[plugin-gantt] tooltip number / currency / percent rows render in the MACHINE locale, next to date rows that follow the display locale

[yinlianghui]

Found while adding tenantCurrency to the tasks memo's dependency array for #4542. Filed rather than fixed — outside that card's ruled surface (which is the dependency array only).

The gap

packages/plugin-gantt/src/ObjectGantt.tsx, inside formatFieldValue (the tooltip value formatter in the tasks memo). #4272 / PR #4544 threaded useDisplayLocale() into this function's four TEMPORAL call sites. The three NUMERIC ones next to them were left as they were:

case 'number':
case 'integer':
case 'float':
case 'decimal':
  return formatNumber(Number(value));
case 'currency':
  return formatCurrency(Number(value), resolveFieldCurrency(def as any, tenantCurrency));
case 'percent':
  return formatPercent(Number(value));

All three formatters accept a locale that nothing passes here:

• formatNumber(value, decimals?, locale?) — third parameter omitted
• formatCurrency(value, currency?, locale?) — third parameter omitted (the currency CODE is resolved correctly; the locale that renders it is not)
• formatPercent — same shape

Each ends up at formatDisplayNumber with locale: undefined, which reaches new Intl.NumberFormat(undefined, ...). Per useDisplayLocale's own doc that is the MACHINE's locale — neither of the repo's two locale channels.

Why it can be observed

One tooltip, two conventions. On a zh or de session a gantt tooltip whose fields include a date and an amount renders the date row through the display locale (correct since PR #4544) and the amount row through whatever locale the browser happens to run in. A de user reading a tooltip sees 5. Jan. 2024 on one line and 1,234.50 on the next, where the German convention is 1.234,50 — the separators are inverted, so the amount is not merely unstyled but readable as a different number.

Measured on this runner (machine locale en-US), for 1234.5:

• de expected 1.234,50, rendered 1,234.50
• zh expected 1,234.50 (agrees with en-US here, so zh alone does not discriminate)

Fix sketch

Thread the displayLocale already read at component level (L435, added by PR #4544) into all three numeric call sites, exactly as the four temporal ones were threaded. The dependency array already carries displayLocale, so no memo change is needed. Red-first case: a de session asserting the tooltip amount uses . for grouping and , for the decimal mark, with an en byte-identical pin beside it.

Relations

• Same family as Console: number cell/display renderers hardcode a grouping Intl.NumberFormat('en-US') — an ordinal Field.number (a year) always renders 2,026, and no field property can turn it off #4033 (the fields-side origin) and Date formatter is half-localized: future relative forms and absolute timestamps stay en-US on a non-English locale #4272 / PR fix(fields): the date formatter's last three en-US channels thread the display locale (#4272) #4544 (the temporal half of this very function).
• [plugin-grid] ObjectGrid's record-detail date fallback renders the MACHINE locale #4541 is the plugin-grid twin of the same class, at a different site; this is not a duplicate of it.
• Not a duplicate of [plugin-gantt] the tasks useMemo omits tenantCurrency, so tooltip currency can render stale #4542 either: that card is the memo's dependency array (WHEN the string is rebuilt), this one is the formatting call itself (WHICH locale it is built in). [plugin-gantt] the tasks useMemo omits tenantCurrency, so tooltip currency can render stale #4542 shipping does not move this.

---

Generated by Claude Code

[yinlianghui]

CLAIM — session_017Qqyix2QcnpUC9XeYVDzx3, branch claude/issue-4553-gantt-tooltip-number-locale. Dispatching a dev agent now (#4554 has landed — no gate).

PM ruling (delegated decision authority; maintainer veto window open — record objections here): the numeric completion of the gantt tooltip's one-channel work (#4544 dates, #4554 currency dep):

1. Thread displayLocale into the three numeric formatter calls at the tooltip sites (formatNumber / formatCurrency / formatPercent, ObjectGantt.tsx ~:663/:665/:667) — measure each formatter's existing options signature in packages/fields first and use the established parameter shape; fields src is read-only (if any of the three lacks a locale parameter, that is a premise inversion — STOP and report, do not edit fields under this card).
2. Red-first on a de session (the card's own repro: 1,234.50 vs 1.234,50 — inverted separators are the un-fakeable signal); en pinned byte-identical; the fix(fields): the date formatter's last three en-US channels thread the display locale (#4272) #4544/fix(plugin-gantt): tooltip currency re-formats when the tenant currency resolves (#4542) #4554 pins (dates, currency dep, locale dep) stay green.
3. Changeset: '@object-ui/plugin-gantt': patch (consumer threading, module-local). Never major.

Mutual exclusion: plugin-gantt free (#4554 landed). Do not touch packages/fields, plugin-grid (#4541 in flight), studio/app-shell (#4527-p2 in flight), plugin-kanban (#4254 in flight).

---

Generated by Claude Code

---

Generated by Claude Code

[yinlianghui]

Signature measurement (ruling item 1) — the premise held for two of the three formatters, not three. Draft PR #4557 threads those two; the third is escalated unstarted rather than guessed at.

formatter (packages/fields/src/index.tsx)	signature	locale?
formatNumber L452	(value, decimals = 2, locale?)	yes, param 3
formatCurrency L371	(value, currency?, locale?)	yes, param 3
formatPercent L473	(value, precision = 0)	no

The card states all three reach formatDisplayNumber with locale: undefined. formatPercent does not — its entire body is:

const displayValue = percentDisplayValue(value);
return `${displayValue.toFixed(precision)}%`;

It constructs no Intl.NumberFormat and never reaches formatDisplayNumber. So its output is not the machine's locale — it is no locale: toFixed emits an ASCII decimal mark and never a grouping separator, identically on every machine. Measured: 1235% on en, de and zh alike, where German wants 1.235 %.

That makes the percent row a different defect class from the other two, and one that threading cannot reach: closing it requires a new parameter on a @object-ui/fields export, which this card's surface holds read-only. Per the ruling I stopped there rather than editing fields.

Two consequences worth a decision:

1. It is not a call-site fix, so it does not belong to this card as scoped. formatPercent is shared with the list cell renderer and the dashboard measure formatter through percentDisplayValue, so a locale-aware reimplementation inside plugin-gantt would fork percent formatting away from both — the reason I did not patch around it locally.
2. Whoever picks it up inherits a second question the separator issue hides: at precision = 0 a percent is also ungrouped in every locale, so 1235% is wrong in en-US too, not only in de. Grouping and locale would land together.

The current percent behaviour is pinned by a labelled test in #4557 (ObjectGantt.numberLocale.test.tsx), marked as evidence of the inversion rather than an endorsement, so the gap cannot drift unnoticed and goes red the day formatPercent grows a locale.

PR #4557 is therefore Part of #4553, not Fixes — the card should stay open for the percent half. Its title also drops "percent" from the ruled wording, since the PR does not make percent follow the display locale and the title would otherwise assert something the measurement disproved.

---

Generated by Claude Code

---

Generated by Claude Code

[yinlianghui]

PM percent ruling after PR #4557's measured escalation (session_017Qqyix2QcnpUC9XeYVDzx3; delegated decision authority, maintainer veto window open):

Option A — extend formatPercent at the producer, grouping and locale together. The dev's three-axis recommendation is adopted in full, and option B is rejected on its own reasoning: a plugin-local percent formatter forks percent away from the two surfaces sharing percentDisplayValue (the list cell renderer and the dashboard measure formatter), recreating the drift that helper's doc comment exists to prevent.

Binding details for phase 2 (same dev continues, same card):

1. formatPercent in packages/fields gains an optional locale and routes through formatDisplayNumber — the exact formatDateTime playbook from Date formatter is half-localized: future relative forms and absolute timestamps stay en-US on a non-English locale #4272/PR fix(fields): the date formatter's last three en-US channels thread the display locale (#4272) #4544: the parameter lands WITH its consumers, never speculatively. Grouping lands in the same change (the 1235%-ungrouped-everywhere finding means en-US output MOVES — that is the fix, not a regression; pin the new grouped en form explicitly and note the old form in the changeset).
2. Consumers threaded in the same PR: the gantt percent tooltip row (flipping fix(plugin-gantt): tooltip numbers and currency follow the display locale (#4553) #4557's inversion-evidence pin, an authorized pin move to be declared), plus the other formatPercent consumers as measured (the list cell renderer inside fields, the dashboard measure formatter — census them; each pinned).
3. Red-first: de percent (12,5 %-family form per Intl) and the en grouping change are both genuine reds; predictions in writing; the fix(plugin-gantt): tooltip numbers and currency follow the display locale (#4553) #4557 pins for number/currency stay green.
4. Changesets: '@object-ui/fields' MINOR (entry-visible optional param + en output move — the fix(plugin-timeline): dates follow the active locale instead of hardcoded en-US (#4513) #4523/fix(fields): the date formatter's last three en-US channels thread the display locale (#4272) #4544 precedent line, quoted); consumers per their own .d.ts diffs. Never major.
5. Mutual exclusion: packages/fields' formatPercent region is now THIS card's surface. ⛔ finding(react): SchemaRenderer carries the #4422 prop erasure in a spelling every sweep and both guards are blind to — Record< string, any > instead of [key: string]: any #4548 (in flight, packages/react + repo-wide canary) — if its canary and your work collide on a file, coordinate by STOP-and-report rather than racing. [a11y] The RLS USING/CHECK predicate editors render a label bound to nothing — CelPredicateField's id is optional and three call sites omit it #4533 (CelPredicateField), Deleting a draft-only saved view no-ops — deleteView has no draft qualifier, and the mirror of #4139 is NOT mechanical (it would break published+draft pairs) #4479 (data-objectstack) disjoint.

PR #4557 (numbers + currency) is accepted and armed; this card closes when phase 2 lands.

---

Generated by Claude Code

---

Generated by Claude Code