Skip to content

Windows e2e: w_fastturn's work-bound kill is not-observed on win32 (sut.ts:M91, every outcome class) #114

Open

Description

@KyleAMathews

Summary

On win32, the witness mutant src/sut.ts#w_fastturn:if-false:if ( x > 2 ) { out . push ( 's' + x ) ; return ; } (sut.ts:M91, line 113) is reported as the gap not-observed. On Linux and macOS the same mutant is killed by its work bound. The oracle's model owes killed. This is a cross-platform classification difference. It is not a stream defect.

Under the old greppable-stream comparison (a count of stdout gap lines against the model's gaps), this difference showed as "the stdout has 122 gap lines for 121 gaps". With the stream law compared against results.json by id (the PR that fixes the gap-line count), the same run fails under classification and names this site.

Traces

  • win32: main at 4d3c8a9, run 43693786590, job 132749781500 (Windows gap report end-to-end oracle (fixed seed, 2/10)), unit fixed seed 20261105: every outcome class. stdout and stderr list the same 122 gap lines, with no duplicates. The extra line is oracle-gaps: gap not-observed L113 sut.ts:M91 + "x > 2 → false" + `.
  • macOS, same unit at 4d3c8a9, local: results.json holds {"id":"sut.ts:M91","outcome":"killed","by":["t36 > fixed seed 37"],"laws":["work bound"]}, and the run lists 121 gaps, the owed number.

What is not known

The cause on win32 is not identified. A kill by work bound needs four things: the two unmutated runs agree on the test's work, the route is reproducible or random, the row is not clocked, and the call count rises past the bound (classify.mjs judgeMutant, rose). One of these differs on win32. A clocked row is one candidate, since w_fastturn's tests wait a turn. This is not confirmed.

Authority

P4 (register): a counterexample is filed with its trace. "Determinism under load" applies: an outcome must not depend on the platform's timing. P3: a declared work bound is a law.

Activity

  1. KyleAMathews commented on Nov 16, 2026

    @KyleAMathews
    ContributorAuthor

    #87 has the same cause. Its list (run 43661845394, job 132652674703, fixed seed noTypescript, win32, head 104376d) has one gap line that is not in a non-Windows run of the same unit at 4d3c8a9: oracle-gaps: gap not-observed L121 sut.ts:M112 x > 2 → false ``. Here M112 is src/sut.ts#w_fastturn:if-false:…, which is `killed` by work bound (`t38 > fixed seed 39`). Every other id is the same in both runs: 169 lines on win32, 168 here, the owed count. So #87's extra line is this classification difference too. It is not a duplicated stream line: stdout and stderr list the same 169 distinct ids.

  2. KyleAMathews commented on Nov 16, 2026

    @KyleAMathews
    ContributorAuthor

    Another trace: #121's CI run 43893246349, job 133389791441 (Windows e2e 3/10). Unit fixed seed 20261105: a declaration fault: noTypescript fails with src/sut.ts:w_fastturn:121:if-false#0: not-observed, owed killed. Main fac63df's 3/10 passed in its run, so this is intermittent on win32. #121 doesn't touch w_fastturn, which has no probe code.

    This comment was written by Isaac.

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

Metadata

Metadata

Assignees

No one assigned

    Labels

    windowsWindows-specific behaviour

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions