[objectstack-fleet]

Requested by Warren · project thread
Fixes #26891

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:@object-ui/react 的 extractFieldErrors 现在识别 409 UNIQUE_VIOLATION 里单个 field,返回一条带服务端句子的字段错误;表单因此把重复值标在对应输入框下。
• 为什么改:服务端(types(data-error-classification): a unique refusal on a single-column index tells the user "A record with this gsku already exists" — the field by its API key, while developerMessage carries its label #26867)已在拒绝里点名字段并用标签造句,但前端只读字段数组,唯一索引冲突仍只显示横幅加 toast,用户要猜是哪个字段重复。
• 风险与回滚:只在 code 为 UNIQUE_VIOLATION 且带字符串 field 时生效;无 field(复合索引、索引名)照旧走横幅。回滚 = revert 本 PR。
• 先看:ui/packages/react/src/utils/error-message.ts 的 extractFieldErrors 新分支;form-server-field-errors.test.tsx 新增的 unique-index 两例。
• 两分钟验证:pnpm --dir ui exec vitest run packages/react/src/utils/error-message.test.ts packages/components/src/renderers/form/__tests__/form-server-field-errors.test.tsx,预期全绿;在 console 新建一个 SKU 重复的产品,SKU 输入框下出现 "A record with this SKU already exists",无 toast。
• 需要你做的:无,席位落地。

What changes

• extractFieldErrors (@object-ui/react): when no field-error list is present and the refusal's code is UNIQUE_VIOLATION (read from code, then details.code, the same pair isPermissionError reads), the single string field (details.field on the raw @objectstack/client error, field on a hand-rolled one) becomes one WriteFieldError whose message is the server's sentence (extractWriteErrorMessage). Without field the answer stays null.
• No change to form.tsx: its existing extractFieldErrors then form.setError path marks the input, and a fully attributed refusal shows no banner and no toast.
• Changeset: @object-ui/react patch.

Why

The server half shipped in #26867: the 409 UNIQUE_VIOLATION body is { error, code, developerMessage?, field?, object? }, with error naming the field by its label. @objectstack/client parks that flat body on err.details, and normaliseClientError (@object-ui/data-objectstack) passes a unique 409 through unchanged. The last hop read only arrays (validationErrors, details.fields, fields), so it answered null and the dialog showed the generic banner and toast.

Side effect, measured: InlineEditSaveBar (@object-ui/plugin-detail) calls the same helper, so on the atomic save path a unique refusal with field is now attributed to that field (labelled in the bar and published as the per-field hint) instead of the generic text; in callback mode it was already attributed to the in-flight field. Its suites stay green (below).

Tests

• error-message.test.ts: a unique refusal with field yields one entry carrying the server's sentence (asserts code and httpStatus 409 on the refusal); a hand-rolled one with top-level field; without field it answers null; an INVALID_FIELD carrying field still answers null.
• error-message.normalisation-boundary.test.ts: the refusal survives normaliseClientError with code, httpStatus and the field entry intact.
• form-server-field-errors.test.tsx: a rejected submit marks the sku input (aria-invalid, sentence under it, no toast); without field the toast carries the generic sentence and no input is marked.

Acceptance notes

• ApiDataSource, which embeds the body as a JSON tail in the message string, is not read for field; the ObjectStack adapter is the console's path.

Verification

Tree: a17ebf4e on base 7eb613c3.

Tests and ablation

• Three touched files: Test Files 3 passed, Tests 46 passed.
• Ablation via scripts/ablation-replace.mjs (unique branch replaced by return null): exactly the four new positive tests red (two helper, one boundary, one form render), 42 passed; restored, blob equals HEAD, git diff HEAD empty. Tests resolve @object-ui/react to src through the ui vitest alias, so no dist leg.
• Consumer sweep under the lock: packages/plugin-detail/src/__tests__/, packages/react/, apps/console/src/pages/settings/, apps/console/src/pages/system/: Test Files 385 passed, 1 skipped; Tests 3607 passed, 18 skipped.
• type-check for @object-ui/react and @object-ui/components: exit 0. eslint on the four files: 0 errors.

Derived gates

node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack: 34 families, each run, all exit 0, including node scripts/check-sdui-manifest.mjs and pnpm --dir ui exec vitest run scripts/. --ran reconciliation: 34 derived, 34 run, 0 NOT-MEASURED, 0 UNRUN.

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 passes. (eslint on the touched files: 0 errors; the repo-wide run is CI's)
• 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: its generated artifacts are regenerated and committed. (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 Before Submitting list in CONTRIBUTING.md is done.

---

Generated by Claude Code

[objectstack-fleet]

席位核对

ACCEPT · domain:ui#2 · session_01JQwxbXpF772qWJjojJvRXw · head a17ebf4eb3 · 2026-10-28T02:56Z

席位意见:只认 UNIQUE_VIOLATION 且带单个字符串 field 的拒绝;服务端说不清是哪一列时照旧走横幅,不会把错误标到无辜的输入框上。行内编辑的保存条也随之点名字段,PR 已写明。

1. Shape: draft, base main; the body says Fixes #26891, which matches the order (whole card). 5 files, +152/−2; not governed (check-governed-merges --pr 27043); ≤ 3000 lines.
2. Premise measured: seat read of extractFieldErrors at head. The new branch runs only when no field list exists, and only when code (top-level or on details) is UNIQUE_VIOLATION and details.field / field is a non-empty string (firstString). Without field it still returns null, so the banner stays. The message is the server's sentence via extractWriteErrorMessage. The code check reuses errorCodeIs from @object-ui/types, as the delegation-scope branch does.
3. Gate check-runs (pinned to head a17ebf4eb3): in progress at this record; landing waits for every check green. The dev's local reading: 34/34 derived families exit 0, --ran 0 NOT-MEASURED, 46/46 tests, ablation turns exactly the 4 new positive tests red, consumer sweep 3607 passed.
4. Accepted deviations: none. form.tsx is untouched; only a test was added beside it.
5. Out of scope: none. The InlineEditSaveBar atomic path now attributes the refusal to the field; the order named this side effect and it is intended.
6. Landing clause: not governed. When every check on the current head is green, the seat lands it (pr_ready + automerge_enable), or asks the maintainer to click if the classifier refuses.

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 40655321196 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）：

• Test UI (7/16) — 失败步骤: Run the ui tests (shard 7/16)

  × zh-CN: the header reads the served 销售代表, and a save keeps the authored label 1137ms
  × ja-JP: the header reads the label the served row carries 1078ms
  × a set the published list does not hold keeps its draft label 1063ms
  × an edit to the label shows the edit, not the served label 1053ms
  FAIL   dom  packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx > PermissionMatrixEditPage — header shows the served label (objectstack#26884) > zh-CN: the he
    ↳ 失败原因: (这条 FAIL 之后 12 行内没有可识别的原因行 —— 点进 job 看)
  FAIL   dom  packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx > PermissionMatrixEditPage — header shows the served label (objectstack#26884) > ja-JP: the he
    ↳ 失败原因: (这条 FAIL 之后 12 行内没有可识别的原因行 —— 点进 job 看)
  FAIL   dom  packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx > PermissionMatrixEditPage — header shows the served label (objectstack#26884) > a set the pub
    ↳ 失败原因: (这条 FAIL 之后 12 行内没有可识别的原因行 —— 点进 job 看)
  FAIL   dom  packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx > PermissionMatrixEditPage — header shows the served label (objectstack#26884) > an edit to th
    ↳ 失败原因: (这条 FAIL 之后 12 行内没有可识别的原因行 —— 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）：

• packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx — 24h 内已弹出 5 个 PR（fix(app-shell): permission-set editor header shows the served (translated) label #27019、fix(app-shell): Home, the app switcher and launcher blocks offer only the apps /me/apps lists, so a tabPermissions-hidden app draws no tile #27021、fix(console): every console form entry point renders the form view's buttons, layout and defaults, and a tabbed form with subforms opens on its defaultTab #27027、fix(app-shell): a locked metadata item's designer canvas is read-only #27035、fix(react): mark the field a unique-index refusal names in create/edit forms #27043），但它们同属 1 条投机栈 ⇒ 1 次独立命中。汇总 issue：Queue-flake anchor: packages/app-shell/src/views/metadata-admin/PermissionMatrixEditor.servedLabel-26884.test.tsx #27150（已刷新）
• ⚠️ 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 32 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)