Bug: timeoutafter-tests-race-real-wall-clock-deadline #516

• Work Mode: full-bug

Summary

UtilitiesCS.Test/Threading/TimeOutTask_AdditionalTests.cs contains tests that race a real 100 ms wall-clock deadline against exception propagation. On a loaded CI runner the deadline can win, so the test observes TimeoutException instead of the expected InvalidOperationException and fails. This is a flaky test, not a product defect, and it fails CI on unrelated pull requests.

Environment

• OS/version: windows-latest GitHub Actions runner (also reproducible in principle on any loaded machine)
• Runtime: .NET Framework 4.8.1, MSTest via vstest.console.exe
• Command/flags used: the CI job "Format, build, analyze, and test", which runs the MSTest suite with coverage
• Data source or fixture: UtilitiesCS.Test/Threading/TimeOutTask_AdditionalTests.cs

Steps to Reproduce

1. Open any pull request that does not touch UtilitiesCS.Test/Threading/ and let CI run the full MSTest suite.
2. Observe the run under runner load.
3. Compare against a local run of the same test on an idle machine.

Observed on 2026-08-08 in run 31275248280 for PR #514 (issue #438): 6350 total, 6349 passed, 1 failed.

Expected Behavior

The test deterministically asserts that a source exception set after TimeoutAfter is attached propagates as InvalidOperationException, independent of machine load. Per .claude/rules/general-unit-test.md, tests must be deterministic and real wall-clock waits are a banned API in test code.

Actual Behavior

TimeoutAfter_GenericTask_ShouldPropagateFaultedSourceException_WhenSourceFaultsLater failed on the CI runner with:

System.TimeoutException: The operation has timed out.

The same test passes locally on an idle machine on the same commit, confirming a load-dependent race rather than a functional regression.

The test attaches a 100 ms timeout and then sets the exception:

var source = new TaskCompletionSource<int>();
var proxy = source.Task.TimeoutAfter(100);
source.SetException(new InvalidOperationException("boom"));
Func<Task> act = async () => await proxy;
await act.Should().ThrowAsync<InvalidOperationException>().WithMessage("boom");

If the runner does not schedule the continuation within 100 ms, the timeout path completes first and the assertion sees the wrong exception type.

Logs / Screenshots

• Attached minimal logs or screenshot
• Snippet: Failed TimeoutAfter_GenericTask_ShouldPropagateFaultedSourceException_WhenSourceFaultsLater [174 ms] / Error Message: System.TimeoutException: The operation has timed out. Full log: https://github.com/drmoisan/TaskMaster/actions/runs/31275248280/job/93147648583

Note the reported duration of 174 ms against a 100 ms deadline, which is the race made visible.

Impact / Severity

• Blocker
• High
• Medium
• Low

This fails CI on pull requests that have nothing to do with the code under test, and the only available remedy today is to re-run the job. That is corrosive in two ways: it blocks unrelated work, and it normalizes re-running a red suite until it turns green - the exact habit that allows a genuine regression to be waved through. The sibling test TimeoutAfter_NonGenericTask_ShouldPropagateFaultedSourceException_WhenSourceFaultsLater has the identical structure and the same latent failure mode.

Source

From: docs/features/potential/2026-08-08-timeoutafter-tests-race-real-wall-clock-deadline.md