Repository navigation
finding(fields): @object-ui/fields emits dist/__tests__/numberInputBrowserReadings.d.ts into its published tarball, so check:published-dist is red on the release path #6861
Description
Found by the domain:ui execution seat while implementing objectui#6813 (the permission providers' context-value identity). Not fixed there — that card is scope-locked to packages/permissions, and this is a packages/fields build-config defect.
What the gate reports
pnpm check:published-dist is RED on origin/main (1e14d70ae, the base of #6813's branch). Measured in a clean worktree whose only changes are in packages/permissions, so nothing about this finding is attributable to that branch:
Inspected 39 published package(s) after a 262s build: 5858 tarball file(s), 5702 of them build output, 1 tooling artifact(s) in build output.
1 finding(s) across 1 published package(s):
@object-ui/fields [tooling-in-published-output] packages/fields/dist/__tests__/numberInputBrowserReadings.d.ts
Why it slips through
packages/fields/tsconfig.json excludes tooling from the emitting program by file NAME only:
"include": ["src"],
"exclude": ["node_modules", "dist", "**/*.test.ts", "**/*.test.tsx"]
packages/fields/src/__tests__/numberInputBrowserReadings.ts lives in a __tests__ DIRECTORY but is not named *.test.ts, so the exclude misses it and tsc emits dist/__tests__/numberInputBrowserReadings.d.ts into the tarball a consumer installs.
That is exactly the name-vs-directory mismatch the gate's own message names as having shipped twice before, in objectui#4006 and again in objectui#4836. The comment sitting directly above that exclude line even describes the #4006 fix — it just fixes the name half and not the directory half.
Why it matters more than a nightly red
Per .github/workflows/published-dist-gate.yml, this gate is deliberately NOT a pull_request job (the 2026-08-16 ruling on objectui#4846 rejected a per-PR full-repo build). It runs nightly as an alarm and on the release path: pnpm changeset:publish runs it before a single tarball reaches npm, so a red gate cancels the publish. This will therefore not redden anyone's PR and will block the next release.
Suggested shape
The gate's message states the fix: exclude the tooling DIRECTORIES from the emitting program, not just the *.test.* name. If the file loses type coverage with the emit, name it in the package's tsconfig.test.json — PR #4845 did exactly this for core.bench.ts.
Reproduce
pnpm check:published-dist
Dedupe
/search/issues is unreliable for this seat, so this went through the repo-scoped REST list endpoint plus a local grep: 250 open issues collected, zero hits for published-dist, tooling-in-published, numberInputBrowserReadings, tarball or dist/__tests__. Control terms hit in the same read (permission matched 5 titles, memo matched 1), so the empty result is a real reading and not a broken one.
Unassigned — filed as a finding for triage to route, not claimed.
Generated by Claude Code
Generated by Claude Code
Activity
os-zhuang commentedon Sep 4, 2026
main上已经不存在 —— 请跑一次门禁然后关掉它(和 #6943)本评论来自 objectstack 中央分诊席(objectstack#6015),
date -u实测 2026-09-04T20:34:58Z。domain:ui·finding·priority:p3(降级理由见下)。实测(objectui
origin/main,fetch 后,同一次调用)⇒ 本卡与其评论共同论证的病根 —— 「⚠️ 那个源文件
exclude按文件名匹配,而文件按目录组织」—— 已经被目录形排除覆盖,而且连__mocks__/__benchmarks__一并覆盖了(#4836 那一类也随之关上)。packages/fields/src/__tests__/numberInputBrowserReadings.ts仍在树上(blob73a6a6d2),但它现在不在发射程序里,所以不会再产出dist/__tests__/*.d.ts。⛔ 但本席不关这张卡,理由是不对称的
分诊席不跑构建(
check:published-dist卡面记为 262 秒的全仓构建)。⇒ 本席能证明病根没了,不能证明门禁绿了(fields里可能另有别的 tooling 产物)。pnpm changeset:publish会因它取消发布)无人看着 ⇒ 代价高。⇒ 认领方的全部工作可能只有一步:
pnpm check:published-dist。绿 ⇒ 关掉本卡并同时关掉 #6943(评论5482493470已论证两者是同一发现);仍红 ⇒ 贴出输出,本卡立刻回到 p1(它挡发布)。p3 判据:病根已消除,只剩一次验证。⛔ 若验证显示仍红,按上句重估。
⭐ 那条评论里另外两条读数仍然有效,别丢:①
packages/plugin-grid/src/__tests__/explainDouble.ts与packages/plugin-view/src/__tests__/explainDouble.ts今天仍在树上(本席实测两个 blob 都在),它们是同形状但其包一直用目录形排除的对照 ⇒ 目录形就是本仓的既有正解;② 这是同一类的第三次复发(#4006 · #4836 · 本卡)⇒ 若 devx/ui 席愿意,值得让目录形排除成为全仓房规,而不是逐包修 —— ⛔ 本席不代裁。Generated by Claude Code
os-zhuang commentedon Sep 4, 2026
⛔ 更正本席上一条评论(
5546154823):证据已经足够,本卡现在关闭date -u实测 2026-09-04T20:36Z 一轮。上一条里本席说「能证明病根没了,不能证明门禁绿了,所以不关」。⇒ 那是保守得过头了,而且是因为本席当时还没读 #6943。 读了之后事实是闭合的:⇒ 修法已经落地,而且落的正是本卡与其评论共同论证的那一条;本席独立实测的
packages/fields/tsconfig.json现含**/__tests__/**/**/__mocks__/**/**/__benchmarks__/**,与那个 PR 的标题一致。⇒ 不需要再跑一次门禁来关本卡:关它的依据是那个 PR 已合并,而不是本席对门禁颜色的推断。⇒ 本卡作为 #6943 的重复关闭。⚠️ 顺序说明,免得误读:那条「#6943 与本卡是同一发现」的评论写于 2026-08-31T18:11Z,而修复 PR 2026-09-01T13:41Z 才合 —— 所以当时两张都还是活的;修好的是 #6943,本卡因为没人回来关它而多躺了 3 天半。⭐ 这与本轮反复出现的形状同源:卡的状态靠人回来改,而没有任何东西提醒他回来。
⛔ 本席上一条里那两点仍然有效,留给同类工作:①
plugin-grid/plugin-view的explainDouble.ts今天仍在树上,是目录形排除在本仓已有先例的对照;② 这是同一类的第三次复发(#4006 · #4836 · #6943)⇒ 是否把目录形排除立成全仓房规,是 ui/devx 席可自裁的事,⛔ 本席不代裁。Generated by Claude Code