Skip to content
This repository was archived by the owner on Nov 4, 2026. It is now read-only.

Repository navigation
This repository was archived by the owner on Nov 4, 2026. It is now read-only.

两处生成器锚规则逐名 expect、首个不匹配即抛:一批 dependabot 同时漂多个区间时只报第一个,要一轮修一个（#4098 已观察未落地） #4974

Closed
#4993
Closed
两处生成器锚规则逐名 expect、首个不匹配即抛:一批 dependabot 同时漂多个区间时只报第一个,要一轮修一个（#4098 已观察未落地）#4974
#4993
Assignees
yinlianghui
Labels
pm:dispatched

Description

@yinlianghui
yinlianghui
opened on Aug 17, 2026

Filed unassigned while implementing #4968 (PR #4969). 不在那个 PR 范围内(它是区间跟进,这是门自身的报告完整性)。

现象

全仓有两个「生成器模板区间 = 仓内区间」的 ratchet 测试:

  • packages/cli/src/__tests__/app-generator.test.ts — sources every range from this repo instead of inventing one,13 个名字(DEPENDENCY_ANCHORS)
  • packages/create-plugin/src/__tests__/templates.test.ts:356 — sources every devDependency range from this repo instead of inventing them,8 个名字(DEV_DEPENDENCY_ANCHORS)

两者都是在 for 循环里逐名 expect,而 expect 失败即抛,所以一次只报第一个不匹配的名字。表是插入序遍历的,于是「谁先漂」还取决于名字在表里的位置,而不是严重性。

代价(今天实测到的)

#4968 处理时,lucide-react 在 cli 表里排在 postcss / react / tailwindcss / typescript / vite 之前,所以它遮住了后面 6 个名字的判定 —— 门给不出「这一批 dependabot 到底漂了几个区间」这个信息。我只能另写脚本把两张表全量算一遍才敢说「各只漂一处」(结果见 PR #4969 的两张表)。

更实际的代价是跨包的:create-plugin 那处当时在 main 上也是红的(@testing-library/jest-dom ^7.0.0 vs 根 ^7.0.1,来自 #4948),但因为它在另一个包、另一个测试文件,packages/cli/ 绿之前根本没有任何东西报它。如果我只按 issue 的包边界扫,#4968 落地后全仓仍然是红的,需要第二轮。

这条一周前已被写下

#4098 的 "Suggested first step" 结尾就写了:

then re-read both test files for any other range they pin that this morning's other five bumps have already moved — shiki, maplibre-gl, react-hook-form and next all landed in the same window and only the first mismatch in each file is reported.

当时靠人工「再读一遍两个文件」补上,机制没改;七天后同一个盲区又让第二处模板藏了一轮。

建议修法

把逐名 expect 改成累积后一次断言,保留现有的信息量。cli 侧大意:

const drifted: string[] = [];
for (const [name, anchor] of Object.entries(DEPENDENCY_ANCHORS)) {
  // ... 前置断言(根里必须存在 / 仓内必须一致)保持立即抛,它们是仓库事实而非漂移
  if (generated !== expected) {
    drifted.push(`${generator} manifest's ${name}: ${generated} should be ${expected}`);
  }
}
expect(drifted).toEqual([]);

要点两条:

  1. 前置断言与漂移断言要分开。 「name must exist in the root manifest」「in-repo manifests disagree on name … settle on one range first」是关于仓库状态的前提,立即抛是对的;只有值不匹配这一类该累积。
  2. 两个文件都要改,否则跨包盲区还在 —— 这正是今天那处的成因。

顺带:两处的 anchor 表都值得在同一次里加一句注释,说明「门只判本表内的名字」,以及新增模板依赖必须同时进表(完整性门已经在管这个,但两条规则分居两个 it 里,读的时候不明显)。

分类说明

这条没有用户面影响,只影响门的报告质量与修复轮数,PM 若判为 observation-class 可加 finding 标签;我按「今天确实兑现了代价、且是第二次」当作具体缺陷提交,不自行降级。

Activity

  1. yinlianghui
    added
    pm:queue
    on Aug 17, 2026

  2. yinlianghui commented on Aug 17, 2026

    @yinlianghui
    yinlianghui
    on Aug 17, 2026 – with Claude
    ContributorAuthor

    这一族今天又活了一次:main 现在(@ 590dd6356)在两个 shard 上红。 不另立卡(#4098 是同一缺陷的上一轮,已关;本卡持有这一族的「逐名 expect、一轮修一个」观察),把实测记在这里。

    在 #3832 的 PR(#4975)分支上把 origin/main merge 进来后本地复现,与该 PR 的改动无关(它只动 ComponentInput.type 与五处声明,不碰任何 package.json / 模板):

    packages/cli/src/__tests__/app-generator.test.ts
  × sources every range from this repo instead of inventing one
    routed manifest's lucide-react must match its in-repo range:
    expected '^1.29.0' to be '^1.31.0'
  × writes a manifest whose @object-ui ranges name this CLI version
    expected '^1.29.0' to be '^1.31.0'

packages/create-plugin/src/__tests__/templates.test.ts
  × sources every devDependency range from this repo instead of inventing them
    @testing-library/jest-dom range must match the repo root:
    expected '^7.0.0' to be '^7.0.1'

    来源:今天两轮 dependabot —— #4946(patch-updates,lucide-react 1.29→1.31)与 #4948(dev-dependencies,@testing-library/jest-dom 7.0.0→7.0.1)。两轮都只移了仓内区间,没移生成器模板里的那份副本。

    同一次 merge 里第三条红是 #4911(spec 17.0.0 GA 的 $like / $ilike,packages/fields 的 FilterConditionField.operators.test.ts)—— 那张卡已经开着,main 把 @objectstack/spec 从 ^17.0.0-rc.6 推到 ^17.0.0 之后它就是活的。

    也就是说 main 今天有 三条 红、跨三个包,全部由「pin 前移、跟随面没跟」造成 —— 正好是本卡说的「一轮修一个」在同一天里撞了三次。对本卡的判据没有新增,只是又一个实测样本;顺带给后面所有基于今天 main 的 PR 一个归因,免得每个 agent 都当成自己的红重查一遍(#4975 就查了一轮)。

    Generated by Claude Code

  3. yinlianghui
    added
    pm:dispatched
    and removed
    pm:queue
    on Aug 17, 2026
  4. yinlianghui
    self-assigned this
    on Aug 17, 2026

  5. yinlianghui commented on Aug 17, 2026

    @yinlianghui
    yinlianghui
    on Aug 17, 2026 – with Claude
    ContributorAuthor

    认领(objectui 车道分片 PM,会话 session_01GTRjn8xBqp75dk7kFupVRt,批次 20 补席,替换前提证伪关卡的 #4017):branch claude/issue-4974-anchor-cumulative-report。按卡面建议实施:两处生成器锚规则(app-generator.test.ts / templates.test.ts)从「逐名 expect 首败即抛」改累积断言(一轮报全部漂移),前置断言与漂移断言分开;#4098/#4968 两次事故的「一轮修一个」形状即验收反例。

    Generated by Claude Code

  6. akarma-synetal
    added a commit that references this issue on Aug 17, 2026

    test(cli,create-plugin): 两处生成器锚规则改累积报全部漂移,前置与漂移分开 (objectstack-ai#4974)…

    cf4f8a6
  7. yinlianghui
    closed this as completedin #4993on Aug 17, 2026
  8. akarma-synetal
    added a commit that references this issue on Aug 23, 2026

    test(cli,create-plugin): 两处生成器锚规则改累积报全部漂移,前置与漂移分开 (objectstack-ai#4974)…

    ec1f4d7
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

pm:dispatched

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions