ci(release): unblock the two gates that fail every release PR (#4894) - #4896
Merged
Conversation
Both red checks on #4422 (`chore: version packages (rc)`) are defects in the gates themselves, and both recur on every Changesets release PR. Check Changeset was structurally unsatisfiable for the release PR. The gate counts changesets a PR *adds* (`git diff --diff-filter=A` vs the base) — right for an ordinary PR, and the fix #3373 landed after a global `find | wc -l` proved unable to ever go red in RC mode. But the release PR is the *consuming* side: it applies pending changesets into versions and CHANGELOGs and adds none, by construction. Nobody labels a bot-authored PR `skip-changeset`, so the release sat blocked on a check that could only be red. `changeset-release/main` is now exempt at the job level, pinned to the bot author as well as the branch name so a hand-pushed branch of that name cannot borrow the exemption. Scaffold E2E skewed the protocol major against itself during an RC window. The install step already falls back to `latest` when the repo's version is not yet published (`@objectstack/cli@^17.0.0-rc.2` -> ETARGET -> retry as `latest`). That fallback rewrote the generated project's dependencies but not its manifest, and the template stamps the repo's protocol major (`engines: { protocol: '^17' }`, written at version time by sync-template-versions.mjs) while `latest` still pointed at 16.x. The ADR-0087 D1 handshake then correctly refused to boot the artifact — the gate working, on a skew the step had introduced: package 'e2e-app' targets protocol ^17 (engines.protocol) but this runtime is protocol 16.0.0 The fallback now re-stamps `engines.protocol` to the major actually installed, read off node_modules/@objectstack/spec (PROTOCOL_VERSION is kept in lockstep with that package's own major, asserted by protocol-version.test.ts), and logs a `::notice` so the run's true protocol is visible rather than silently rewritten. Confined to the fallback branch: on the normal path the project installs the repo's own version, the majors agree by construction, and a template stamping the wrong major must still fail — which is what template-consistency.test.ts is for. Re-stamping runs before `npm run build`, so the artifact and the Docker image (already pinned to the resolved CLI version by the same reasoning) stay in step. CI configuration only; releases nothing. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BbNVKv6KgPzuQ5p76nMgnf
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
github-actions
Bot
added
size/m
documentation
os-zhuang
marked this pull request as ready for review
August 3, 2026 13:59
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #4894。修 #4422(
chore: version packages (rc))上的两处红灯 —— 两处都是闸门自身的缺陷,不是该发布 PR 的内容问题,并且每一个发布 PR 都会重现。1.
Check Changeset对发布 PR 结构性必失败闸门数的是「本 PR 新增了几个 changeset」(
git diff --diff-filter=A对 base)。这个口径对普通 PR 是对的 —— #3373 正是因为全局find | wc -l在 RC 预发布模式下永远非空、永远不能变红才改成这样的。但 changesets 的发布 PR 是消费侧:它把待发布的 changeset 应用成版本号和 CHANGELOG,自己按定义新增数为 0。发布 PR 由
github-actions[bot]推送,不会有人去给它打skip-changeset标签,于是mergeable_state: blocked一直挂着 —— 卡在一个原理上不可能满足的检查上。改为在 job 级豁免
changeset-release/main,分支名和作者一起钉,这样手工推一个同名分支也拿不到这个逃生口。2.
Scaffold E2E / Scaffold with repo dist在 RC 窗口把协议大版本对错了install 步骤本来就有兜底:仓库版本尚未发布时(
@objectstack/cli@^17.0.0-rc.2→ ETARGET),把所有@objectstack/*改写成latest再装一次。但兜底只改依赖、没改 manifest —— 脚手架模板的objectstack.config.ts由scripts/sync-template-versions.mjs在版本时刻死了仓库自己的协议大版本engines: { protocol: '^17' },而 17 的 RC 窗口里latest仍指向 16.x。ADR-0087 D1 握手于是正确地拒绝启动:也就是说这是闸门在正常工作,拦下的是这一步自己引入的偏移。同一个 workflow 的 Docker 步骤早就认识到这一类偏移(它读取实际解析到的 CLI 版本来 pin 运行时镜像,而不是硬编码
latest),只是 boot 这条路没做对应的对齐。兜底分支现在在第二次
npm install之后,把engines.protocol重新盖成实际装上的大版本,取自node_modules/@objectstack/spec(PROTOCOL_VERSION与该包自身 major 锁步,由protocol-version.test.ts守着),并打一条::notice,让日志能看出这一轮跑的是降级后的协议版本,而不是静默改写。三点边界是刻意的:
template-consistency.test.ts的职责,这里不放松它。npm run build之前重盖,所以产物里带的是修正后的区间,后面 Docker 那条路也跟着一致。::error::退 1,不静默跳过 —— 模板哪天不再打这个戳,这里要响。验证
if:折叠成单行表达式(续行与首行同缩进,避免>-保留换行)。bash -n。templates/blank/objectstack.config.ts跑过:^17→^16(喂一个version: 16.4.2的假 spec 包),并验证了「无 stamp →::error::+ 退出码 1」这条失败路径。check:nul-bytes、check:doc-authoring本地通过。未在真实 GitHub Actions 上端到端跑过第 2 条 —— 它只在 RC 窗口且仓库版本未发布时才进入兜底分支,本 PR 触碰
.github/workflows/scaffold-e2e.yml,会命中该 workflow 的 path filter,由本 PR 的 CI 实跑给出结论。只动 CI workflow,不涉及运行时和发布产物;changeset 为空 frontmatter(releases nothing)。
🤖 Generated with Claude Code
https://claude.ai/code/session_01BbNVKv6KgPzuQ5p76nMgnf
Generated by Claude Code