[objectstack-fleet]

Requested by Steve · project thread
Fixes #25898

Clause-②: no
Metadata-protocol: none

What changes

An empty list chose its copy from the user's query and the view's declared filter alone. Two cases from the card's table were wrong:

• No filter, caller may not create. A member who may read none of an object's records and holds no create grant was told "Nothing here yet / Create your first record to get started." beside no Create button. Once the permission answer has loaded, a caller without the listNew verdict of the affordance-to-grant map (managed-object policy, effective API operations and the caller's create grant, the same verdict the page's own New reads) now gets "No records you can see / There are no records here that you have access to. If you expected to see some, ask your administrator for access." The add-record action is not drawn for that caller, even on a view whose addRecord is enabled.
• A "my" view. When the view's declared filter names the signed-in user (a comparand equal to the caller's id, which is what a host-resolved {current_user_id} becomes, or the placeholder itself), the empty state reads "Nothing of yours yet / This view lists only records that are yours, and you don't have any yet." instead of "No records match this view's filter.", with the add-record action only under the create grant.

Unchanged: the first-run copy and action for a caller who may create, the user-query copy, the plain view-filter copy, an authored emptyState (it still wins), the 403 error panel (its own state, untouched), and every surface with no permission provider mounted (fail-open, as the affordance map is).

Where it lands, following the card's measured producers:

• ListEmptyState (plugin-list, module-internal) makes the decision: two new optional props, baseFilterNamesCurrentUser and mayCreate (false is a known denial, absent is "not known"), and the exported helper filterNamesCurrentUser.
• ListView (grid and gallery visualizations) passes both from the empty-state call site: the resolved declared filter against the session scope's currentUserId, and resolveAffordance('listNew', …) once usePermissions().isLoaded.
• ObjectGallery (standalone) passes the same two facts.
• ObjectGrid's default empty state makes the same decision in its own empty-state region (it draws no add-record action, as before). plugin-grid and plugin-list do not depend on each other, so the filter walk is restated there in a few lines.
• @object-ui/i18n: list.ownedNoneTitle, list.ownedNoneMessage, list.noneVisibleTitle, list.noneVisibleMessage in all ten packs, and in the two in-code English fallback tables (LIST_DEFAULT_TRANSLATIONS, ObjectGrid's).
• scripts/sdui-manifest.record.json: the registry-source hash re-recorded by node scripts/gen-sdui-manifest-node.mjs, which every ui/packages/** source change requires (check-sdui-manifest); the manifest itself is byte-identical.

The copy is generic rather than naming the object ("你没有可查看的工单" in the card) so it needs no plural forms across ten locales; the zh pack reads 「没有你可查看的记录」 and 「你还没有负责的记录」.

Tests

At head 8878ebe8 (implementation unchanged since be788ede; later commits are the main merge and the manifest record):

• New: ListView.emptyStateAccess-25898.test.tsx (9), ObjectGallery.emptyStateAccess-25898.test.tsx (4), ObjectGrid.emptyStateAccess-25898.test.tsx (5), through the real MePermissionsProvider and FilterScopeProvider: 3 files, 18 tests passed.
• Reverse verification: the four source files and the locale packs restored to the base commit 38050ced with the tests kept: 11 failed / 7 passed, exactly the predicted split (every no-access, "mine" and zh row red; the create-grant, no-provider and plain-filter controls green). Restored from HEAD afterwards, git diff HEAD empty.
• vitest run packages/plugin-list/ packages/plugin-grid/ packages/i18n/ packages/plugin-form/src/affordanceGrantMap-12082.test.tsx (the affordance census included): 498 files passed, 5253 tests passed, 13 skipped.
• type-check for @object-ui/plugin-list, @object-ui/plugin-grid, @object-ui/i18n: clean.
• node scripts/pm/dispatch-gates.mjs --commands: 47 derived families run locally, all exit 0 (check-sdui-manifest exited 1 before the record was regenerated and 0 after); --ran reconciliation: 47 accounted, 0 not measured.

Acceptance notes

• The {current_user_id} walk is shape-agnostic: any string comparand equal to the caller's id counts, so a filter like owner = me OR status = open also reads "Nothing of yours yet". Views of that shape are rare; the copy is still true of the caller's own records.
• The toolbar's add-record buttons (top and bottom) still follow addRecord.enabled alone inside ListView; the console host already withholds addRecord without the create grant, so only a non-console host could show them to a caller without the grant. Left out of this card's empty-state scope; noted, not filed.
• A shared home for the decision in @object-ui/core would let ObjectGrid and ListEmptyState share one function; core is outside this card's claimed surface, so the grid restates it.

---

Generated by Claude Code

[objectstack-fleet]

席位核对 · ACCEPT · seat domain:ui#1 · 2026-10-26T08:23Z

Head 8878ebe80ccf (read from the PR object).

• Shape. Draft on main; Fixes #25898; Clause-②: no, Metadata-protocol: none; changeset .changeset/25898-list-empty-state-access.md in the root .changeset/ (patch, plugin-list + plugin-grid + i18n).
• Checks pinned to the head. 61 named check runs, latest per name: 54 success, 7 skipped, 0 failing, 0 pending. Lint & Repo Gates, TypeScript Type Check, Test UI 16/16, Type Check UI all success.
• Governed surface. check-governed-merges.mjs --pr 25987: 0 of 19 paths governed.
• Changeset against the diff. ListEmptyState gains baseFilterNamesCurrentUser and mayCreate; the user's own query still wins, then a filter naming the caller ("Nothing of yours yet"), then another declared filter, then, with no filter and a KNOWN listNew denial, "No records you can see"; first-run otherwise. A known denial withholds the add-record action even when onAddRecord is passed. mayCreate is asked only after the permission answer has loaded, so an unloaded provider keeps today's copy. ObjectGrid's default empty state applies the same order. Each sentence matches the code.
• Serial constraint held. ObjectGrid.tsx changes only the default-empty-state copy and its translation defaults, not the group-header region fix(plugin-grid): order a grouped grid's groups by the level's orderBy #25977 changed.
• Known limit, not blocking. filterNamesCurrentUser is shape-agnostic and does not read the operator, so a filter that EXCLUDES the caller (owner != current user) also reads "Nothing of yours yet". The changeset scopes the feature to "owner equals the current user", and no shipped view is known to use the inverse. If one turns up, it is a card, not a hold on this PR.
• Out-of-surface file. scripts/sdui-manifest.record.json: the derived hash only.

⚠️ Merge order: #25978 and #25984 append to the same ten locale packs, and #25977 / #25978 / #25984 move the same manifest-record line. Whichever lands later merges origin/main and regenerates the record.

Landing: ready + auto-merge through the relay.

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 40317813492 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

分类：failure —— 按下面的日志分诊。

失败的 job（日志抽取，best effort）：

• Build Core — 失败步骤: Build packages (excluding docs)（日志不可读，点进 job 看）

• Build & E2E UI — 失败步骤: Build the Console for E2E

  ✗ Build failed in 5.46s
  

• Type Check UI — 失败步骤: Run type-check over the ui packages（日志不可读，点进 job 看）

• Lint UI — 失败步骤: Build every ui package the gates below read（日志不可读，点进 job 看）

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️ 断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• ⚠️ 本次没有可用的聚合签名（日志里没有能解析出测试文件名的 FAIL 行）—— 这不是「没有同签名的其他 PR」，是这一轮没测到。跨 PR 聚合本次不可用，请手工比对其他 PR 的同类评论。
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 28 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)