[objectstack-fleet]

Requested by Tesla · project thread

Part of #25712 (4b PR 2d-1: the ui/scripts/__tests__ files that read ui/.github/workflows/ci.yml read the root ci.yml instead, or retire).
Part of objectstack#25677.

Clause-②: no
Metadata-protocol: none

维护者速读(草稿)

• 改了什么: ui/scripts/__tests__ 里有 28 个测试文件读的是嵌套的 ui/.github/workflows/ci.yml,而这个文件在 monorepo 里根本不会运行(GitHub 只运行仓库根目录的 .github/workflows/)。本 PR 新增一个小 helper,按路径读根目录的 ci.yml。17 个文件改为断言根 ci.yml 里真正运行的 type-check-ui / test-ui / build-e2e-ui 任务;另外 11 个文件里,凡是断言对象是根目录没有移植的 objectui 任务(coverage、Build Docs、shard 汇总、timings 上传、两个选择器、turbo 缓存规则)的用例,都退役,并在原处写明由 tooling(ui-wiring): objectstack-side inert wiring for the ui/ workspace — workspace globs, root turbo tasks, path-filtered ui test/typecheck/lint/E2E jobs, spec change compiles the Console; pin gates retire at cutover (parent #25677, 3/6) #25697 workflow 清单的哪一行承接。其中 4 个文件没有别的内容,整文件删除。
• 为什么改: 2d-3 要删除 ui/.github/。删之前这 28 个文件必须先不依赖它,否则删除时会红 28 个文件。改完以后,无论 ui/.github/ 在不在,这 28 个文件都是绿的。
• 风险与代价(含回滚): 只动测试和一个测试 ledger,不动任何会运行的代码、workflow 或受管面。退役的 pin 不会悄悄消失:每一条都列在下面的 follow-up 表里。其中一条是真问题:根目录 type-check-ui 的 Turbo 缓存保存没有 objectui#6577 的上限(把那组 pin 指向根任务,6 条里 5 条红)。按 PM 的裁定不在本 PR 里改根 workflow,只列为 follow-up。回滚 = revert 本 PR。
• 席位意见:(留空)
• 你要做的: 无。Tier S,由席位落地。

席位核对

(left for the dispatch seat)

Why

GitHub runs only the repository root's .github/workflows/. Since the cutover (#25825), objectui's own ci.yml sits at ui/.github/workflows/ci.yml and runs nowhere. Its four jobs that matter were ported into the root ci.yml by #25697: Type Check became type-check-ui, Test became test-ui, Build & E2E became build-e2e-ui, and lint.yml's Lint became lint-ui.

The 2d census (local, main e167956e plus 2c, with ui/.github/ deleted; per-file table below) found that deleting ui/.github/ reds 81 ui/scripts/__tests__ files. Thirty of them read ui/.github/workflows/ci.yml. Two of those thirty, markdown-test-inputs.test.ts and select-affected-tests-wiring.test.ts, belong to #25851 and are not touched here. This PR is the other 28. The 51 files that scan the workflow directory or read other single workflows are 2d-2 and 2d-3.

The rule for each pin, per the dispatcher's ruling: retarget it to the root workflow by path, or, when the root did not port its subject, retire it and name the roster row (#25697, comment 6388182608) or the root job that now carries the subject. Nothing is skipped, and nothing is folded into a root workflow here; work worth folding is listed as follow-up rows.

What changes

• New ui/scripts/__tests__/helpers/root-ci.ts. It reads <workspaceRoot()>/.github/workflows/ci.yml and never a copy under ui/. It provides:
  • jobBlock(ci, key), which throws on a missing job, so a renamed job is a red pin and never an empty block;
  • triggerBlock and withoutComments;
  • uiFilterGlobs, the filter job's ui: paths filter, which throws on an empty list;
  • globMatches and repoPath;
  • uiRun(alias), which returns pnpm --dir ui run <alias>, the root's spelling of objectui's commands.
• 17 files retargeted to the root jobs. The ordering pins keep their meaning inside the root job: "after pnpm install --frozen-lockfile", "before or after turbo's type-check task", "fetch-depth: 0 above the step". The three path-filter pins asserted that objectui's push paths-ignore did not hide a surface. They now assert two things: the root triggers carry no path filter at all, and the filter job's ui: globs cover each surface the gate judges, which is what the job's if: reads.
• 11 files retire pins on jobs the root did not port. Each retirement is a comment where the pin was, naming its row. Four files had nothing else and are deleted: coverage-shard-reporter-readability, coverage-shard-width, docs-build-trigger and turbo-cache-save-bound. The pure-module cases in the other seven stay, for example check-test-shard-results's budget sizing, which reads only constants.
• ui/scripts/markdown-test-inputs.mjs drops two ledger rows: docs-build-trigger.test.ts (deleted) and check-action-forward-parity.test.ts, whose only markdown literal (content/docs/guide/ci-cd-pipeline.md, an argument to the old matcher self-check) went with the retargeted pin. markdown-test-inputs.test.ts itself is not touched. Without these two removals it reports stale-test.

Disposition table (28 files)

#	File	What it read in ui/.github/workflows/ci.yml	Disposition	Carried by now
1	check-action-forward-parity	step after install; push paths-ignore	retarget	type-check-ui; no trigger path filter; ui: globs cover every SURFACES file
2	check-doc-example-shared-reader	the step	retarget	type-check-ui
3	check-handler-key-read-sites	the step	retarget	type-check-ui
4	check-i18n-call-site-keys	step after install	retarget	type-check-ui
5	check-i18n-designer-table-parity	step after install; paths-ignore	retarget	type-check-ui; ui: globs cover the designer table and the gate
6	check-i18n-en-drift	step after install; fetch-depth: 0; paths-ignore	retarget	type-check-ui (fetch-depth: 0); ui: globs cover the 10 packs, the ledger and the designer table
7	check-lucide-icon-record-names	pnpm <alias> anywhere	retarget	root ci.yml run: pnpm --dir ui run <alias>
8	check-object-metadata-write-doors	the step	retarget	type-check-ui
9	check-package-self-import	step after install, in type-check	retarget	type-check-ui
10	check-phantom-dependencies	step after install, in type-check	retarget	type-check-ui
11	check-unreferenced-sources	step after install, in type-check	retarget	type-check-ui
12	check-published-tsconfig-tooling-exclude	the step; published-dist-gate.yml	retarget; workflow half retired	type-check-ui; published-dist-gate.yml is roster row 23 (kept, follow-up port)
13	check-pending-changeset-literals	the Decide whether this change needs a full run exclude pathspecs	retarget	test-ui reads the ui output; ui: globs cover scripts/ and packages/ test sources
14	scripts-type-check	type-check job, step after install	retarget	type-check-ui
15	vitest-config-type-check	after install, before pnpm type-check	retarget	type-check-ui, before pnpm exec turbo run type-check
16	vitest-setup-type-check	after pnpm type-check	retarget	type-check-ui, after pnpm exec turbo run type-check
17	ensure-chromium-ready	ci.yml and live-e2e.yml use the script, no --with-deps	retarget; live-e2e half retired	build-e2e-ui; live-e2e.yml is roster row 17 (kept, follow-up port)
18	check-doc-expression-carriage	node scripts/check-doc-expression-carriage.mjs in docs	1 case retired	roster row 7 Build Docs (@object-ui/site) (kept, follow-up port); the page-row pin stays
19	check-test-shard-results	test-aggregate wiring; the Test (shard N/M) name	8 cases retired, budget case kept	roster row 7 Test (the shard aggregate) (dropped until cutover; port with objectstack's shard attestation)
20	shard-headroom	headroom step in test-aggregate and coverage-report	2 cases retired	roster row 7 aggregate and coverage
21	test-file-timings	test job name, timing reporter, upload; refresh --shards	4 cases retired	roster row 28 test-shard-timings-refresh.yml (→ sub-issue 4; "restore Test UI's timings reporter and upload"); test-ui's comment records the wait
22	select-affected-tests	the Select the tests this change affects wiring	4 cases retired	roster row 7 Test (shard n/16): test-ui runs the whole suite and does not wire the selector
23	select-type-check-scope	tc-scope step, TURBO_FORCE, root-programs build	5 cases retired	roster row 7 Type Check: type-check-ui uses turbo ...[base] and its own root-programs build step
24	report-turbo-cache	type-check prune and save; every workflow's turbo cache	7 cases retired	the four root * UI jobs carry objectstack's cache steps (job key, build-core fallback, push-only save, no prune or report); follow-up F2
25	coverage-shard-reporter-readability (deleted, 4 cases)	test-coverage reporter flags	retired	roster row 7 coverage (kept, follow-up port)
26	coverage-shard-width (deleted, 5 cases)	test-coverage and coverage-report width	retired	roster row 7 coverage (kept, follow-up port)
27	docs-build-trigger (deleted, 17 cases)	docs job's docs-changes decision step	retired	roster row 7 Build Docs (kept, follow-up port)
28	turbo-cache-save-bound (deleted, 6 cases)	type-check's split, bounded cache save	retired	type-check-ui carries the save but not the bound; measured below; follow-up F1

Follow-up rows (fold-worthy, not done here)

Row	What	Why it is not in this PR
F1	Bound the root * UI jobs' Turbo cache saves the objectui#6577 / #7048 way: split restore and save, timeout-minutes, continue-on-error, the cache-hit guard, and a ceiling that leaves the verdict room. Then restore turbo-cache-save-bound's pins against type-check-ui.	A real finding. Its six pins, pointed at the root type-check-ui job by a scratch copy, fail 5 of 6. Only "save after the last command" holds. The other five fail because the job keeps a combined actions/cache (the pnpm store), its save has no bound, continue-on-error or cache-hit guard, and its ceiling is 40 minutes. Fixing it is a root ci.yml edit (Tier H), and the dispatcher ruled no folds into root workflows in these PRs.
F2	Decide whether the root ui jobs run report-turbo-cache.mjs (prune before save, restored-key notice), or retire the script.	Root cache design is objectstack's; a fold.
F3	Test UI shard aggregate at the required set: port with objectstack's shard attestation, or retire check-test-shard-results.mjs and shard-headroom.mjs.	Roster row 7, the maintainer's required-set call.
F4	Coverage lane: push lane vs objectstack's coverage-nightly.yml, then restore the reporter and width pins.	Roster row 7, kept, follow-up port.
F5	Build Docs (@object-ui/site): port, then restore docs-build-trigger's derived closure and the carriage-census wiring pin.	Roster row 7, kept, follow-up port.
F6	Timings reporter and upload on test-ui with the refresh workflow, then restore test-file-timings's ci pins.	Roster row 28 (sub-issue 4).
F7	Wire select-affected-tests.mjs into test-ui once its git diff paths are cwd-relative, or retire it, and the same for select-type-check-scope.mjs.	Roster row 7.
F8	live-e2e.yml (row 17) and published-dist-gate.yml (row 23) ports bring their workflow halves back.	Kept, follow-up port.

Verification

Run from ui/ per ui/AGENTS.md, at this head (main c475b45b plus this commit):

Run	Result
The 24 modified test files plus markdown-test-inputs.test.ts, with ui/.github/ present	25 files, all passed
The same 24 files with ui/.github/ moved aside (the 2d-3 layout)	24 passed (779 tests). markdown-test-inputs.test.ts fails at collection on ENOENT ui/.github/workflows/ci.yml; that file is #25851's.
Whole scripts/__tests__ suite, --shard=1/4 through 4/4, under scripts/pm/os-verify-lock.sh	5,447 passed, 26 failed, 2 skipped
pnpm type-check:scripts	the one error is packages/components/src/renderers/action/resolve-icon.ts(265,69) TS2339, identical on unmodified main here; none in a touched file
pnpm exec eslint (touched files), pnpm lint:root, lint:coverage, check:test-path-roots, check:ui-root-lookups, check:lint-rule-coverage, check:entry-guard, check:vi-mock-override-shape; root pnpm check:test-source-alias	all exit 0; no lint warning in a touched file

Of the 26 suite failures, one is the owned row from 2c (check-pre-install-import-graph, unresolved: scripts/objectstack-pin.mjs; it clears in 2d-3 with the nested workflows). The other 25 are date-only-zone-pins-10183 ×23 and unit-registry-absence-collision ×2. They fail identically on a clean main checkout here ("ran no case at all" on this unbuilt box), so they are not this diff's.

Derived gates (node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack from this diff): 18 families, all exit 0 at this head. --ran reconciles 18 run, 0 NOT-MEASURED, 0 UNRUN. The slowest were check:query-options-erasure (308 s), check:type-check-debt (287 s) and check-comment-mask-corpus (208 s).

The 2d census: what deleting ui/.github/ reds (81 files; 30 read ci.yml)

Local, main e167956e plus 2c, whole suite in two shards with ui/.github/ deleted. 83 files fail. Two of them (date-only-zone-pins-10183, unit-registry-absence-collision) fail identically on clean main on that box. The other 81 read ui/.github/.

• ui/.github/workflows/ci.yml (30 files): rows 1–28 above, plus hourly full run: red on main (Lint & Type Check) #25851's markdown-test-inputs and select-affected-tests-wiring.
• The workflow directory as a population (27 files, 2d-2): check-doc-component-types, check-doc-example-ids, check-doc-example-types, check-doc-json-fences, check-doc-snippet-types, check-hand-rolled-comment-mask, check-i18n-dead-keys, check-lint-rule-coverage, check-lockfile-dedupe, check-lockfile-integrity, check-pre-install-import-graph, check-prompt-component-keys, check-published-dist-tooling, check-readme-exports, check-shell-escape-residue, check-skill-eval-tokens, check-skill-examples, check-skills-paths, check-vi-mock-inherit, check-vi-mock-specifiers, ci-setup-pnpm-wiring, dependabot-merge-gate, docs-links-workflow, merge-queue-reporting, workflow-cache-save-bound, doc-version-claims, labeler-package-coverage (.github/labeler.yml).
• Single workflows (24 files, 2d-2/2d-3):
  • repo-gates.yml: check-action-ref-convention, check-control-bytes, check-doc-fence-languages, check-doc-snippet-emitted-census, check-docs-route-eager-closure;
  • lint.yml: bash32-floor-wiring, check-test-path-roots, check-vi-mock-override-shape, entry-guard-wiring, lint-workflow;
  • changeset-presence.yml: check-changeset-claims, check-changeset-presence, render-changeset-claims-comment;
  • changeset-guard.yml: check-changeset-no-major, check-changeset-overwrite;
  • performance-budget.yml: check-eager-closure-budget, ci-cd-pipeline-doc, render-budget-comment;
  • one file each: check-governed-queue-guard (governed-surface-guard.yml), check-links-workflow (check-links.yml), check-merge-queue-head (merge-queue-head-patrol.yml), check-new-cross-file-line-citations (line-citation-gate.yml), check-required-check-set (dependabot-auto-merge.yml, required-check-set-patrol.yml), shadcn-check-report (shadcn-check.yml).

🤖 Generated with Claude Code

https://claude.ai/code/session_01A6SEitNngAMugGbuJHWsDS

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 40261620902 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 (11/16) — 失败步骤: Run the ui tests (shard 11/16)

  × finds no pre-install gate reaching a package 11ms
  FAIL   unit  scripts/__tests__/check-pre-install-import-graph.test.ts > the real tree — every step objectui#6148 measured is still derived > finds no pre-install gate reaching a package
    ↳ 失败原因: AssertionError: a script this repository runs before `pnpm install` needs `node_modules` to load: expected [ Array(1) ] to deeply equal []
  

↳ 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）：

• scripts/__tests__/check-pre-install-import-graph.test.ts — 24h 内已弹出 6 个 PR（fix(ui-merge): two tests that #25745 broke on main read the merged tree — the Changesets config through changesetConfigPath(), the SDUI html tier by its tags (#25711) #25852、ci(ui-merge): Type Check UI builds the five vitest-setup packages before that step, so an empty ui scope no longer reds it #25856、ci(ui-merge): register Lint UI and Build & E2E UI in REQUIRED_CONTEXTS #25857、fix(ui-merge): the CI page and two of its tests stop reading the two workflows #25745 deleted #25861、test(ui): three ui/ script tests spell their temp bases so the whole-tree scratch-dir sweep reads them (PM dispatch-gates red on main) #25865、test(ui-merge): the ui gate-wiring pins read the root ci.yml that runs them; pins on jobs it did not port retire #25879），但它们同属 1 条投机栈 ⇒ 1 次独立命中。汇总 issue：Queue-flake anchor: scripts/__tests__/check-pre-install-import-graph.test.ts #25863（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 8 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)