Repository navigation
finding(dashboard): ObjectPivotTable carries the same i18n-resolver ref workaround #5587 is removing from ObjectChart #5625
Description
Observation-class finding, filed unassigned while implementing #5587. Found by the over-delete sweep for that card, not by looking for it.
What is there
#5587 removes a ref wrapped around an i18n resolver in packages/plugin-charts/src/ObjectChart.tsx. Sweeping the repo for other readers of the symbol being deleted turned up a second, independent instance of the same workaround in a different package:
packages/plugin-dashboard/src/ObjectPivotTable.tsx:83-93
// i18n: translate field display labels and select option labels via the
// standard object/field translation conventions. Held behind refs so the
// metadata-derivation effect doesn't need them in its dep array (the i18n
// hook returns fresh function identities each render).
const { fieldLabel, fieldOptionLabel } = useSafeFieldLabel();
const fieldLabelRef = useRef(fieldLabel);
const fieldOptionLabelRef = useRef(fieldOptionLabel);
useEffect(() => {
fieldLabelRef.current = fieldLabel;
fieldOptionLabelRef.current = fieldOptionLabel;
}, [fieldLabel, fieldOptionLabel]);
Read at lines 108 and 116, inside the metadata-derivation effect whose dep array is [dataSource, schema.objectName] (line 127).
The stated reason — "the i18n hook returns fresh function identities each render" — is the same condition #5564 measured and PR #5585 fixed, so the same precondition is gone here too.
Why it is filed rather than folded into #5587
Deliberately not included in #5587's PR. It is a different package (plugin-dashboard, which declares @object-ui/i18n at package.json:30), a different test suite, and — this is the part that makes it a separate review rather than a mechanical repeat — the ref hides the resolvers from a metadata-derivation effect rather than from a fetch callback. Removing it changes when that effect re-runs and what it re-derives, which is not the same behaviour question #5587 answers, and it deserves its own before/after test rather than riding along.
Suggested shape, if queued
Mirror #5587: drop both refs, depend on fieldLabel / fieldOptionLabel directly in the effect's dep array, and pin it with a test that counts effect runs (or getObjectSchema calls) across re-renders both inside and outside a provider. #5587's ObjectChart.fieldOptionLabelRefetch.test.tsx is a usable template, including the ordering note about createI18n registering a process-global instance.
Worth checking as part of that work whether the two derived state maps (setFieldLabelMaps / setFieldNameLabels) can settle, since a direct dependency puts them in a loop's path the same way fetchData was in #5587's.
Activity
huangyiirene commentedon Aug 22, 2026
Concentration-round first-touch (triage seat): promoted
finding→pm:queue, Task S. The precondition for the ref workaround is measured gone (#5564 / PR #5585), and the removal shape mirrors #5587 with the template test it names. The card's own two cautions are the acceptance bar: an effect-run-counting pin across re-renders (in and out of a provider), and the settle check on the two derived state maps — a loop here is a stop-and-report, not a force-through. Sequencing: after #5587 merges (its template test is the reference); not hard-blocked.Generated by Claude Code
claude commentedon Aug 23, 2026
Claim —
domain:uiexecution seat. PM round 30, sessionsession_01EuPCi56cnGyykygi3z9w4m, branchclaude/issue-5625-pivottable-i18n-ref-workaround. Default tier (Opus).Triage graded this Task S with the precondition already measured discharged (#5564 / PR #5585) and the removal shape mirroring #5587 plus the template test it names — read that landed diff rather than inferring the shape from this card.
Generated by Claude Code
os-zhuang commentedon Aug 23, 2026
{ "issue": 5625, "status": "done", "branch": "claude/issue-5625-pivottable-i18n-ref-workaround", "pr": "https://github.com/objectstack-ai/objectui/pull/5851", "premise_still_valid": true, "summary": "Removed both i18n-resolver refs (fieldLabelRef / fieldOptionLabelRef and the useEffect keeping them current) from packages/plugin-dashboard/src/ObjectPivotTable.tsx, reading fieldLabel / fieldOptionLabel directly in the metadata-derivation effect and declaring them in its dep array — the shape #5587 landed in ObjectChart (PR #5628), which I read rather than inferring from the card. Precondition re-verified on current main before removing anything: useObjectLabel pins both memo deps to module constants (NO_INSTANCE_T / NO_INSTANCE_I18N) while no i18next instance is bound, and useObjectLabel-identity-5564.test.tsx already pins that through useSafeFieldLabel specifically. The card's settle question is answered rather than assumed: the two derived state maps DO settle, because the effect now runs exactly once per genuine resolver change; the removal additionally restores re-derivation on a language switch, which the ref suppressed. eslint reports no react-hooks/exhaustive-deps finding on the new dep array, so nothing was suppressed to make it pass.", "tests": "All at b83dd427c (clean tree, git diff HEAD --stat empty), every heavy run through scripts/pm/os-verify-lock.sh, exit codes captured before any pipe. NEW PIN packages/plugin-dashboard/src/__tests__/ObjectPivotTable.i18nResolverDeps.test.tsx (3 cases: derivation count across mount + 3 forced re-renders with NO provider asserting RAW labels; same INSIDE a provider asserting localized labels; language-switch re-derivation) — 'Test Files 1 passed (1) / Tests 3 passed (3)'. PACKAGE SUITES `pnpm exec vitest run packages/plugin-dashboard/ packages/i18n/` from the repo root (never --filter, never a package cwd) — 'Test Files 129 passed (129) / Tests 1607 passed (1607)', lock VERDICT command-exit 0. TYPECHECK `pnpm --filter @object-ui/plugin-dashboard type-check` after `pnpm --workspace-concurrency=2 --filter '@object-ui/plugin-dashboard^...' build` — VERDICT command-exit 0; output echoed '> tsc --noEmit && tsc -p tsconfig.test.json', so it was not a zero-match silent no-op. LINT repo-wide `pnpm lint` (eslint . --no-inline-config) ran in full under the foreground cap — 0 errors repo-wide (warnings only, all pre-existing no-explicit-any); no narrowing to declare. GATES check-changeset-presence.mjs OK ('1 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'), check:control-bytes OK (scanned 4870 tracked text files), check:phantom-deps OK, check:self-import OK, check:i18n-keys OK — each quoted from its own printed verdict line, not from a bare $?. REVERSE VERIFICATION: two legs, both on the real commit, each driven by a script carrying trap '<restore>' EXIT INT TERM. NO REBUILD WAS NEEDED AND THAT IS ESTABLISHED, NOT ASSUMED: vitest.config.mts aliases @object-ui/i18n to packages/i18n/src (and @object-ui/plugin-dashboard to its src), so both mutations are live from source rather than through a dist the ablation could silently miss. Each mutation was confirmed ON DISK by grep, never by an editor exit code — the leg-B mutator was a python script that exits non-zero on anchor miss, and the counts were printed both ways. Leg A (restore the ref workaround in ObjectPivotTable.tsx via git checkout of the base rev; on disk: 'fieldLabelRef' 3 occurrences, direct-deps line 0) → RED, \"AssertionError: expected 'Stage (en)' to be 'Stage (ja)'\", 1 failed | 2 passed. Leg B (revert the #5585 memo-stability fix in useObjectLabel.ts; on disk: 'NO_INSTANCE_T;' 0 occurrences, 'const t = boundT;' 1) → RED, 'AssertionError: expected 258 to be 1' — 258 derivations instead of 1, i.e. the unbounded derive loop returns, 1 failed | 2 passed. Restored tree → GREEN 3 passed with git diff HEAD --stat empty. LEGS THAT DO NOT DISCRIMINATE, named: under leg A the two counting cases stay GREEN (the ref version also derives exactly once — that was the point of the ref), so only the language-switch case speaks for the source change; under leg B the two with-provider cases stay GREEN (that identity path was stable before #5585 too), so only the no-provider case speaks for it. Neither leg alone covers the change; the pair does.", "open_questions": [], "out_of_scope_findings": [] }Generated by Claude Code