Bug: engine-toggle-prime-fault-logging-test-races

[drmoisan]

• Work Mode: minor-audit

Summary

EngineToggleStateCoordinatorTests.GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse failed once in CI and passed on rerun. The fault logging it asserts appears to race the task the test awaits.

Environment

• OS/version: windows-latest (GitHub Actions)
• Python version: n/a (C# / MSTest)
• Command/flags used: required check MSTest with coverage
• Data source or fixture: n/a

Steps to Reproduce

1. Run TaskMaster.Test under the parallel regime (Workers=0, Scope=ClassLevel).
2. Observe GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse (TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs:213). It fails intermittently.

Expected Behavior

The test is deterministic: the error log it asserts is written before the awaited task completes, or the test awaits the logging continuation itself.

Actual Behavior

It failed once on PR #939 head 9624376dc, passed on a single rerun, and passed in two local runs. PR #939 does not touch this code.

Logs / Screenshots

• Attached minimal logs or screenshot
• Snippet: CI run for PR test(931): remove scheduler and file-handle dependence from breadcrumb thread-affinity and FileInfoWrapper tests #939 at head 9624376dc (first attempt).

Impact / Severity

• Blocker
• High
• Medium
• Low

Source

From: docs/features/potential/2026-09-29-engine-toggle-prime-fault-logging-test-races.md

[drmoisan]

Restoring sections dropped by the promotion tooling.

Suspected Cause / Notes

The prime fault is probably observed and logged in a continuation that is not part of the awaited task, so the assertion can run before the log call. This is a determinism defect on a required check. Seen once on PR #939 at head 9624376dc; it passed on rerun, and PR #939 does not touch this code.

Fix it by awaiting or injecting the continuation, not by retries, sleeps, [DoNotParallelize], or Workers=1.

Proposed Fix / Validation Ideas

• Write a regression test that forces the ordering deterministically (for example a controllable scheduler or a TaskCompletionSource gate), then fix the coordinator or the test seam.
• Negative control: show that the test fails when the ordering is inverted.

[drmoisan]

Delivered by PR 946 (merge commit b305903), parallel run bugs-2026-09-28. All 14 acceptance criteria in spec.md are checked off; all seven checks green on head 0448b92 (run 36713444412) before merge. Follow-ups named in the PR body are filed separately by the run coordinator.