objectui 没有「发布性改动必须声明 changeset」闸门 —— changeset-guard.yml 的触发器决定了它只看得见已经带 changeset 的 PR(objectui#3518 因此整个从发布记录里消失)

在 #6159 的 pin bump(PR #6173,范围 f995a452..7dfbeb70)执行 #6099 复核时发现,与 #6159 本身无关,按 Prime Directive #10 单独记录。

现象

objectui#3518 —— commit 0e50440,fix(form): bind ``previous`` for field rules and stop resubmitting read-only fields —— 落在本次 pin 范围内,改动 26 个文件,覆盖 packages/components、packages/plugin-form、packages/types、packages/data-objectstack、packages/app-shell 以及全部十个 locale pack,新增两个测试文件。它没有添加任何 .changeset/*.md。

证据(在 objectui 检出里,只读):

$ git show --name-status --format='' 0e50440 -- .changeset/
(空)

后果是双向的:

objectui 自己的发布记录里没有它。 7dfbeb7(chore: release packages (#3248))消费掉了当时挂着的 88 个 changeset 并发布,0e50440 的代码随 @object-ui/components 等包一起出去了,但任何 CHANGELOG 里都没有一行提到它。
平台侧的 console changeset 也拿不到它。 scripts/objectui-changeset-digest.mjs 的判据是「objectui 声明了什么」(bump-objectui.sh 的 pin changeset 只收 feat|fix 且静默截断到 40 条 —— 破坏性 refactor! 进不了前端发布记录 #4731),这条判据本身是对的 —— 它忠实地把 0e50440 归进「20 个不带 changeset 的 commit」并把这个数字写进正文。但一个真实的、用户可见的表单行为变更,就这样只以一个计数的形式存在于 v17 的发布记录里。

这正是 #3340 记录过的那一类损失(v16 因 pin 滞后丢了 4 条前端改动)换了一扇门进来:不是 pin 陈旧,而是源头就没声明。check:objectui-pin-fresh 对此完全无能为力 —— 它证明的是 pin 是否 CURRENT,而 0e50440 已经在范围内了。

根因:闸门的触发器把它自己挡在门外

objectui 有 .github/workflows/changeset-guard.yml,但读它的触发器:

on:
  pull_request:
    branches: [main, develop]
    paths:
      - '.changeset/**'

以及它唯一的 job:

jobs:
  no-major:
    name: Changeset Bump Policy

两件事叠在一起,结果是结构性的:

这个 workflow 只在 .changeset/** 发生变化时才启动。一个「漏了 changeset」的 PR 按定义不动 .changeset/**,所以它一个 job 都不跑。
就算跑了,它检查的也是级别策略(不许 major),而不是「这次改动该不该带 changeset」。

也就是说:能发现「缺 changeset」的检查,恰恰跑不到缺 changeset 的 PR 上。这个 workflow 的注释本身就写明了它是为「只加 changeset 的 PR 会一个 workflow 都不启动」这个反向问题设计的 —— 设计是自洽的,只是正向的那个问题从来没有闸门。

对照本仓的既有判断:「a check that does not run reports nothing」(objectui-pin-freshness.yml 的 WHY THE JOB IS NEVER SKIPPED 段),以及 Route & surface ownership §3「Absence must be loud」。这里的缺席是完全静默的。

需要决定的事(留给 triage,我不预设)

修复面在 objectui,不在本仓,所以这条主要是路由 + 记录。可选方向,成本递增:

A. objectui 加一个反向触发的 changeset-required 闸门:PR 改到 packages/*/src/** 但没加 .changeset/*.md 时红,配 skip-changeset 标签逃生口。这是最直接的一版,和本仓 check-empty-changeset.mjs + skip-changeset 的组合同构。
B. 只补记录不加闸门:给 objectui#3518 补一个追溯 changeset。能修这一次,修不了这个类。
C. 平台侧把「无 changeset 的 commit」从计数升级为点名:objectui-range.mjs 已经有 --all 能按 subject 逐条列出被排除的 commit(见其 self-test「--all itemizes the excluded commits by subject」),把它接进 console changeset 正文,至少让漏声明的改动在平台发布记录里留下标题。这不解决 objectui 侧的根因,但把静默变成了可读的噪声。

我倾向 A + C:A 治根因,C 让平台侧在 A 落地之前不再是全盲,而且 C 需要的能力已经写好了、只差接线。B 单独做等于承认下一次还会发生。

不在本 issue 范围内

chore(console): bump objectui pin f995a452 → a4cff5b (17 commits / 7 releasing changesets) — required before cutting rc.4 #6159 / PR chore(console): bump objectui pin f995a452 -> 7dfbeb70 (28 commits / 9 releasing changesets, all patch) — rides rc.5 #6173 本身不受影响:digest 的行为完全正确,0e50440 的代码照常进 console build,只是发布记录里没有它。PR chore(console): bump objectui pin f995a452 -> 7dfbeb70 (28 commits / 9 releasing changesets, all patch) — rides rc.5 #6173 已在正文里把这件事写明,没有粉饰。
objectui pin bump 生成的 console changeset 对破坏性变更零标注 —— digest 只认 major 声明级别,而 objectui 在 v17 窗口内把破坏性声明为 minor + 正文标注 #6099(digest 只认 major 声明级别)是另一个问题:那条讲的是声明了但级别标低,这条讲的是根本没声明。两者互不覆盖。

Triage: pm:queue + domain:devx — but scoped to the objectstack half only (option C). Option A already has a dispatch entry elsewhere; do not implement it here.

1. Dedup — this problem is already tracked, and this issue is the third record of it

Keyword sweep across all three repos (open issues + PRs) found the existing chain:

issue	state	what it holds
objectstack#4904	open, `pm:blocked`, `domain:devx`, `Blocked-by: objectstack-ai/objectui#3387`	the same defect, filed 2026-08-03 off #4843, with two earlier instances (`19716b5bf`, `5e7ef1141`)
objectui#3387	open, `pm:queue`, unassigned	the actual gate work — "PR touches `packages//src/` and adds no `.changeset/.md` → fail; empty frontmatter is the explicit exemption". This is option A, already transferred to the objectui seat on 2026-08-05

So option A must not get a second dispatch entry here — one thing, one entry. What this issue adds over #4904 is genuinely new and has been pushed to where it is actionable (see §3):

the root cause — neither objectui 可以在改动已发布包源码的同时不带 changeset —— 这些修复不会出现在任何发布记录里 #4904 nor objectui#3387 mentions changeset-guard.yml at all; they read as "there is no gate", when in fact there is a workflow whose trigger structurally excludes the PRs it would need to see;
a fresh instance inside the current pin window (objectui#3518 / 0e50440, 26 files), i.e. the class is still firing after objectui-range.mjs 默认仍按 feat|fix 类型过滤 —— 发布页 Console 段落同样会丢掉破坏性 refactor! #4843 made it audible;
option C, which lands in this repo and is covered by neither of the two issues above.

2. Stale-premise check — every premise still holds on `origin/main` (checked at 07:5xZ)

objectui .github/workflows/changeset-guard.yml — trigger is still paths: ['.changeset/**'] on both pull_request and push. Last touched by 71be406; nothing since. The reporter's reading is exact, and the file's own header comment confirms the inverse trigger is deliberate.
objectstack scripts/objectui-range.mjs — --all / SHOW_EXCLUDED present (const SHOW_EXCLUDED = has('--all')), with the self-test `--all` itemizes the excluded commits by subject. The capability option C needs is indeed already written.
objectstack scripts/objectui-changeset-digest.mjs — still counts without naming (${noChangeset} commit(s) without a changeset, and the self-test commits without a changeset are counted, not dropped in silence). The gap option C describes is real and current.

3. Scope of this issue: option C only

Landing site: scripts/objectui-changeset-digest.mjs — wire the already-implemented objectui-range.mjs --all itemization into the console changeset body so a commit that skipped its changeset appears in the platform release record by subject rather than only inside a count.
Domain: scripts/ (gate class) ⇒ domain:devx per the domain table. Anchored on the landing site, not on the title vocabulary — the title says "objectui", the fix this issue keeps lands in this repo's scripts.
⛔ Out of scope here: option A (the objectui gate) — it belongs to objectui#3387. Option B is not queued; the reporter's own argument against it ("fixes this once, not the class") stands, and it is subsumed anyway once A lands.
No Blocked-by: line, deliberately: C is explicitly the measure that makes the platform side non-blind before A lands, so serialising it behind A would defeat its only purpose. The two are independent.

4. Not applying `target:v17`

Applying the release-board criterion honestly: not fixing this does not stop the current RC from shipping — the code in 0e50440 is in the console build either way; what is missing is a line in the release record. That is a completeness gap in the notes, not a user-facing defect, so it does not meet any of the four blocking classes. Recorded explicitly because it is a judgment call, and because the sibling issues of the same class (#4904, #6099) are unlabelled for v17 too — consistent, not accidental. Maintainer can veto.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

Generated by Claude Code

更正本座位 07:51Z 那条分诊评论里的一处证据 —— 结论不变,但引用的判据有一半是假读数,按纪律公开更正。

那条评论的 §2 写:「objectui .github/workflows/changeset-guard.yml … Last touched by 71be406; nothing since.」

这句话是错的。 当时本地 objectui 检出是浅克隆,git log -- <path> 把浅边界 commit 报成了该文件的最近改动。git fetch origin main --deepen=400 之后重读:

该文件的真实最近改动是 c319420(2026-08-03,fix(release): the pending release is 17.2.0, not 18.0.0 / objectui#3237);
71be406 根本不触碰这个文件 —— git show --stat 71be406 -- .github/workflows/changeset-guard.yml 零输出。

这正是 Operational notes 6 第四条点名的浅检出假读数形态,我在同一轮里对 objectstack 侧也踩了一次(7e791e5 被报成 engine.test.ts 的最近改动,--stat 显示「2528 insertions」,即整文件被当作新增)。

为什么结论不受影响: §2 里真正承载判断的三条都是内容读数(git show origin/main:<path> 读文件当前内容),不是历史读数 —— 触发器仍是 paths: ['.changeset/**']、objectui-range.mjs 的 --all/SHOW_EXCLUDED 仍在、objectui-changeset-digest.mjs 仍只计数不点名。这三条与克隆深度无关,复核后逐条仍成立。受影响的只有那句「谁最后动的、此后没人动过」的溯源。

分诊判级(pm:queue + domain:devx,范围限于选项 C)因此不变。

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

Generated by Claude Code

认领(pm-dispatch devx 座位) — 本单进入派发,范围严格限于分诊 07:51Z 钉死的选项 C(串行链:PR #6335 已于 14:5xZ 合并,objectui-changeset-digest.mjs 面锁释放)。

Session: session_01BDmDsu2575gDxeMCxXhDE3
分支: claude/issue-6174-name-changeset-less-commits
文件面: scripts/objectui-changeset-digest.mjs(把已实现的 objectui-range.mjs --all 逐条点名能力接进 console changeset 正文)。⛔ 不动 scripts/objectui-range.mjs(能力已在,只消费不改造)。
⛔ 选项 A 不在本单:objectui 侧的 changeset-required 闸门归 objectui#3387(已在该仓 pm:queue,2026-08-05 转过去的),⛔ 不得在本仓开第二个入口;⛔ 选项 B(给 objectui#3518 补追溯 changeset)分诊未入队,不做。
依据: 分诊 2026-08-07T07:51Z 判级 + 07:57Z 的自我更正(那条更正只否定了「谁最后动过该 workflow」的溯源读数,三条内容读数复核后仍成立,判级不变)。
⚠️ 同文件前序两棒刚落地,须从合并后的 main 切出并按内容定位:fix(scripts): digest 的破坏性判据扩到「声明 major 或作者正文标注」(#6099) #6289(objectui pin bump 生成的 console changeset 对破坏性变更零标注 —— digest 只认 major 声明级别,而 objectui 在 v17 窗口内把破坏性声明为 minor + 正文标注 #6099,破坏性判据扩到作者正文标注)与 fix(scripts): digest 的回落读取不再刷一屏 git fatal:,改说一句可读事实 (#6175) #6335(finding(bump-objectui): pin 范围跨过 objectui 的 release commit 时,digest 会先刷一屏 fatal: 再报成功 —— 一个完整的结果长得像一次失败 #6175,readAt 回落静默 B+C)。后者刚在 accounting 行旁加了一句回落事实——本单的点名输出与它是相邻但不同的事实(一个讲「工具怎么读到文件」,一个讲「哪些 commit 没声明 changeset」),落点与措辞都不要混。
互斥核对: 与在飞 check-single-authz-resolver 检查 (1) 的判据词表已被 ADR-0090 D3 改名废掉:sys_user_role 全仓 0 命中,门禁结构上抓不到任何重复解析器 #6286(check-single-authz-resolver.mjs)、[finding] quick-reference.mdx 的协议索引与 packages/spec 现状漂移:三处小节计数不符 + connector-auth 有 schema 无参考页 #6319(quick-reference.mdx + 可能的新校验脚本)零交集;队列中 fix(lint): object/missing-name-field 认 nameField、不再采信已退役的 titleFormat (#6108) #6338/docs(lint): 按实测改正 normalized 输入层的三条依据,并补回归 pin (#6073) #6340(packages/lint/**)无交集。

Generated by Claude Code

feat(scripts): digest 把「无 changeset 的 commit」从计数升级为按 subject 逐条点名 (ob…

Navigation Menu

Repository navigation

objectui 没有「发布性改动必须声明 changeset」闸门 —— changeset-guard.yml 的触发器决定了它只看得见已经带 changeset 的 PR(objectui#3518 因此整个从发布记录里消失) #6174

现象

根因:闸门的触发器把它自己挡在门外

需要决定的事(留给 triage,我不预设)

不在本 issue 范围内

1. Dedup — this problem is already tracked, and this issue is the third record of it

2. Stale-premise check — every premise still holds on `origin/main` (checked at 07:5xZ)

3. Scope of this issue: option C only

4. Not applying `target:v17`

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

objectui 没有「发布性改动必须声明 changeset」闸门 —— changeset-guard.yml 的触发器决定了它只看得见已经带 changeset 的 PR(objectui#3518 因此整个从发布记录里消失) #6174

Description

现象

根因:闸门的触发器把它自己挡在门外

需要决定的事(留给 triage,我不预设)

不在本 issue 范围内

Activity

claude commented on Aug 7, 2026

1. Dedup — this problem is already tracked, and this issue is the third record of it

2. Stale-premise check — every premise still holds on origin/main (checked at 07:5xZ)

3. Scope of this issue: option C only

4. Not applying target:v17

claude commented on Aug 7, 2026

hotlong commented on Aug 7, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

2. Stale-premise check — every premise still holds on `origin/main` (checked at 07:5xZ)

4. Not applying `target:v17`