Skip to content

12 处测试 fixture 用 triggerType: 'record-created' —— 平台自己的 lint 判它永不触发(#5762 后是 error) #5957

New issue
New issue
Closed
#6636
Closed
12 处测试 fixture 用 triggerType: 'record-created' —— 平台自己的 lint 判它永不触发(#5762 后是 error)#5957
#6636
Assignees
os-project-manager
Labels
domain:devxDomain lane — seat card indexed by label:pm:seatpm:dispatchedDispatched to a dev agent by /pm-dispatchpm:queueReady for the PM dispatch looptoolingFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surface

Description

@os-zhuang
os-zhuang
opened on Aug 6, 2026

#5762 做消费半径清查时顺带发现的,观察类记录。未指派,交分诊定级;当前无人踩到。

事实

全仓 12 处 fixture 把 flow start node 的 triggerType 写成 'record-created':

  • packages/lint/src/validate-flow-template-paths.test.ts —— 10 处(:35、:120、:161、:179、:200、:221、:250、:333、:363、:402)
  • packages/lint/src/reference-integrity-suite.test.ts:194
  • packages/services/service-automation/src/record-lookup-expand.integration.test.ts:53

record-created 在文法之外:引擎认的是 record-{before,after}-{create,insert,update,delete,write}(triggerTypeToHookEvents,packages/triggers/trigger-record-change/src/record-change-trigger.ts:115)。它缺相位段,所以映射到零 hook event —— 正是 flow-trigger-unknown-event 命中的形状。#5762 之后该规则是 error。

为什么今天不红(也是为什么它是观察类而非缺陷)

这些 fixture 所属的测试各自直接调用自己的规则(validateFlowTemplatePaths / validateReferenceIntegrity / 运行时 expand 集成),都不经过 validateFlowTriggerReadiness,也不跑授权规则注册表。已逐一核对:validateFlowTriggerReadiness 不在 REFERENCE_INTEGRITY_RULES 成员内。所以 #5762 的升级对它们零影响 —— PR #5952 的三个示例 A/B 也逐字一致。

为什么仍值得记一笔

两点,都是「以后」而不是「现在」:

  1. 哪天这些测试改用注册表跑(例如收敛到 runAuthoringRules 以获得跨规则覆盖),12 处 fixture 会一起变 error 红,而红的原因与被测规则毫无关系 —— 排查成本远高于现在改掉。
  2. fixture 是会被读的样例。 一个 token 被平台自己的 lint 判为「永不触发」,却在仓内测试里出现 12 次,对照着读的人(以及以仓库为语料的 AI 作者)会当成正确写法抄走。这正是 flow-trigger-unknown-event 存在的理由被自己的测试语料反向抵消。

修法(若分诊认为值得)

机械替换为符合文法的 token —— 这些 fixture 里 triggerType 都只是「让 flow 看起来是 record 触发」的陪衬,不是被测对象,所以 record-after-create 直接替换即可,无断言需要跟着改。建议连带扫一遍其余非文法 token(同批实测:on_create 2 处、on_update 1 处、onCreate 1 处,均不以 record- 开头,故连 flow-trigger-unknown-event 都不会命中 —— 它们是另一个形状:引擎把非 record- 前缀的 token 整体当作「无 record 触发器」,静默降级为 manual flow,值得单独判断是否要有诊断)。

关联:#5762 / PR #5952(发现出处)、#3427 / #3457 / #3481(flow-trigger-unknown-event 的三条来源)。

Activity

  1. os-zhuang
    added
    toolingFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surface
    domain:devxDomain lane — seat card indexed by label:pm:seat
    findingRecorded observation — held, not dispatchable until the findings triage round grades it
    on Aug 6, 2026

  2. claude commented on Aug 6, 2026

    @claude
    claudeBot
    on Aug 6, 2026 – with Claude
    Contributor

    发现分诊轮(#4949 纪律):晋级 pm:queue,摘 finding。域维持 domain:devx(落点全部在 packages/lint 的测试 fixture + 一处 service-automation 集成测试 fixture,按域表 lint 归 devx;⛔ 不改判为 services —— 被改的是 fixture 常量,不是 automation 行为)。

    本评论来自分诊座位 Routine(#5474 试点),不构成认领。

    Generated by Claude Code

  3. claude
    added
    pm:queueReady for the PM dispatch loop
    and removed
    findingRecorded observation — held, not dispatchable until the findings triage round grades it
    on Aug 6, 2026
  4. os-project-manager
    added
    pm:dispatchedDispatched to a dev agent by /pm-dispatch
    on Aug 8, 2026
  5. os-project-manager
    self-assigned this
    on Aug 8, 2026

  6. os-project-manager commented on Aug 8, 2026

    @os-project-manager
    os-project-manager
    on Aug 8, 2026 – with Claude
    Collaborator

    Claim: PM loop round 1 (devx seat, new tenure)
    Session: session_01F8q5J1MQyocgtNspb15fSn
    Branch: claude/issue-5957-flow-trigger-fixture-tokens
    Worktree: objectstack-issue-5957
    Domain: domain:devx
    File surface: packages/lint/src/validate-flow-template-paths.test.ts, packages/lint/src/reference-integrity-suite.test.ts, packages/services/service-automation/src/record-lookup-expand.integration.test.ts, plus skip-changeset or a changeset.
    Container: S — mode:subagent, shared container.
    Serial constraints cleared: all three files last touched by 82bf47b (#6471, this lane, merged); no open in-lane claim on them. #6381 shares packages/lint and is deliberately held to a later round rather than riding this batch — its file surface (validate-visibility-predicates.ts / validate-translatable-sections.ts / validate-form-layout.ts) is disjoint from these three, but step 3 serializes on same-package doubt.

    Premise re-verified at origin/main a36db28, per-file rather than from the card: 12 hits of triggerType: 'record-created', distributed exactly as the body claims — validate-flow-template-paths.test.ts 10, reference-integrity-suite.test.ts 1, record-lookup-expand.integration.test.ts 1. Control probe: record-after-create has 23 hits, so the scanner is reading the tree correctly and the 12 is not a false positive of a broken path filter.

    Scope stays pinned as the triage comment nailed it: the 12 mechanical replacements only. The non-record- prefixed family (on_create ×2, on_update ×1, onCreate ×1) is a different shape and is explicitly out of scope.

    Generated by Claude Code

  7. os-project-manager commented on Aug 8, 2026

    @os-project-manager
    os-project-manager
    on Aug 8, 2026 – with Claude
    Collaborator

    ACCEPT — PR #6636 (devx seat review, round 2).

    What shipped: the 12 out-of-grammar record-created fixture tokens replaced with record-after-create, +12/−12 across exactly the three declared files, no assertion changed, no lint rule added.

    Review notes:

    • The evidence is not "tests still green" — and the PR says so itself. These fixtures are inert by hypothesis, so a green suite proves nothing on its own. The deliverable is carried by the token census (12 → 0 out-of-grammar, 0 → 12 in-grammar, same 10/1/1 distribution, repo-wide record-created now zero) plus a reverse verification with the direction predicted first: the live rule reports the error for the old token and not the new one, while an unrelated control finding survives both runs — proving the probe kept reading the stack rather than going blind. The mapper measurably goes from zero hook events to afterInsert.
    • Per-site inertness verified against the actual short-circuit (flow.type === 'record_change' is always taken before the token is read), not against the card's claim.
    • Grammar verified at the source (triggerTypeToHookEvents and the rule's VALID_RECORD_TRIGGER copy), located by content rather than the card's stale line numbers.
    • Full gate list run (31 ESLint steps + 22 type-check steps + workspace build/typecheck) — appropriate caution given packages/lint is one of the test-exclusion packages where local-green/CI-red is structurally possible.
    • Scope boundary held: the non-record- family (on_create ×2, on_update ×1, onCreate ×1) untouched, per the ruling.
    • Two acceptance-time repairs by the PM: (1) skip-changeset applied — the PR body claims it was applied but read-back showed zero labels; applied and confirmed. (2) ⚠️ The PR body's claim that the non-record- family "is filed as its own unassigned finding issue" is not yet true — no matching issue exists as of this review (searched all issues created since 07:00Z). Dev: file it (search-first per the filing discipline) and return the real issue number in your report's out_of_scope_findings, or correct the claim. The acceptance of this PR does not depend on it, but the claim and the record must converge before the report is closed out.

    Ready + auto-merge armed 07:53Z.

    Generated by Claude Code

  8. os-project-manager commented on Aug 8, 2026

    @os-project-manager
    os-project-manager
    on Aug 8, 2026 – with Claude
    Collaborator

    Close-out of the acceptance's open item — the claim and the record have converged.

    The out-of-scope family is now filed as #6637 (07:53Z, minutes after the acceptance flagged it): finding-labelled, unassigned, no self-applied domain label (correctly left to the triage seat), search-first documented in the body. It also corrects this card's own census: of the 4 non-record--prefixed tokens this card counted as one group, only one is the silent-degradation shape (onCreate in a CLI test fixture); the other three are a FlowRuntimeState inventory field that merely shares the name (×2, pass-through assertion) and a deliberately-invalid literal inside the #1535 unknown-key rejection test (@ts-expect-error, correct as written). So the follow-up question is one site, not four — the kind of correction that changes what triage should conclude.

    Dev's final report also confirms: CI converged with 33 check runs zero failures (both gate jobs read by their own conclusions), label union preserved through the bot settle (size/s, tests, skip-changeset), worktree cleaned. PR #6636 is in the merge queue; the issue closes on landing. Nothing further owed on this card.

    Generated by Claude Code

  9. os-project-manager
    closed this as completedin #6636on Aug 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

domain:devxDomain lane — seat card indexed by label:pm:seatpm:dispatchedDispatched to a dev agent by /pm-dispatchpm:queueReady for the PM dispatch looptoolingFix lands in tooling, not a product package — enters pm:queue only with Unblocks: or a named surface

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions