fix(plugin-grid): inline cells read the record-grained update verdict the row kebab reads - #27384

Requested by Warren · project thread
Fixes #27365

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:列表行内编辑现在按「这一行」的服务端写权限判定;服务端拒绝的行点不开单元格编辑器,和该行的 ⋮ 菜单给出同一个答案。
• 为什么改:对象级有编辑权、但共享/归属/RLS 拒绝某行时,⋮ 菜单已隐藏「编辑」,单元格却仍可编辑,保存时被服务端整单拒绝。
• 风险与回滚:只收窄不放宽;没拿到行级答案(未返回、请求失败、行无 id)时保持原行为。回滚即 revert 本 PR。
• 先看:ui/packages/plugin-grid/src/ObjectGrid.tsx(isRowEditable 与 update: canEdit || inlineEditable)、ui/packages/components/src/renderers/complex/data-table.tsx(rowMayEdit)、ui/packages/types/src/data-display.ts(新运行时槽位)。
• 两分钟验证:pnpm --dir ui exec vitest run packages/plugin-grid/src/__tests__/inlineEditRecordVerdict-27365.test.tsx,预期 8 个用例全绿。
• 需要你做的:无,席位落地。

What changes

An inline-editable grid gated its cells on the object-level affordance only (inlineEditable), while the row kebab on the same rows also read the server's record-grained update verdict (useRecordCrudVerdicts). A row refused by sharing, ownership or RLS lost Edit from its kebab but still opened a cell editor, and the save was refused whole. Now the cells read the same verdict.

• @object-ui/types: DataTableSchema declares a new runtime slot, isRowEditable?: (row: any) => boolean. The zod mirror refuses it by name (handlerKeyRefusal, runtime-slot), like renderCellEditor and rowGroups. The ledgers move with it: the KnownDrift entry for data-display.zod.ts#DataTableSchema gains the key (51 entries, 88 keys), and the read lists in the body/children tombstones name it.
• @object-ui/components: data-table reads isRowEditable through one rowMayEdit(row) check in startEdit (the door click, double-click and Enter all pass through), in the Enter key path, and in the per-cell isEditable. A refused row reads like a row of a read-only table: no editor, no edit cursor, and a click reaches the row's own click handler. With the slot absent, every row behaves as before.
• @object-ui/plugin-grid: ObjectGrid supplies isRowEditable as resolveRowRecordCrudAffordance(inlineEditable, recordVerdict(rowRecordId(row), 'update')), the exact accessor the kebab reads. It asks the batched probe whenever the grid is inline-editable (update: canEdit || inlineEditable), so an editable grid whose kebab offers no Edit is narrowed too. That is still one probe, read by both affordances, with no second evaluator (ruling 6457467702 B).
• ListView reaches the cells through ObjectGrid (type: 'object-grid', editable: inlineEdit && inlineEditOffered), so it has no separate door and needs no plugin-list change.
• Test doubles: editable-grid suites now get the explain probe. Nine plugin-grid suites and one app-shell suite (ListView.gridRowColorSelect-12393) answer it from an explain double instead of reaching the network. This is test-only, per claim amendment 6487558539. The assertions are unchanged.

Why

The card: one grid gave two answers for one row. Fail-open is preserved exactly as the useRecordCrudVerdicts docblock requires: an unknown verdict keeps the object-level behaviour.

Tests

• plugin-grid/src/__tests__/inlineEditRecordVerdict-27365.test.tsx (8): a refused row opens no editor on click or Enter and shows no edit cursor. CONTROLs: an allowed row stays editable; a non-OK response, a network failure, a response with no records[], and the pre-answer window all keep the row editable. The kebab and the cells agree on both rows from ONE update probe. An editable grid with no kebab Edit still asks and still narrows.
• components/src/__tests__/data-table-row-editable-27365.test.tsx (7): the seam on its own covers click, double-click and Enter, a click that reaches onRowClick, an allowed-row control, the no-predicate control, and narrows-only.
• types/src/__tests__/data-table-declared-keys-6882.test.ts: the declared exact shape, a widened-return control (@ts-expect-error), and the zod mirror refusing an authored isRowEditable by name.

Acceptance notes

• Ablation 1, ObjectGrid's per-row read replaced by (_row: any) => true via scripts/ablation-replace.mjs: 4 of 8 tests in the new grid suite red (the repro, agreement, no-kebab and pre-answer legs). Restored blob equals HEAD and git diff HEAD is empty.
• Ablation 2, data-table's rowMayEdit replaced by () => true: 3 of 7 tests in the seam suite red (click, double-click, Enter). Restored blob equals HEAD.
• ui/apps/site was retired on main. After merging main, an ignored leftover directory (node_modules) in the worktree reddened quick-reference-commands-4149 locally. That was a local-state artifact only: the test is green once the leftover is removed, and nothing in the tree changed.

Verification

All readings were taken on head 3b46f5a1 (origin/main c4c515ad merged in), with the @object-ui/console^... closure built under the lock.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic.
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint was not run locally (CI Lint UI and Lint & Repo Gates own it).
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset, and both declaration lines are filled in here and in the changeset.
• 6. packages/spec touched: n/a, not touched.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English (the 维护者速读 block is the dispatch's required Chinese summary).
• The Before Submitting list in CONTRIBUTING.md is done.

🤖 Generated with Claude Code

Generated by Claude Code