TSan gate: the #1144 scope-boundary probe raced nothing again on a hosted runner, after the barrier of #1489

未关闭
#2,383 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
5/5
预计耗时
一周以上
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
c, github-actions
领域
ci-cd, testing

调研方向

Start with the failing row in tests/test_tsan.sh and the fixture tests/loader_mt_userrace.eigs; inspect how the ready/release barrier coordinates the workers and how the hosted TSan job runs that fixture. Find a repeatable way to trigger the missing report, then verify the row judges the race reliably under that reproduction. The issue asks for 500 successful fixture runs in one hosted job and an updated comment describing the row’s behavior.

由索引模型根据 Issue 内容生成。

描述

area:concurrency area:gates kind:bug

The row that #1489 made deterministic has failed as vacuous once more, on phase2-numbers:

=== #1144 scope boundary: the user-level race names NO loader structure ===
  FAIL (vacuous): the probe captured NO ThreadSanitizer report (rc=0): its
        deliberate race did not run, so the boundary check examined nothing
...
Results: 75 passed, 1 failed

Run 43955204249 (the full lane, dispatched on 3450dff5, the branch's head after #2379), job tsan / x86_64 (concurrency race gate), 133607937065. tsan (concurrency race gate) fails with it. Every other row of the job passed.

The same tree passed this job an hour before: the head of #2379, de363064, has the tree of 3450dff5, and its run is 43945412729. So did 7a4dc010 before it (run 43921441839). #2379 changes src/ext_http.c and tests of the HTTP extension; tests/loader_mt_userrace.eigs uses neither.

On a machine with two cores, the fixture by itself, 40 runs on a make tsan-http build of that tree: 40 of 40 with a report. So the miss is rare, and I could not make it happen here.

tests/test_tsan.sh says of this row: "The race is deterministic now: a ready/release barrier rendezvous workers before their one-shot accesses to a separate race word, so one run suffices and no retry can hide a control that stopped racing." On that runner one run did not suffice.

Done when

  • The cause of a run with no report is named, with a reproduction that makes it happen on demand (a delay at a named place, one core, or what it turns out to be), and not by repeating until it happens.
  • On that reproduction the row has its report, or the row is changed so that what it judges does not depend on what made the report go missing.
  • Shown once in the pull request: 500 runs of the fixture in one job on a hosted runner, 500 with a report.
  • The comment in tests/test_tsan.sh says what is true of the row after the change.

— Claude DevBox

主要语言
C
星标
3
派生
7
平均合并
3 小时 28 分钟
30 天内合并 PR
270

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

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。

InauguralSystems/EigenScript 的其他 Issue

查看 InauguralSystems/EigenScript 的全部 Issue

相似的 Issue

更多 C Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。