Gap CLI: a nondeterministic test's no-progress verdict kills, and a copy run alone has no idle detection

未关闭
#149 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
56/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
javascript
领域
testing-qa

调研方向

Start with classify.mjs (outcomeOf and cutOf) and check.mjs (countOnly, runUnit, and captureLines) to trace the two reported cases. Then inspect the existing constructed-history tests and add coverage for a nondeterministic idle copy and a batch error followed by a hanging copy. Done means the model and runtime agree on the nondeterministic verdict, and runUnit applies whole-row capture checks and idle detection.

由索引模型根据 Issue 内容生成。

描述

Two cases that the merge-gate review of #99 found (L1 and L3). #99 records both in its Limits with this issue as their owner.

1. A nondeterministic test's no-progress verdict kills (L1)

  • A test whose observations differ between the two unmutated runs is nondeterministic. Its ordinary failures judge nothing (classify.mjs outcomeOf, the nondeterministic.has check).
  • A built-in verdict skips that filter: outcomeOf adds each builtIn verdict of a cut copy without it.
  • #99 closes the spin half: a spin kills only when the test is in baseline.work (check.mjs countOnly), and that set leaves out every nondeterministic test.
  • The no-progress half stays open: a copy of a nondeterministic test that goes idle is killed by no progress.
  • The classify oracle's model makes the same prediction (cutOf gives a flaky test's runIdle the law), so it does not see the case.
  • What would close it: the model judges a nondeterministic test's built-in verdict as it judges the test's ordinary failures, and outcomeOf filters builtIn verdicts by their key. The cut rows already carry the key.

2. A copy run alone has no idle detection, and its spin row has no whole-row guard (L3)

  • A batch whose run reports an error outside any test reruns each copy alone (unitsBatch, then runAlone, then unit, then runUnit).
  • runUnit gives its state no idleMs, so a copy that hangs there runs to its backstop: a gap where R50 owes a no progress kill.
  • runUnit's spinRow splits the raw capture text and parses the spin line. It does not use captureLines' whole-row guard, which runner law L1 requires.
  • Reach is low: the path is entered only after a batch completes with an unattributed error.
  • What would close it: runUnit reads its cut rows through captureLines with partial: true, and it cuts an idle row as the batch watchdog does. Add a constructed history whose batch reports an error outside any test beside a hanging copy.
主要语言
JavaScript
星标
5
派生
1
平均合并
11 小时 18 分钟
30 天内合并 PR
66

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

TanStack/oracles 的其他 Issue

查看 TanStack/oracles 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。