refactor(plugin-dashboard): retire the written-never-read FieldMeta.decimals

.changeset/6625-retire-fieldmeta-decimals.md

@@ -0,0 +1,41 @@
+---
+'@object-ui/plugin-dashboard': minor
+---
+
+Retire `FieldMeta.decimals` and the now-unfed `BuildFieldMetaParams.overrides.decimals`
+from `plugin-dashboard`'s shared field-rendering helpers (objectui#6625,
+enforce-or-remove).
+
+`buildFieldMeta` computed `decimals: overrides.decimals ?? meta?.decimals ?? meta?.scale`
+on **every** call and the value **reached nothing**. Re-measured on this branch's base
+(`efdc6c62`): **zero `.decimals` member reads** across `@object-ui/fields`,
+`@object-ui/i18n`, `@object-ui/components`, `@object-ui/core` and `plugin-dashboard`
+itself — the only non-comment occurrence was the write being removed here. The
+**positive control in the same query shape** fires: `.scale` member reads hit
+`NumberField.tsx`, `GridField.tsx` and `fields/src/index.tsx`. So the zero is a
+finding, not a broken query. The `overrides.decimals ??` head of that chain had already
+lost its only feeder when objectui#6425's ruling removed the authored read from
+`ObjectDataTable.enrich()`; `RecordDetailDrawer`, the only other `buildFieldMeta`
+caller, passes no overrides at all. Both halves retire together, so the key leaves in
+one move.
+
+Behaviour is unchanged — no reader existed to notice, pinned by the unchanged runtime
+assertion in `ObjectDataTable.overrideSource-6425.test.tsx` that an authored `decimals`
+renders byte-identical to its absence.
+
+**The refusal did NOT leave with the member.** `ObjectDataTable` derives *two* refusal
+bands from `keyof FieldMeta` — `EnrichedColumn`'s write-side tombstones (objectui#6373)
+and `AuthoredColumnOverrides`' read-side band (objectui#6425) — so deleting the member
+would have dropped `decimals` from both as a side effect, silently un-enforcing
+objectui#6425's retire. A new hand-written `ObjectDataTableRetiredDecimalsTombstone`
+(`{ decimals?: never }`) is intersected into both halves of the seam, the same shape and
+for the same reason as `ObjectGrid`'s `ObjectGridRetiredOptionsTombstone`. The verdict
+is unchanged since 2026-08-27; only the artefact enforcing it moved, from derived to
+hand-written.
+
+Marked `minor` per this repo's version-alignment rule (AGENTS.md 版本号策略), which
+reserves `major` for following `@objectstack` across a major. Scope note, measured rather
+than assumed: `FieldMeta`, `BuildFieldMetaParams`, `AuthoredColumnOverrides` and
+`EnrichedColumn` are **absent from `dist/index.d.ts`** — they are not re-exported by the
+package barrel, and the `exports` map publishes only `"."`, so no consumer can name them.
+No downstream type moves; this is a package-internal contract change.

packages/plugin-dashboard/src/ObjectDataTable.tsx

@@ -131,14 +131,52 @@ interface NormalizedColumn {
  * seam instead of arriving anonymously inside a spread. When #5120 retires the
  * consumer alias, this member becomes a tombstone with it.
  */
+/**
+ * ⭐ `decimals` — RETIRED from `FieldMeta` ITSELF (objectui#6625), and this
+ * hand-written tombstone is now the ONLY enforcement of that key's refusal at
+ * BOTH halves of this seam. ⛔ Do not "tidy" it away as redundant.
+ *
+ * Until objectui#6625 the key needed no tombstone: it was a `FieldMeta` member,
+ * so BOTH bands below — {@link EnrichedColumn}'s write-side tombstones
+ * (objectui#6373) and {@link UnheldFieldMetaOverrideKey}'s read-side refusal
+ * (objectui#6425) — derived it for free. Refusal by MEMBERSHIP. #6625 then
+ * retired the member (written from the schema def, read by nothing), which
+ * removed it from `keyof FieldMeta` and silently ended that enforcement at both
+ * ends. Nothing would have gone red at the moment of loss: on the read side the
+ * suite's `@ts-expect-error` would merely have turned TS2578-unused, and on the
+ * write side the hand-written pin would have gone on passing on the
+ * excess-property check alone — a pin passing because its subject stopped
+ * existing.
+ *
+ * This is the exact mirror of the rule `ObjectGrid`'s sibling tombstone
+ * records (`ObjectGridRetiredOptionsTombstone`, objectui#6425): *a pin enforced
+ * by a key's non-membership silently stops enforcing the moment the key becomes
+ * a member.* Read from this end: **a refusal DERIVED from a key's membership
+ * silently stops enforcing the moment that member is deleted.** Same seam, same
+ * blindness, opposite direction — which is why the retirement had to be
+ * re-stated by hand rather than inherited.
+ *
+ * objectui#6425's verdict is unchanged — no reader for an authored `decimals`
+ * existed then and none exists now — and so is the runtime behaviour: the value
+ * the retired `buildFieldMeta` write resolved reached nothing. Only the
+ * refusal's mechanism moves, from derived to hand-written. It is intersected
+ * into BOTH types below, because the retirement belongs to the SEAM, not to one
+ * of its two halves — the same reason `ObjectGrid` intersects its own tombstone
+ * into both its draft and its post-fold column. ⭐ A future reader for decimal
+ * places reads `scale`, never this key (objectui#6625; `NumberCellRenderer`
+ * already does).
+ */
+type ObjectDataTableRetiredDecimalsTombstone = { decimals?: never };
+
 export type EnrichedColumn =
  TableColumn
  /** HELD alias, objectui#5120 — see above. Not declared by `TableColumn`. */
  & { name?: string }
  /** RETIRED at this emit seam, objectui#6373 — derived, never hand-listed, so
  * a future `FieldMeta` member is tombstoned by default and has to be
  * adjudicated to escape. */
- & { [K in Exclude<keyof FieldMeta, keyof TableColumn | 'name'>]?: never };
+ & { [K in Exclude<keyof FieldMeta, keyof TableColumn | 'name'>]?: never }
+ & ObjectDataTableRetiredDecimalsTombstone;
 
 /**
  * What this widget's column producer is allowed to READ off the AUTHORED
@@ -167,8 +205,10 @@ export type EnrichedColumn =
  * - `currency` — DECLARED (kept in production, never promised before;
  * declaring makes the existing behaviour honest). Same `Pick`.
  * - `decimals` — RETIRED, immediately (neither promised nor kept: zero
- * readers measured, and the authored read below is gone). The derived
- * band now refuses it like any other unadjudicated `FieldMeta` member.
+ * readers measured, and the authored read below is gone). Refused by
+ * {@link ObjectDataTableRetiredDecimalsTombstone} since objectui#6625
+ * retired the `FieldMeta` member itself and took the key out of the derived
+ * band's pool; the verdict is unchanged, only its mechanism moved.
  * - `referenceTo` — ⛔ NOT declared as spelled; still HELD, owned by
  * objectui#6597 (the enforce-or-remove channel: fix the spelling chain so
  * the promise becomes real, or withdraw the README line).
@@ -182,8 +222,12 @@ export type EnrichedColumn =
  * `FieldMeta` is the override vocabulary — `buildFieldMeta`'s `overrides` is a
  * subset of it — so it is the pool a new tolerance would come from, and the
  * same pool {@link EnrichedColumn}'s tombstones derive from. Deriving means a
- * seventh `FieldMeta` member is refused here on the day it is added, without
- * anyone remembering to extend a hand-written list.
+ * NEW `FieldMeta` member is refused here on the day it is added, without
+ * anyone remembering to extend a hand-written list. (Stated without a count on
+ * purpose: it read "a seventh" while the type had eight members, and
+ * objectui#6625 has since retired one. The property is that ADDING is covered
+ * by derivation — ⚠️ REMOVING is not, which is why that card had to leave
+ * {@link ObjectDataTableRetiredDecimalsTombstone} behind by hand.)
  *
  * ⚠️ It derives from the OVERRIDE VOCABULARY, not from the authored input
  * type, and that difference is forced rather than stylistic. `plugin-grid`'s
@@ -230,7 +274,9 @@ export type EnrichedColumn =
  * left: objectui#6425's ruling (maintainer, 2026-08-27) declared `format` /
  * `options` / `currency` on `TableColumn` itself (they are read via
  * `Pick<TableColumn, …>` below, no hold needed) and retired `decimals`
- * outright (the derived band refuses it now).
+ * outright — refused by {@link ObjectDataTableRetiredDecimalsTombstone} since
+ * objectui#6625 retired the `FieldMeta` member that used to carry it into the
+ * derived band.
  */
 export interface ObjectDataTableColumnHolds {
  /**
@@ -253,9 +299,20 @@ export interface ObjectDataTableColumnHolds {
  * {@link ObjectDataTableColumnHolds} to escape. Keys `TableColumn` declares
  * leave the pool by declaration: `type` (objectui#5853 owns its VALUE set,
  * folded below by `normalizeTableColumnType`) and, since objectui#6425's
- * ruling, `format` / `options` / `currency`. `decimals` is the member the
- * same ruling RETIRED — it lands here, refused at the read site, which is
- * exactly the "held-band verdict flips to RETIRED" the ruling asked for.
+ * ruling, `format` / `options` / `currency`.
+ *
+ * ⚠️ `decimals` USED TO land here — that is how objectui#6425's ruling was
+ * enforced at the read site. It no longer can: objectui#6625 retired the
+ * `FieldMeta` member itself, so the key is out of this Exclude's POOL rather
+ * than out of its exclusion list, and a derived band cannot refuse a key that
+ * is not in what it derives from. The refusal is carried by
+ * {@link ObjectDataTableRetiredDecimalsTombstone} instead, intersected below.
+ * ⛔ Do not read this band's silence about `decimals` as a softening — the two
+ * artefacts together are the same verdict, unchanged since 2026-08-27.
+ *
+ * The pool is what shrank, so what THIS band still refuses is `name` and
+ * `label` — both of them `FieldMeta` members with answers this seam already
+ * has (see the docblock above).
  */
 export type UnheldFieldMetaOverrideKey =
  Exclude<keyof FieldMeta, keyof TableColumn | keyof ObjectDataTableColumnHolds>;
@@ -271,7 +328,10 @@ export type AuthoredColumnOverrides =
  * their published types, straight off the declaration. */
  & Pick<TableColumn, 'format' | 'options' | 'currency'>
  & ObjectDataTableColumnHolds
- & { [K in UnheldFieldMetaOverrideKey]?: never };
+ & { [K in UnheldFieldMetaOverrideKey]?: never }
+ /** RETIRED, objectui#6425's verdict — re-stated by hand because objectui#6625
+ * took the key out of the derived band's pool. See the tombstone's docblock. */
+ & ObjectDataTableRetiredDecimalsTombstone;
 
 /**
  * Shared empty fallback for the resolved row list (objectui#4629).

packages/plugin-dashboard/src/__tests__/ObjectDataTable.emitBoundary-6373.test.tsx

@@ -210,10 +210,14 @@ describe('ObjectDataTable emits only what the columns slot declares (#6373)', ()
  it('names the six keys that retired, and keeps the one that is held', async () => {
  const cols = await emit({ type: 'object-data-table', objectName: 'account' });
  for (const col of cols) {
- // `Object.keys`, not a value read: `buildFieldMeta` always returns all
- // eight members, so before this card every one of these keys EXISTED on
- // every emitted column — carrying `undefined` where the schema said
- // nothing, which is its own small lie about the shape.
+ // `Object.keys`, not a value read: `buildFieldMeta` returns all of its
+ // members unconditionally, so before this card every one of these keys
+ // EXISTED on every emitted column — carrying `undefined` where the schema
+ // said nothing, which is its own small lie about the shape. (It returned
+ // eight then and returns seven now: objectui#6625 retired `decimals` from
+ // `FieldMeta` itself, so that key can no longer be written from here even
+ // by accident. The verdict below is unchanged — it was already retired
+ // from the EMIT by this card.)
  for (const key of RETIRED) {
  expect(Object.keys(col), `${col.accessorKey}.${key}`).not.toContain(key);
  }
@@ -296,12 +300,15 @@ describe("the emit type can FAIL — otherwise the annotation is decoration (#63
  // — so on its own it pins "the spread is refused" without pinning WHY, and
  // would have gone on passing after the enforcement was removed.
  //
- // `Omit<FieldMeta, 'name' | 'type'>` spans the retired six: `name` is
+ // `Omit<FieldMeta, 'name' | 'type'>` spans the retired members: `name` is
  // the held alias, `type` carries #5853's own refusal. Since the #6425
- // ruling declared `format` / `options` / `currency` on `TableColumn`,
- // the members still refused by tombstones are `label`, `referenceTo` and
- // `decimals` — enough to keep this spread an error, and nothing but the
- // tombstones refuses it.
+ // ruling declared `format` / `options` / `currency` on `TableColumn`, the
+ // members still refused by the DERIVED tombstones are `label` and
+ // `referenceTo` — enough to keep this spread an error, and nothing but the
+ // tombstones refuses it. (`decimals` used to be a third; objectui#6625
+ // retired the `FieldMeta` member, so it is no longer spanned by this
+ // `Omit` at all. Its own refusal is pinned separately below, because a
+ // hand-written tombstone is now what carries it.)
  // @ts-expect-error objectui#6373 — the still-tombstoned members are refused by the tombstones alone.
  const retiredRefused: EnrichedColumn = { header: 'h', accessorKey: 'a', ...({} as Omit<FieldMeta, 'name' | 'type'>) };
  expect(retiredRefused.accessorKey).toBe('a');
@@ -313,9 +320,36 @@ describe("the emit type can FAIL — otherwise the annotation is decoration (#63
  // by different machinery. This pin carried `format` until the #6425
  // ruling declared it (writing it by hand is now legal, see the accepted
  // fixture above); `decimals` — the key the same ruling RETIRED — takes
- // its place, refused by the derived tombstone.
- // @ts-expect-error objectui#6373/#6425 — `decimals` retired from this emit seam.
- const writtenRefused: EnrichedColumn = { header: 'h', accessorKey: 'a', decimals: 2 };
+ // its place.
+ //
+ // ⚠️ The source is a VARIABLE, not a fresh object literal, and that is
+ // objectui#6625's doing rather than style. A fresh literal is refused by
+ // the excess-property check whether or not any tombstone exists, so as a
+ // literal this pinned "something refused this" without pinning what — and
+ // it would have gone on passing after #6625 removed `decimals` from
+ // `FieldMeta` and thus from the DERIVED tombstone band, which is a pin
+ // passing because its subject stopped existing. A non-fresh source skips
+ // the freshness check and reaches
+ // `ObjectDataTableRetiredDecimalsTombstone` and nothing else. The control
+ // below proves that is what answers.
+ const carriesDecimals: { header: string; accessorKey: string; decimals?: number } =
+ { header: 'h', accessorKey: 'a', decimals: 2 };
+ // @ts-expect-error objectui#6373/#6425/#6625 — `decimals` refused by the explicit retired-key tombstone.
+ const writtenRefused: EnrichedColumn = carriesDecimals;
  expect(writtenRefused.accessorKey).toBe('a');
  });
+
+ it('the TOMBSTONE is what refuses the retired `decimals` at this emit', () => {
+ // ⭐ objectui#6625's counter-control for the pin above: `EnrichedColumn`
+ // minus the retired-key tombstone and nothing else ACCEPTS the very source
+ // the directive refuses. So the refusal is the tombstone's — not the
+ // derived band's (which no longer reaches `decimals`, since the key left
+ // `keyof FieldMeta`), not the excess-property check's (the source is not
+ // fresh), and not weak-type detection's (`header` / `accessorKey` are in
+ // common, and both are required here).
+ const carriesDecimals: { header: string; accessorKey: string; decimals?: number } =
+ { header: 'h', accessorKey: 'a', decimals: 2 };
+ const untombstoned: Omit<EnrichedColumn, 'decimals'> = carriesDecimals;
+ expect(untombstoned.accessorKey).toBe('a');
+ });
 });