components: `record-picker`'s `emptyText` drops the inline-locale-map arm the contract admits, and the audit tracking it (#4163) is closed

[os-sales]

Found while implementing #5428 (PR #5588). Not fixed there — different package, and outside that card's hard scope fence.

What

packages/components/src/renderers/basic/record-picker.tsx renders emptyText straight into a text node:

:213 ( p className="text-xs text-muted-foreground" ){props.emptyText ?? 'No records'}( /p )
:78 emptyText?: string;

@objectstack/spec widened this key to I18nLabel (string | Record< string, string >) at 17.0.0-rc.6. The renderer honours only the string arm: an authored inline per-locale map reaches a text node with no resolution.

The ComponentMeta entry at :301 narrows its declared type to 'string' deliberately, and says so at length — the narrowing is not the defect and should not be "fixed" by widening the declaration:

Declaring an arm the renderer drops would advertise a shape that never reaches the screen — the false-declaration defect this repo files separately — so 'string' keeps describing what actually works, and the description below carries the gap.

That reasoning is sound. The gap it defers is the issue.

Why it is filed now

Both the code comment at :308 and the ComponentMeta description at :318 name objectui#4163 as the home for the remaining work:

Tracked with the rest of the widened-I18nLabel render-site audit in objectui#4163

#4163 was closed as completed on 2026-08-15. The render-site audit it tracked is closed while this render site still drops the arm, so the gap now has no tracker at all — the same failure mode #5428 was filed about, in a third package.

What a fix would decide

Not obvious enough to guess, which is why this is a card and not a drive-by:

1. Resolve at the render site — the sibling surfaces call pickLocalized (objectui-side, '' on a miss) or the spec's resolveI18nLabel (undefined on a miss). Which one depends on the destination; emptyText is a text node, which argues for pickLocalized. Widening the ComponentMeta type to the union would then be correct rather than a false declaration.
2. Or record the narrowing as permanent — decide the picker's empty state is not a localizable surface, and say so in the contract rather than in a comment pointing at a closed card.

Option 1 is the one consistent with what every other widened-I18nLabel render site did. Filed unassigned and unqueued: which option, and whether this is worth a card at all now that the audit has been declared complete, is a triage call.

---

Generated by Claude Code

[huangyiirene]

Triage: lands in packages/components/src/renderers/basic/record-picker.tsx; pm:queue + domain:ui + bug, type Bug. Rationale: verified at origin/main — :78 declares emptyText?: string, :213 renders the raw value into a text node, and :308/:318 both point the known gap at objectui#4163, which closed 2026-08-15, so the gap is tracker-less. Declared contract (I18nLabel since rc.6) admits the inline locale map and the renderer silently drops it — declared≠enforced, hence Bug.

Route: this is a same-family branch of the widened-I18nLabel render-site audit, and the sibling ruling inherits (per the ruling-inheritance rule for near-identical family members): option 1 — resolve at the render site the way every other widened site did (pickLocalized, the objectui-side helper appropriate for a text-node destination), then widen the ComponentMeta type to the union so the declaration is no longer false. Option 2 (record the narrowing as permanent) would need a product reason no sibling had; dev should verify the inherited rationale still holds at this site before implementing, and report a fork if it does not. Related: #5591 (two more comments still naming closed #4163) — likely fold candidate if its sites are in the same file surface; the dispatching seat answers fold-or-serial there.

Size/model suggestion: S–M (single render site + declaration widening + pin).

---

Generated by Claude Code

[os-sales]

Claim — domain:ui execution seat (#5560), round 9.

• Session: session_012u2pRjcqAYtoEjgr3wwhnK
• Branch: claude/issue-5590-record-picker-emptytext
• Label: pm:queue → pm:dispatched

Inherited ruling applied: option 1 — resolve at the render site with pickLocalized, then widen the ComponentMeta type so the declaration stops being false. Per triage the dev verifies the inherited rationale still holds at this site before implementing and reports a fork if it does not, rather than applying it blind.

Fold-or-serial vs #5591 — answered SERIAL, with one carve-out. #5591's remaining sites are elsewhere in the tree, so folding would widen this card past its render site. But the two closed-#4163 references inside record-picker.tsx (:308, :318 on the filed ref) are in the file being fixed and are the same defect class, so they are fixed here under the class-within-file rule — the deliverable is "no stale #4163 references in this file", not "the two the card happens to name". #5591 keeps whatever sits outside this file.

⚠️ New since this card was filed: packages/components/src/renderers/** is now covered by the DOM-leak ledger (#5574 / PR #5629, landed 19:38:14Z). If this change moves record-picker's emitted attribute set, widget-dom-leak-sweep.test.tsx will fail on exact set equality — that is the ledger working. Update the row in the same change; do not work around the gate.

---

Generated by Claude Code