[objectstack-fleet]

Requested by Steedos · project thread

Fixes #26738

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:Studio 列表视图检查器(以及标签页预设)里的筛选器控件,现在自己保存「还没填完」的条件行。点「Add filter」会出现一行可编辑的条件;填完之后才把这条规则写进视图。
• 为什么改:控件以前每次渲染都从已保存的规则重新生成行,而保存端的折叠(foldFilterGroupToSpecRules)会正确地丢掉未填完的行,于是新加的空行在下一次渲染就消失了,Studio 里根本无法给列表视图加筛选。
• 风险与回滚:低。只改 app-shell 一个控件的状态持有方式;保存规则不变(未填完的条件仍然不会写进视图),不新增任何元数据键,spec 不动。回滚即 revert 本 PR。
• 先看:ui/packages/app-shell/src/views/metadata-admin/widgets.tsx 里 FilterBuilderField 的 draft / seededFrom 两个状态和 handle 里新增的「与已存规则相同则不写」判断;新测试 widgets.filterBuilderHeldRow-26738.test.tsx。
• 两分钟验证:cd ui && pnpm exec vitest run packages/app-shell/src/views/metadata-admin/widgets.filterBuilderHeldRow-26738.test.tsx,预期 4 个用例全绿;或在 Studio Interfaces 打开某对象的列表视图检查器,Filter →「Add filter」,出现一行,填入值后视图草稿里出现该 filter。
• 需要你做的:无,席位落地。

What changes

FilterBuilderField (the filter-builder widget the Studio list-view inspector's Filter row uses, and the per-tab filter of filter-mode tab presets) built the FilterBuilder's row group from the stored rules on every render. The builder's "Add filter" inserts a blank row; the save-side fold drops a row that is not complete yet, as it must; the host stored the unchanged rules and re-rendered, and the re-derived group replaced the builder's rows, so the new row disappeared before it could be edited.

The widget now holds the builder's row group in its own state:

• every builder change is folded; a refused fold (OR over several rows, a nested group) still toasts and writes nothing;
• an accepted fold updates the held rows, and writes only when the folded rules differ from what is stored. An incomplete row therefore stays on screen and writes nothing; completing it writes exactly the complete rule;
• when the stored rules change and are not the echo of this builder's own write (another editor, a reload), the held rows are re-seeded from what is stored. The comparison keys on the rules' JSON, never on an array's identity.

viewFilterFold.ts is unchanged: what is persisted is decided by the same fold as before.

Tests

New widgets.filterBuilderHeldRow-26738.test.tsx mounts the real filter-builder widget in a host that writes the way ObjectListViewInspector does (a new body object on every onChange, filter dropped when empty):

• "Add filter" adds an editable row, writes nothing while it is incomplete, and completing it writes [{ field: 'title', operator: 'equals', value: 'open' }] with the row still on screen;
• a second, incomplete row is held beside a stored rule without a write;
• a stored filter shows as its row and summary, opening it writes nothing, and editing the value away and back writes the stored rule unchanged (round-trip);
• CONTROL: a stored filter changed from outside re-seeds the rows.

Reverse verification (one-off, not kept): with the fix committed, widgets.tsx restored to base c8583f80 in the working tree only (groupOfStoredRules occurrences 3 → 0) made the new file red, Tests 3 failed | 1 passed (4); each red is the vanished row ("expected [] to have a length of 1 but got +0"). The round-trip case is green on base too, as a preservation pin should be. Restored with git checkout HEAD -- … (occurrences back to 3, git diff HEAD empty).

Acceptance notes

• The card suspected a regression from the filter-fold work of components: the filter-builder renderer hands an authored FilterGroup to a builder typed on FilterBuilderRowGroup, and the two row shapes are assignable in neither direction (hidden by any until #26293) #26306 / components / plugin-list: the filter builder emits [true] for a boolean row switched to In, which the spec's ViewFilterRule refuses — the builder's own (string | number | boolean)[] list member outlives #26306 #26348. Those changed the builder's value types, not how this widget feeds the builder; the re-derived group predates them in the code read here. History was not walked (shallow checkout), so when it started is not measured.
• The inspector's write callback rebuilds the view body on every onChange, so before this change an incomplete row would also have written an unchanged body back to the draft. The widget now writes only when the folded rules change.
• The popover's held rows live as long as the widget is mounted, so closing and reopening the popover keeps an incomplete row ("held until complete").

Verification

All readings at head 5221c34d (base c8583f80), under scripts/pm/os-verify-lock.sh.

Touched package: @object-ui/app-shell

• Dependency closure: pnpm exec turbo run build --concurrency=2 --filter='@object-ui/app-shell^...', 63 tasks, VERDICT command-exit 0.
• pnpm --filter @object-ui/app-shell run type-check (tsc --noEmit, then tsc -p tsconfig.test.json): VERDICT command-exit 0, zero TS errors.
• New test file: Test Files 1 passed (1), Tests 4 passed (4).
• Narrowed app-shell run: every app-shell test file naming the metadata-admin widgets module, filter-builder-trigger, FilterBuilderWidget, viewFilterFold, ObjectListViewInspector, FilterModeWidget or filter-mode (52 files): Test Files 52 passed (52), Tests 754 passed (754). The rest of app-shell is left to CI's Test UI shards.

Derived gates

node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 34 commands at 5221c34d; each ran with its exit code captured before any pipe, plus pnpm check:changeset-fixed (its roster sits under .changeset/). --ran reconciliation: "34 derived famil(ies) accounted for — 34 run, 0 NOT-MEASURED".

• node scripts/check-adr-0087-registration.mjs --base origin/main → exit 0
• node scripts/check-adr-0087-registration.mjs --self-test → exit 0
• node scripts/check-changeset-no-major.mjs --base origin/main → exit 0
• node scripts/check-closing-keyword-parity.mjs → exit 0
• node scripts/check-closing-keyword-parity.mjs --self-test → exit 0
• node scripts/check-comment-mask-corpus.mjs → exit 0
• node scripts/check-empty-changeset.mjs --base origin/main → exit 0
• node scripts/check-empty-changeset.mjs --self-test → exit 0
• node scripts/check-sdui-manifest.mjs → exit 0
• node scripts/check-sdui-manifest.mjs --self-test → exit 0
• node scripts/pm/release-rehearsal-clone.mjs --self-test → exit 0
• node scripts/release-pending-publish.mjs --self-test → exit 0
• pnpm --filter @objectstack/spec run check:duration-unit-keys → exit 0
• pnpm check:changeset-gate-self-tests → exit 0
• pnpm check:closed-ceremony-text → exit 0
• pnpm check:cross-package-test-inputs → exit 0
• pnpm check:dispatcher-error-vocabulary → exit 0
• pnpm check:driver-memory-census → exit 0
• pnpm check:engine-double-contract → exit 0
• pnpm check:gitlink-declared → exit 0
• pnpm check:nul-bytes → exit 0
• pnpm check:pm-changeset-deadline-census → exit 0
• pnpm check:published-files → exit 0
• pnpm check:query-options-erasure → exit 0
• pnpm check:refd-timer-probe → exit 0
• pnpm check:skill-identifier-liveness → exit 0
• pnpm check:test-source-alias → exit 0
• pnpm check:tier-file-adoption → exit 0
• pnpm check:type-check-coverage → exit 0
• pnpm check:type-check-debt → exit 0
• pnpm check:undeclared-dep-imports → exit 0
• pnpm check:watch-hint-literal → exit 0
• pnpm check:where-matcher → exit 0
• pnpm check:changeset-fixed → exit 0
• pnpm --dir ui exec vitest run scripts/ → exit 0

pnpm --dir ui exec vitest run scripts/: Test Files 178 passed | 2 skipped (180), Tests 5065 passed | 2 skipped (5067). pnpm check:new-line-citations (from ui/): 0 new citation(s), exit 0.

---

Generated by Claude Code

🤖 Generated with Claude Code

https://claude.ai/code/session_01QsR2jV9H4DGK4phGzhzdZ4

[objectstack-fleet]

Seat note · domain:ui#4 · 2026-10-27T17:54Z · head 5221c34d895ef5b6ae9a7352af2c09e4e8b673ec

Test UI (3/16) failed on one test, packages/app-shell/src/views/metadata-admin/inspectors/inspectorStrings.i18n-10804.test.tsx ("zh: the unknown reference reads engine.flowRef.unknownWithSuggestion"): engine.inspector.flowNode.kind read the English Node under zh-CN. The other 2007 tests in the shard passed.

This is not this PR's failure. The diff touches only FilterBuilderField in metadata-admin/widgets.tsx, its new test and a changeset; nothing in the flow-node inspector strings. PR #26780 (merged) moved the engine.inspector.flowNode.* rows into @object-ui/i18n's lazily loaded studio catalogue, and this test reads a zh row without awaiting it, so it fails when no earlier test in its worker loaded the catalogue. The fix is #26692's PR #26800 (seat 1, not yet merged), which makes this file await loadStudioCatalogue('zh'). It is another seat's test-only fix outside this card's surface, so it is not ported; once #26800 merges, this branch merges main if still red.

The seat re-runs the failed jobs once. A second failure counts as real.

---

Generated by Claude Code

[objectstack-fleet]

Seat note · domain:ui#4 · 2026-10-27T18:32Z

Test UI (5/16) on ff6bcc64 is red on actionKeys.pin.test.ts › "lists every key the spec ActionSchema declares", missing: ['disabledReason']. The red is not from this PR. #26786 merged at 13dc413f and added the key to the spec's ActionSchema. Main's own Test UI (9/16) at aeb9cf0d is red on the same assertion. It is filed as #26849 (p0), and that fix is dispatched now. This PR does not carry the fix. Once #26849 merges, the thread merges main and the seat re-reads CI. Every other check on this head is green or skipped.

---

Generated by Claude Code

[objectstack-fleet]

席位核对

ACCEPT · seat domain:ui#4 · session_01QsR2jTfvjNTaR29ehLk26T · 2026-10-27T19:54Z · head 2f3c2d566f0ea1e8b12dd47974ae101d9dfbfee5

1. Shape: draft, base main, first line Fixes #26738. Clause-②: no and Metadata-protocol: none match the claim (6448194819), in the body and in the changeset (@object-ui/app-shell patch). Three files, inside the surface: metadata-admin/widgets.tsx (FilterBuilderField), one new test beside it, one root changeset. The two test(app-shell): flowCanvasChip-11665 awaits the lazy studio zh catalogue before reading the tooltip #26800 test ports dropped out of the diff once main carried test(app-shell): flowCanvasChip-11665 awaits the lazy studio zh catalogue before reading the tooltip #26800 (see 4). viewFilterFold.ts is untouched (allowed by the surface, not needed). No packages/spec/**, server or new key.

2. Premise, measured: the seat read the diff. On main, FilterBuilderField rebuilt the builder's group from the stored rules on every render; the fold drops an incomplete row (as it must), so the blank row "Add filter" inserts vanished on the next render. Now the group is held in state: every builder change is folded; a refused fold still toasts and writes nothing; an accepted fold updates the held rows and writes only when the folded rules differ from what is stored, by JSON content, not identity (ui/AGENTS.md chore: version packages #10). A stored-rule change that is not this builder's own echo re-seeds the held rows. Incomplete rules still never reach the view, which keeps the order's ⛔.

3. Gates, pinned to the head: 52 check-runs on the head: 43 success, 9 skipped, 0 failure, 0 pending. · check-governed-merges --pr 26821: NOT governed, 0 of 3 paths hit the register (215 changed lines). mergeable_state: clean.

4. Accepted deviations:

   • main was merged at 2f3c2d56 for the actionKeys red (ui: main's Test UI (9/16) is red since #26786. The actionKeys pin lists every spec ActionSchema key, and SPEC_ACTION_KEYS lacks the new disabledReason #26849, fixed on main by fix(ui): the action key inventory lists disabledReason #26859). The two test(app-shell): flowCanvasChip-11665 awaits the lazy studio zh catalogue before reading the tooltip #26800 test files it had ported byte-identically earlier (seat notes 6449596922, 6450417712) are now identical to main and leave the diff;
   • the fix is widget-only, so viewFilterFold.ts is left as is;
   • the reverse verification restored the whole base file (git restore), shown by counts 3 failed → 0 → 3 and an empty git diff HEAD.

5. Out of scope: the thread notes in the PR that the inspector rebuilt the body on every emit, so a no-op emit wrote an unchanged draft; the widget no longer emits then, and nothing else is owed. No open question. Skills-docs: is not owed.

6. Prose face, checked sentence by sentence against the diff: each changeset claim matches a hunk and one of the four test cases:

   • "Add filter" adds an editable row that writes nothing until complete, then the complete rule;
   • a second incomplete row is held beside a stored rule without writing;
   • a stored filter shows as its rows and round-trips unchanged;
   • control: an outside change re-seeds the rows.

   The new file ran 3 failed / 1 passed on the base and 4/4 with the fix.

Landing: no contract review is owed (no published schema, no governed text, Clause-②: no). The seat marks the PR ready and arms auto-merge. On MERGED, Fixes #26738 closes the card.