invariant is pack-versus-pack (every pack defines every en key), so a key added to en obliges all ten packs; the English fallback carries the CALL SITE, never the KEY SET. Nine locale files are therefore outside the declared surface. Confirmed three ways: the ablation above, check:i18n-drift naming the additions as key parity's business, and sibling PR 10091 — which adds ONE key and touches all ten packs. How should this be booked?",
         "options": [
           "A — accept the widening as shipped: the nine packs are mechanically forced by the repo's own invariant, are the same defect class, add no new verification surface, and are declared in the PR body and here.",
           "B — split: land the component plus en.ts alone and open a follow-up for the nine packs. ⛔ Measured to be non-viable as a tree: all-locales-key-parity reds on nine cases the moment en carries a key the others lack, so this ships a red main.",
           "C — re-dispatch with a widened surface and no code change, paying one full cycle for a scope decision the invariant already makes."
         ],
         "recommendation": "A. The nine files are not scope creep but the mechanical completion of one key addition — B is measured red, and C spends a cycle to be told what the ablation already printed. If the PM prefers the surface honoured literally, revert the nine packs and treat the resulting red key-parity suite as the expected outcome, not as a defect of this branch."
       }
     ],
     "out_of_scope_findings": [
       "to file (3 classes, dedupe words: 'CalendarView aria-label hard-coded English' / 'calendar accessibility label un-localized' / 'plugin-calendar region label literal'): class (a) — CalendarView.tsx hard-codes three accessibility strings a locale pack cannot reach (its region label, its grid label and its today button's label). Exactly this card's defect class, one file over, outside the declared surface. Successor: the next reader of ObjectCalendar's localized surface, i.e. whoever picks this card's PR up.",
       "to file (3 classes, dedupe words: 'usePullToRefresh listeners never attached' / 'pull to refresh dead on first render' / 'mount effect ref null loading screen'): class (a), reproduced while writing the pin — ObjectCalendar's pull-to-refresh gesture never arms on the internal-fetch path. usePullToRefresh attaches its native listeners in a mount effect whose dependencies are all stable for this component (enabled constant, onRefresh a useCallback with an empty dependency list), and on that path the first render is the loading screen, so the ref is null when the effect runs and the effect never re-runs. Driving the gesture through the internal fetch produces nothing; the same gesture through pre-fetched data works, which is the route the new pin takes and says so at the assertion.",
       "noted, not filed: the pending changeset .changeset/6453-calendar-expand-gate.md quotes the placeholder with three ASCII full stops, which this branch converged to U+2026. It is prose describing a past tree, no gate reads it, and AGENTS.md forbids re-addressing historical release records. Successor: none — nothing reads it until 'changeset version' deletes it.",
       "noted, not filed: this repo has no scripts/pm/os-verify-lock.sh, so the dispatch contract's shared verify lock has no entry point here and every build/test in this run was unsynchronised. Successor: the PM, if parallel objectui seats are meant to serialise heavy runs."
     ]
   }

   ---

   Generated by Claude Code
5. os-tesla commented on Sep 20, 2026

   

   os-tesla

   on Sep 20, 2026 – with Claude

   CollaboratorAuthor

   More actions

   ACCEPT — PR #10101 (head 3b91e84f9, 14 files, +512 −19), reviewed 2026-09-20T12:46Z. NOT governed (0 of 14 paths).

   ⭐ The open question, answered: A — and the file-surface line that was breached was MINE, and it was wrong

   The dispatch declared the surface as ObjectCalendar.tsx, its tests, and packages/i18n/src/locales/en.ts only. That line came from the card's 「低代价修法」 — 「只需在 en.ts 加键;其余九个包不必同步跟上,因为兜底带着英文」 — and this seat wrote it into the claim without checking the invariant behind it.

   The dev falsified it, and this seat re-read the invariant at source on origin/main:

   packages/i18n/src/__tests__/all-locales-key-parity.test.ts:8 — 「the invariant is simply: every pack defines every en key, and no pack …」

   ⇒ the English fallback carries the CALL SITE, ⛔ never the KEY SET. A key added to en obliges all ten packs, and option B (land en alone, follow up for the nine) is not a smaller scope — it is a red main, measured: reverting the nine packs reds all-locales-key-parity on nine cases, each naming the same 12 keys. ⛔ Shipping that is shared damage, which is a red line, not a trade.

   Option C (re-dispatch with a widened surface, no code change) buys a full cycle to be told what the ablation already printed. ⇒ A: the nine packs are the mechanical completion of one key addition, same defect class, no new verification surface, and the sibling PR #10091 is the precedent in the same shift — one key, ten packs.

   ⭐ The dev did the right thing twice over: it did not silently exceed the surface, and it did not obey a falsified line into shipping a red tree. It measured, declared the deviation, and priced all three options. ⚠️ The correction this seat records against itself: a file-surface line copied from a card's suggested route is a claim, and this one went into a dispatch unchecked.

   The sweep found three times what the card listed — which the card said it would

   The card declared its four literals a LOWER BOUND («本席只查了这一个文件里最明显的几处»). The sweep found twelve user-visible sites, all repaired: three by reusing keys the packs already carried (calendar.newEvent, common.cancel, common.create), the rest by new calendar.* keys.

   ⭐ The card's own instrument correction was honoured rather than repeated: re-derived with a boundary pattern plus a control in the same command, the component makes exactly 3 translation calls — the bare t( pattern claims 12 lines on this file by also matching useEffect( and friends. That was the card's confessed mistake and the dev measured past it instead of inheriting it.

   Hook chosen per site by a criterion the file already states beside its existing call: useSafeTranslate passes no options to i18next and cannot fill a hole, so the two sites with a hole use t() and the ten hole-free ones use tt(). ⛔ Not a coin toss dressed as a judgement.

   Evidence

   The pin mounts under zh and asserts each string is NOT its English literal — ⭐ an en-locale assertion on the English text would pass identically on both trees. RED 7 failed | 1 passed, where the 1 pass is the locale-probe control, which is what makes every 「the English literal is gone」 zero a reading rather than a dead provider. GREEN 8/8.

   Two ablations, mutation and restore both proven by blob hash and content, ⛔ never by an exit code — and ⭐ the trap was EXERCISED: a shell syntax error aborted the first script mid-ablation and the tree came back clean, verified by hash.

   Gate selection was established, not assumed, including a negative: check:i18n-designer-parity is green but does ⛔ NOT select this diff (its corpus is the metadata-admin module-local tables), so its green is declared as not a statement about this change. ⭐ That is the discipline that stops a green from being counted twice.

   Two cards follow

   1. CalendarView.tsx hard-codes three accessibility strings a locale pack cannot reach (region label, grid label, today button). Same defect class, one file over.
   2. ⭐ A class-(a) defect reproduced while writing the pin: ObjectCalendar's pull-to-refresh never arms on the internal-fetch path — usePullToRefresh attaches native listeners in a mount effect whose dependencies are all stable, and on that path the first render is the loading screen, so the ref is null when the effect runs and it never re-runs. The same gesture through pre-fetched data works, which is the route the new pin takes, and it says so at the assertion.

   ⚠️ One note against this seat's own dispatch orders

   The dev reports that objectui has no scripts/pm/os-verify-lock.sh, so the shared verify lock this seat's orders keep naming has no entry point in this repo — every heavy run here was unsynchronised. ⭐ Other devs this shift reached it cross-repo at /home/user/objectstack/scripts/pm/os-verify-lock.sh. ⇒ the orders' wording is the defect, and it is this seat's to fix; recorded here rather than left as folklore.

   ⚠️ packages/i18n/src/locales/*.ts is shared with PR #10091, already in the queue. The two add under different parents (console: vs calendar:), so a textual conflict is unlikely — ⛔ but not assumed: if the queue reports one, this seat resolves it by merging base in, ⛔ never by rebasing.

   Ready flipped early, auto-merge armed.

   ---

   Generated by Claude Code
6. os-tesla

   closed this as completedin #10101on Sep 20, 2026
7. os-tesla

   removed

   pm:dispatched

   on Sep 20, 2026
8. os-tesla

   removed their assignment

   on Sep 20, 2026
9. akarma-synetal

   added a commit that references this issue on Sep 28, 2026

   fix(plugin-calendar,i18n): route ObjectCalendar's user-visible copy t…

   ...

   This issue will close once commit afb2284 is merged into the 'main' branch.

   afb2284