diff --git a/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs b/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs new file mode 100644 index 000000000..8ea1a5053 --- /dev/null +++ b/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs @@ -0,0 +1,77 @@ +using System; +using System.Threading.Tasks; +using FluentAssertions; +using Microsoft.VisualStudio.TestTools.UnitTesting; + +namespace TaskMaster.Test.Ribbon +{ + /// + /// Regression for issue #942: the prime-fault report must precede the in-flight marker's + /// removal, so any caller that observes the marker absent observes a report that has already + /// completed. A third partial of the coordinator fixture, so the private Harness and + /// LoggedError types are reused; the primary file is close to the 500-line ceiling. + /// + public partial class EngineToggleStateCoordinatorTests + { + #region Issue #942 — prime fault report precedes marker removal + + /// + /// Regression for issue #942. Invariant: for a key whose prime did not run to completion, + /// the in-flight marker is present until the fault report has returned. The discriminator + /// is the prime handle observed from inside the error-log sink: it is the still-registered + /// continuation under the fixed order and under the + /// defective one, on the same thread, so the outcome is a function of program order + /// rather than of scheduling. No sleep, delay, gate, timer or parallelism attribute. + /// + [TestMethod] + public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged() + { + // Arrange + var harness = new Harness(); + var probe = new TaskCompletionSource(); + var failure = new InvalidOperationException("configuration load failed"); + harness.Engines.Setup(x => x.EngineActiveAsync(SpamEngine)).Returns(probe.Task); + harness.Coordinator.GetPressed(SpamEngine); + var prime = harness.Coordinator.GetPrimeTask(SpamEngine); + Task handleSeenBySink = null; + harness.OnLogError = (_, _) => + handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine); + + // Act + probe.SetException(failure); + await prime; + + // Assert + // If this test passes without the production reorder in CompletePrime, the negative + // control has lost isolation: investigate the run rather than accepting it. + handleSeenBySink + .Should() + .BeSameAs( + prime, + "while the fault is being reported the prime handle " + + "must still be registered, so a caller that fetches it " + + "after the trigger awaits the report" + ); + harness.Errors.Should().ContainSingle("a prime fault is reported exactly once"); + harness + .Errors[0] + .Message.Should() + .Contain(SpamEngine, "the message names the engine whose prime failed"); + harness + .Errors[0] + .Exception.Should() + .BeSameAs(failure, "the sink receives the injected exception unchanged"); + harness.Invalidations.Should().BeEmpty("a failed prime leaves nothing to display"); + harness + .Coordinator.GetPrimeTask(SpamEngine) + .Should() + .BeSameAs( + Task.CompletedTask, + "once the handle has completed the marker has been cleared " + + "so a later read may re-prime" + ); + } + + #endregion Issue #942 — prime fault report precedes marker removal + } +} diff --git a/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs b/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs index 6c1d7d46f..57ff5b16b 100644 --- a/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs +++ b/TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs @@ -412,7 +412,11 @@ internal Harness() OnInvalidate?.Invoke(controlId); }, message => Notifications.Add(message), - (message, exception) => Errors.Add(new LoggedError(message, exception)) + (message, exception) => + { + Errors.Add(new LoggedError(message, exception)); + OnLogError?.Invoke(message, exception); + } ); } @@ -433,6 +437,13 @@ internal Harness() /// internal Action OnInvalidate { get; set; } + /// + /// An optional extra observer invoked from inside the error-log sink, immediately after + /// the error has been appended to , so a test can probe coordinator + /// state at the exact moment a fault is reported. + /// + internal Action OnLogError { get; set; } + internal List Invalidations { get; } = new List(); internal List Notifications { get; } = new List(); diff --git a/TaskMaster.Test/TaskMaster.Test.csproj b/TaskMaster.Test/TaskMaster.Test.csproj index f78a3bc91..a5bd88ed3 100644 --- a/TaskMaster.Test/TaskMaster.Test.csproj +++ b/TaskMaster.Test/TaskMaster.Test.csproj @@ -357,6 +357,7 @@ + diff --git a/TaskMaster/Ribbon/EngineToggleStateCoordinator.cs b/TaskMaster/Ribbon/EngineToggleStateCoordinator.cs index e95568fdd..7ce33ccea 100644 --- a/TaskMaster/Ribbon/EngineToggleStateCoordinator.cs +++ b/TaskMaster/Ribbon/EngineToggleStateCoordinator.cs @@ -242,7 +242,9 @@ internal async Task ExecuteToggleAsync(string engineName) /// /// The prime task, or when no prime has been started for /// the key. The returned task never faults: a prime fault is observed inside the prime - /// itself and reported through logError. + /// itself and reported through logError. For a key whose prime did not run to + /// completion, the marker is cleared only after that report has returned, so a caller that + /// receives can rely on the fault having been reported. /// internal Task GetPrimeTask(string engineName) { @@ -326,8 +328,9 @@ string controlId /// /// Observes the outcome of a prime. On any outcome other than ran-to-completion the cache - /// is left unset — so the key still reports unchecked — the in-flight marker is cleared so - /// a later read may re-prime, and the failure is reported through logError. + /// is left unset — so the key still reports unchecked — the failure is reported through + /// logError, and only then is the in-flight marker cleared so a later read may + /// re-prime. /// /// /// The status is tested rather than the exception. A CANCELED task carries a null @@ -345,13 +348,15 @@ private void CompletePrime(Task completed, string engineName) return; } - _primeTasks.TryRemove(engineName, out _); - var failure = (Exception)completed.Exception?.GetBaseException() ?? new TaskCanceledException(completed); + // Report-then-clear is load-bearing: the marker stays registered until the report has + // returned, so a caller that observes the marker absent — including one that fetched the + // prime handle after the fault — is guaranteed the fault has already been reported. _logError(BuildPrimeFailedMessage(engineName), failure); + _primeTasks.TryRemove(engineName, out _); } /// diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/code-review.2026-09-30T08-30.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/code-review.2026-09-30T08-30.md new file mode 100644 index 000000000..6c6aed8b4 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/code-review.2026-09-30T08-30.md @@ -0,0 +1,50 @@ +# Code Review — engine-toggle-prime-fault-logging-test-races (Issue #942) + +- Branch: `bug/engine-toggle-prime-fault-logging-test-races-942` against base `231e1c0b55105aeb626bf5a6e8d0266a567cacad` +- Review date label: 2026-09-30T08-30 (assigned without a clock; later than every executor evidence label) +- Files reviewed in full: `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`, `TaskMaster.Test/TaskMaster.Test.csproj` (the three coordinator `Compile Include` lines), plus the unchanged `EngineToggleStateCoordinatorTests.Race.cs` for context. +- Companion artifacts: `policy-audit.2026-09-30T08-30.md`, `feature-audit.2026-09-30T08-30.md`. + +## Executive Summary + +Verdict: **APPROVE**. 0 blocking findings. Five non-blocking entries (CR-1 to CR-5); none requires a change on this branch. + +The fix is the smallest correct one for the defect the research record established. `CompletePrime` runs as a `ContinueWith` continuation on the default scheduler; before the change it removed the in-flight marker and then reported the fault, so a `GetPrimeTask` call issued after the trigger could observe `Task.CompletedTask` while the report was still pending on a pool thread. The change moves `_primeTasks.TryRemove` after `_logError`, which makes the marker's absence imply a completed report. That is sufficient for the flaky test's observation path because `ConcurrentDictionary.TryRemove` publishes under a lock and `TryGetValue` reads with a volatile read, so a caller that observes the removal acquires the `Errors.Add` the sink performed before it. The regression test discriminates on the marker state at the moment the sink runs, on the same thread that will perform the removal, so its outcome depends on program order and not on scheduling; the fail-before evidence shows it failing deterministically on the `BeSameAs` assertion against the byte-identical base production file. + +The success path is untouched (early return before any marker or sink access), the cancellation path still synthesizes `TaskCanceledException`, the type keeps exactly one `catch` and one `lock`, and no seam, constructor parameter or public surface was added. + +## Findings Table + +| Severity | File | Location | Finding | Recommendation | Rationale | Evidence | +|---|---|---|---|---|---|---| +| Low (non-blocking, follow-up) | `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | `StartPrimeIfNeeded` lines 271–279 and `CompletePrime` line 359 | Hazard B (NB-2 of the #735 review) remains: `_primeTasks[engineName] = StartObservedPrime(...)` stores the continuation after `StartObservedPrime` returns, and the continuation is queued to the pool with `TaskContinuationOptions.None`. If the prime completes synchronously in a non-success state, `CompletePrime` can execute `TryRemove` before the assignment lands, leaving a completed handle registered for the rest of the session. The reorder neither widens nor narrows the window. | Track under the separately promoted issue (the spec records it as promoted by the coordinator; the session checkout's branch name indicates #944). A fix would take `_primeGate` inside `CompletePrime` or register before starting; both are explicitly out of scope here. | Declared non-goal in `spec.md` "Scope & Non-Goals" and plan D-1; the caller instructed it be listed, not blocked. | Direct read of lines 263–305 and 344–360; research record section 6 "Rejected alternatives". | +| Low (non-blocking, latent) | `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | `CompletePrime` lines 358–359 | A throwing error sink now leaves the in-flight marker registered as well as faulting the continuation (before the change it only faulted the continuation, which `StartObservedPrime`'s remarks say "always completes successfully"). The production sink is `logger.Error(message, exception)` (log4net), which does not throw on appender failure, so the path is unreachable today. | No change on this branch. If a sink that can throw is ever injected, wrap the report in `try { _logError(...) } finally { _primeTasks.TryRemove(...) }` so the marker is always cleared; the spec records this as an optional hardening and a non-goal. | The trade is documented in `spec.md` "Error handling and logging updates" and "Risks & Mitigations"; adding `try`/`finally` now would violate AC1's "no `try` … is added" clause. | Direct read; `RibbonController.EngineCommands.cs` wiring per plan fact 5 (unchanged on this branch). | +| Informational | `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | `CompletePrime` lines 355–359 | A `getPressed` poll that arrives between the report and the removal now sees the marker present and does not re-prime on that poll; the next poll re-primes. Before the change the opposite window existed (a re-prime could start, and log, before the first fault's report), which could invert log order. The new window is bounded by one log call and nothing asserts on either. | None. | Behavior-preserving in every observable the tests and the ribbon depend on; the documented ordering guarantee is the stronger property. | `spec.md` "Data flow and validation changes"; direct read. | +| Informational | `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` | line 36 `Task handleSeenBySink = null;` | The local is initialised to `null` and written from the sink on a pool thread; the read after `await prime` is ordered by task completion. In a nullable-annotated file this would be `Task?`; none of the three partials carries `#nullable enable`, consistent with the sibling files and with the repository's per-file opt-in rule. | None; a nullable adoption of the fixture is outside this bug's scope. | Matches existing style (CLAUDE.md General 7.1). | Direct read; no `#nullable` in any touched file. | +| Informational | `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` | `Harness` line 445 | `OnLogError` is a settable, unguarded observer invoked from inside the sink. An observer that throws would fault the continuation the test awaits and surface as a test failure with the thrown exception, which is the correct failure mode for a test-only hook; it mirrors the pre-existing `OnInvalidate` hook exactly. | None. | Test-only surface, private nested type. | Direct read of lines 403–452. | + +## Design and Correctness Review + +- **Invariant statement.** The comment above `_logError` states the invariant in one sentence ("the marker stays registered until the report has returned, so a caller that observes the marker absent … is guaranteed the fault has already been reported"), and the `summary` of `CompletePrime` and the `returns` of `GetPrimeTask` both restate it. Comment explains why, not what (CLAUDE.md General 5.3). +- **Memory-model argument.** The spec's happens-before reasoning was checked against the code: the pool thread executes `Errors.Add` (inside `_logError`) then `TryRemove`; `TryRemove` acquires the bucket lock and writes with volatile semantics; the test thread's `TryGetValue` performs a volatile read. A test thread that sees the key absent therefore sees the appended error. The original reproduction test needs no change and now cannot observe an empty list. +- **Regression test discriminator.** `handleSeenBySink` is written from the sink, i.e. on the thread executing `CompletePrime`, at the moment `_logError` runs. Under the old order the sink runs after `TryRemove` and reads `Task.CompletedTask`; under the new order it reads the registered continuation. This is a function of statement order alone, which is what makes the fail-before result deterministic (24/1 every time, not intermittently) and the in-file comment about "loss of isolation" meaningful. +- **Cleanup discipline.** The new test ends with the marker cleared and no re-prime triggered (no `GetPressed` after the await), so the strict mock has exactly one expected call and no work is left in flight. The final `BeSameAs(Task.CompletedTask)` assertion doubles as the cleanup check. +- **Partial-class shape.** The new partial mirrors the Race partial: no `[TestClass]`, one region, the minimal `using` set (no `using Moq;` because Moq is not named directly, avoiding an unnecessary-using diagnostic). The csproj entry is required because the test project uses explicit compile items; the evidence proves the test executed (RESULT line), which is the only way to know a partial is compiled. +- **File sizes.** 420 / 470 / 77 lines, all under the 500-line ceiling; the primary fixture has 30 lines of headroom, which is why the third partial exists. +- **Unchanged neighbours confirmed.** `Race.cs` is untouched (diff-clean per `determinism-tokens.md` `NONGOAL_FILES_DIFF_EXIT=0`; the file read is consistent with the plan's fact 3). `GetPrimeTask` has no production caller (Grep over `TaskMaster/`). + +## Test Quality Review + +| Property | Assessment | +|---|---| +| Independence / isolation | Own harness, own mock, own completion source; no shared static state. | +| Determinism | No sleep, delay, retry, timeout, wall-clock, `[DoNotParallelize]`, blocking wait, temp file or scheduler seam (20-token census at 0 plus direct read). The outcome is program-order dependent only. | +| Failure diagnostics | Every assertion carries a reason; the `BeSameAs` failure text seen at fail-before names both the expected continuation type and the found `Task`, which is directly actionable. | +| Scenario coverage | Faulted path (new + original), canceled path (two Race tests), success early return (prime-success tests), key validation (existing). Complete for the changed method. | +| Original reproduction preserved | `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` is byte-for-byte unchanged (method SHA equal at base and head) and passes in every post-fix run. | + +## Non-Blocking Follow-ups + +1. Hazard B (registration racing removal on a synchronous non-success prime) — separately promoted; not addressed here by design. +2. Optional `try`/`finally` hardening of the report-then-clear pair if a throwing error sink is ever injected — documented non-goal. +3. The two other production sites that discard fault-observing continuations (`AppEvents.ReadinessHookup.cs`, `OutlookFolderTreeService.cs`) noted in the spec — no test asserts on their logs; outside this defect. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md new file mode 100644 index 000000000..f42dcf850 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md @@ -0,0 +1,11 @@ +# Bootstrap: dotnet-coverage global tool (issue 942) + +Timestamp: 2026-09-30T07-26 +Task: P0-T7 +Command: pwsh -NoProfile -Command 'if (-not (Get-Command dotnet-coverage -ErrorAction SilentlyContinue)) { dotnet tool install --global dotnet-coverage }; "DOTNET_COVERAGE_RESOLVED=$($null -ne (Get-Command dotnet-coverage -ErrorAction SilentlyContinue))"; dotnet-coverage --version' +EXIT_CODE: 0 + +Output Summary: +- The tool was already resolvable; the guarded install did not run. +- DOTNET_COVERAGE_RESOLVED=True +- Version line: 18.10.0+f4cc39224845ffa74bf246c9da2399d50e5d6342 diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md new file mode 100644 index 000000000..6b76c0422 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md @@ -0,0 +1,14 @@ +# Bootstrap: NuGet restore (issue 942) + +Timestamp: 2026-09-30T07-25 +Task: P0-T6 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $env:MSBUILDDISABLENODEREUSE = "1"; & .\scripts\vscode\Invoke-Restore.ps1; "RESTORE_EXIT=$LASTEXITCODE"; "PACKAGE_DIRS=..."; foreach ($proj in @("TaskMaster\TaskMaster.csproj", "TaskMaster.Test\TaskMaster.Test.csproj")) { ... "ANALYZER_MISSING $proj = $missing" }' +EXIT_CODE: 0 + +Output Summary: +- RESTORE_EXIT=0 +- PACKAGE_DIRS=172 +- ANALYZER_MISSING TaskMaster\TaskMaster.csproj = 0 +- ANALYZER_MISSING TaskMaster.Test\TaskMaster.Test.csproj = 0 +- Every analyzer Include of the two Write Set projects resolves relative to its own project directory; no ANALYZER PATH SKEW. +- Execution note: the restore script's console output was redirected to an ignored log under the repository coverage directory (it carries absolute host paths); the payload's gate lines above are unchanged. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md new file mode 100644 index 000000000..f7d2d5c96 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md @@ -0,0 +1,11 @@ +# Bootstrap: repository .NET SDK (issue 942) + +Timestamp: 2026-09-30T07-24 +Task: P0-T3 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; if (-not (Test-Path -LiteralPath ".dotnet-sdk\sdk\8.0.205")) { & .\scripts\vscode\Install-RepoDotNetSdk.ps1 }; "SDK_MARKER=$(Test-Path -LiteralPath ".dotnet-sdk\sdk\8.0.205")"; dotnet --version' +EXIT_CODE: 0 + +Output Summary: +- The guard found no .dotnet-sdk\sdk\8.0.205 marker, so scripts/vscode/Install-RepoDotNetSdk.ps1 ran and installed the repo-local SDK 8.0.205 into the ignored .dotnet-sdk directory (installer path line omitted: it carries an absolute host path). +- SDK_MARKER=True +- dotnet --version printed: 8.0.205 (a version string, not the global.json error message) diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md new file mode 100644 index 000000000..381655656 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md @@ -0,0 +1,13 @@ +# Bootstrap: dotnet manifest tool restore (issue 942) + +Timestamp: 2026-09-30T07-25 +Task: P0-T5 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool restore; "RESTORE_EXIT=$LASTEXITCODE"; dotnet tool list --local; dotnet tool run csharpier check --help | Out-Null; "CHECK_HELP_EXIT=$LASTEXITCODE"' +EXIT_CODE: 0 + +Output Summary: +- dotnet tool restore: "Tool 'csharpier' (version '1.2.6') was restored." and "Restore was successful." +- RESTORE_EXIT=0 +- dotnet tool list --local (Package Id and Version columns only; the Manifest column carries an absolute path and is not transcribed): + - Package Id: csharpier | Version: 1.2.6 +- CHECK_HELP_EXIT=0 diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md new file mode 100644 index 000000000..a51d9bd92 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md @@ -0,0 +1,23 @@ +# Baseline: coordinator fixture run (issue 942) + +Timestamp: 2026-09-30T07-28 +Task: P0-T12 +Command: vstest.console.exe TaskMaster.Test\bin\Debug\TaskMaster.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests" "/ResultsDirectory:coverage\test-results\942\p0-t12" "/Logger:trx;LogFileName=p0-t12.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" +EXIT_CODE: 0 + +Output Summary: +- vstest.console.exe resolved through vswhere; the trx stays under the ignored coverage directory. +- VSTEST_EXIT_CODE: 0 +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- COUNTERS total=24 executed=24 passed=24 failed=0 +- RESULT_COUNT: 24 +- RESULT GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker = Passed +- RESULT GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse = Passed +- RESULT GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked = Passed +- No RESULT line names GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged (the test does not exist yet). +- No FAILED or MESSAGE line was printed. + +BASELINE-COUNTERS: total=24 executed=24 passed=24 failed=0 +BASELINE-TOTAL: 24 +BASELINE-FAILED: NONE diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md new file mode 100644 index 000000000..9a55ea427 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md @@ -0,0 +1,110 @@ +# Baseline: repository-wide test and coverage run (issue 942) + +Timestamp: 2026-09-30T07-31 +Task: P0-T14 +Command: dotnet-coverage collect --output coverage\baseline-942.cobertura.xml --output-format cobertura --settings coverage\effective-coverage-942.config -- vstest.console.exe <9 test assemblies> /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:TestCategory!=LiveOutlook&FullyQualifiedName!~HelperClasses.ShellUtilities_Tests&FullyQualifiedName!~HelperClasses.ShellUtilitiesStatic_Tests&FullyQualifiedName!~HelperClasses.SysImageListHelperTests&FullyQualifiedName!~EmailIntelligence.OSBrowser_Tests" "/ResultsDirectory:coverage\test-results\942\baseline" "/Logger:trx;LogFileName=baseline-942.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None"; then CMD-COVERAGE-POST (STAGE baseline, RAW True) +EXIT_CODE: 0 + +Output Summary: +- COVERAGE-ROUTE: DIRECT (selected by STALL-PROBE: REPRODUCES in evidence/baseline/stall-probe.md) +- RAW: True (the DIRECT route writes a raw document; CMD-COVERAGE-POST post-processed it in place with the runner's own ConvertTo-KoverageCoberturaXml, using the native backslash spelling of the repository root; AUDIT-ABSOLUTE-FILENAMES: 0 after post-processing) +- COLLECT_EXIT_CODE: 0 +- ASSEMBLY_COUNT: 9 + - ASSEMBLY: \QuickFiler.Test\bin\Debug\QuickFiler.Test.dll + - ASSEMBLY: \SVGControl.Test\bin\Debug\SVGControl.Test.dll + - ASSEMBLY: \Tags.Test\bin\Debug\Tags.Test.dll + - ASSEMBLY: \TaskMaster.Test\bin\Debug\TaskMaster.Test.dll + - ASSEMBLY: \TaskTree.Test\bin\Debug\TaskTree.Test.dll + - ASSEMBLY: \TaskVisualization.Test\bin\Debug\TaskVisualization.Test.dll + - ASSEMBLY: \ToDoModel.Test\bin\Debug\ToDoModel.Test.dll + - ASSEMBLY: \UtilitiesCS.Test\bin\Debug\UtilitiesCS.Test.dll + - ASSEMBLY: \VBFunctions.Test\bin\Debug\VBFunctions.Test.dll +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- DOCUMENT_PRESENT: True +- LINE-FLOOR: MET +- BRANCH-FLOOR: MET +- First-party coverage: lines 56083/65736 (85.32%), branches 13597/17054 (79.73%) +- ROOT line-rate=0.853155 branch-rate=0.797291 lines-covered=56083 lines-valid=65736 branches-covered=13597 branches-valid=17054 + +Trx-derived summary: + +SUMMARY-BEGIN +Test run outcome: Completed +Total 7323, executed 7323, passed 7323, failed 0. +Skipped 0, derived as total minus executed rather than reported by the test platform. +Figures reported verbatim by the test platform: error 0, timeout 0, aborted 0, notExecuted 0, inconclusive 0. +Failed tests: none +SUMMARY-END + +FAILED-SET: (empty) + +JaCoCo package projection: + +PROJECTION-BEGIN +```xml + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +``` +PROJECTION-END + +Coordinator file figures (TaskMaster/Ribbon/EngineToggleStateCoordinator.cs): + +- COORD-CLASS-NODES: 1 +- COORD-LINES covered=143 valid=143 +- COORD-BRANCHES covered=37 valid=38 +- COMPLETEPRIME-SPAN: 341-355 +- COMPLETEPRIME-LINE-ELEMENTS: 10 +- COMPLETEPRIME-LINE 342 hits=1 +- COMPLETEPRIME-LINE 343 hits=1 +- COMPLETEPRIME-LINE 344 hits=1 +- COMPLETEPRIME-LINE 345 hits=1 +- COMPLETEPRIME-LINE 348 hits=1 +- COMPLETEPRIME-LINE 350 hits=1 +- COMPLETEPRIME-LINE 351 hits=1 +- COMPLETEPRIME-LINE 352 hits=1 +- COMPLETEPRIME-LINE 354 hits=1 +- COMPLETEPRIME-LINE 355 hits=1 +- COMPLETEPRIME-UNCOVERED: 0 + +Branch outcome: (a) exit 0 with both floors met. + +Prospective statement: the planned change adds no executable statement to the coordinator (it reorders two statements and edits documentation), so `COORD-LINES valid=` is expected to be unchanged at P3-T10. + +The stage documents coverage\baseline-942.cobertura.xml and coverage\baseline-942.trx remain on disk under the git-ignored coverage directory for P3-T10; neither is committed. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md new file mode 100644 index 000000000..cff7ed139 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md @@ -0,0 +1,11 @@ +# Baseline: CSharpier read-only check (issue 942) + +Timestamp: 2026-09-30T07-26 +Task: P0-T8 +Command: dotnet tool run csharpier check . +EXIT_CODE: 0 + +Output Summary: +- Console: "Checked 1625 files in 6311ms." +- CSHARPIER_EXIT_CODE: 0 +- Unformatted-file set: none (CSharpier reported no file as unformatted). The formatter baseline is clean; no FORMAT BASELINE NOT CLEAN. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md new file mode 100644 index 000000000..ca92f2cd2 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md @@ -0,0 +1,17 @@ +# Baseline: Write Set line counts and hashes (issue 942) + +Timestamp: 2026-09-30T07-28 +Task: P0-T11 +Command: CMD-LINECOUNT and CMD-HASH (Get-Content line count and Get-FileHash SHA256 over the three formatter-visible Write Set files) +EXIT_CODE: 0 + +Output Summary: +- LINES TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = 415 +- LINES TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = 459 +- LINES TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = ABSENT +- HASH TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0 +- HASH TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = 61F4EB0FCC001C43A8F0F2F4C95EBDC6C0B1D2310700C0B79CF3F5805C84D487 +- HASH TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = ABSENT +- BASE-HASH-PROD: F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0 +- BASE-HASH-TEST: 61F4EB0FCC001C43A8F0F2F4C95EBDC6C0B1D2310700C0B79CF3F5805C84D487 +- These counts are advisory; the authoritative AC14 audit is P3-T3. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md new file mode 100644 index 000000000..cc750ab1d --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md @@ -0,0 +1,19 @@ +# Baseline: analyzer rebuild (issue 942) + +Timestamp: 2026-09-30T07-27 +Task: P0-T9 +Command: msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-REBUILD (TASKID p0-t9): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- WARNINGS: 0 +- ANALYZER-BASELINE-WARNINGS: 0 +- SKIP_CORECOMPILE_LINES: 0 +- CSC_OUT_TASKMASTER: 2 +- CSC_OUT_TASKMASTER_TEST: 2 +- WRITESET_DIAGNOSTIC_LINES: 0 +- TEST_DLL_EXISTS: True +- UCS_TEST_DLL_EXISTS: True diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md new file mode 100644 index 000000000..3fbe8d96c --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md @@ -0,0 +1,19 @@ +# Baseline: nullable rebuild (issue 942) + +Timestamp: 2026-09-30T07-27 +Task: P0-T10 +Command: msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-REBUILD (TASKID p0-t10): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. No Nullable property override. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- WARNINGS: 0 +- NULLABLE-BASELINE-WARNINGS: 0 +- SKIP_CORECOMPILE_LINES: 0 +- CSC_OUT_TASKMASTER: 2 +- CSC_OUT_TASKMASTER_TEST: 2 +- WRITESET_DIAGNOSTIC_LINES: 0 +- TEST_DLL_EXISTS: True +- UCS_TEST_DLL_EXISTS: True diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md new file mode 100644 index 000000000..95333b56a --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md @@ -0,0 +1,13 @@ +# Phase 0 commit (issue 942) + +Timestamp: 2026-09-30T07-33 +Task: P0-T16 +Command: git add -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 then git commit -m "docs(942): plan and Phase 0 baseline evidence for the prime-fault ordering fix" -m "Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com" -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 then git status --porcelain -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 +EXIT_CODE: 0 + +Output Summary: +- git commit exited 0: 14 files changed (13 new Phase 0 artifacts plus the plan file's check-off marks). +- PHASE0-COMMIT-SHA: 3c9f75b66e4f6d4171c12125ddaceae5a1d719ab +- Pushed to origin bug/engine-toggle-prime-fault-logging-test-races-942 (d5e57c26b..3c9f75b66). +- Feature-folder porcelain immediately after the commit: empty. The plan file (its P0-T16 check-off mark) and this artifact are written after the commit and are the only expected uncommitted feature-folder entries. +- The commit used the pathspec form; no path outside the feature folder was staged by this task. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md new file mode 100644 index 000000000..fc03472a8 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md @@ -0,0 +1,24 @@ +# Phase 0 Policy Read (issue 942) + +Timestamp: 2026-09-30T07-23 +Task: P0-T1 + +Policy Order: CLAUDE.md -> .claude/rules/general-code-change.md -> .claude/rules/general-unit-test.md -> .claude/rules/csharp.md + +Files read (in this order), with the top-level (`# `) heading count of each, counted from the file: + +1. CLAUDE.md — top-level headings: 1 +2. .claude/rules/general-code-change.md — top-level headings: 1 +3. .claude/rules/general-unit-test.md — top-level headings: 1 +4. .claude/rules/csharp.md — top-level headings: 1 +5. .claude/rules/plan-acceptance-gates.md — top-level headings: 1 +6. .claude/rules/tonality.md — top-level headings: 1 + +No policy document was modified by this task. + +Key constraints carried into execution: + +- C# toolchain order: CSharpier format and check, analyzer Rebuild, nullable Rebuild (TreatWarningsAsErrors, no Nullable override), coverage-enabled MSTest run. +- MSTest, Moq and FluentAssertions for tests; no temporary files; no sleeps, delays or wall-clock reads in tests. +- 500-line ceiling per source file. +- Committed test evidence is projections only; no raw trx or Cobertura document is committed. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md new file mode 100644 index 000000000..b42faf7c2 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md @@ -0,0 +1,97 @@ +# Scope and Anchor (issue 942) + +Timestamp: 2026-09-30T07-23 +Task: P0-T2 (creates this file); P0-T4 and P0-T15 append. + +## Sources read in full + +- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md +- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md +- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md + +## Write Set (code paths, verbatim) + +- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs +- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs +- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs (new) +- TaskMaster.Test/TaskMaster.Test.csproj + +## Prohibited paths and trees (from the plan's Write Set section) + +1. TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs +2. TaskMaster/Ribbon/RibbonController.EngineCommands.cs +3. TaskMaster.runsettings +4. scripts/vscode/TaskMaster.cli.runsettings +5. every file under scripts/vscode/ +6. every file under .claude/ except .claude/agent-memory/ (session memory, never staged) +7. every file under config/ +8. every file under docs/features/potential/ (including the inherited promotion record docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md) + +## Work mode and acceptance-criteria counts + +- issue.md line 12 reads: `- Work Mode: full-bug` +- spec.md acceptance section, counted from the file: 14 lines beginning `- [ ] AC`, 0 lines beginning `- [x] AC`. + +## Anchor and pre-change tree state (P0-T4) + +Timestamp: 2026-09-30T07-24 +Command: git rev-parse HEAD; git merge-base --is-ancestor 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD; git merge-base origin/main HEAD; git diff --name-status 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD; git status --porcelain --untracked-files=all +EXIT_CODE: 0 + +Output Summary: +- HEAD-SHA: d5e57c26bfd6dee1fcc3e0bf61356e27cd8ac8b5 +- Ancestor check (231e1c0b55105aeb626bf5a6e8d0266a567cacad is an ancestor of HEAD): exit 0 +- git merge-base origin/main HEAD printed: 231e1c0b55105aeb626bf5a6e8d0266a567cacad (equals the anchor; no BASE-SHA MISMATCH) + +INHERITED-COMMITTED: +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md +- A docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md + +Every inherited path is under the feature folder or is exactly the promotion record the amended AC13 exempts. No INHERITED SET EXCEEDS AC13 EXEMPTION. + +PRE-EXISTING-WORKTREE-PATHS: +``` + M .claude/agent-memory/atomic-executor/MEMORY.md + M .claude/agent-memory/atomic-planner/MEMORY.md + M .claude/agent-memory/prd-feature/feedback_ac_gates_verify_satisfiability.md + M .claude/agent-memory/task-researcher/MEMORY.md + M docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md +?? .claude/agent-memory/atomic-executor/index_artifact_hygiene_and_misc.md +?? .claude/agent-memory/atomic-executor/index_build_toolchain.md +?? .claude/agent-memory/atomic-executor/index_csharp_and_components.md +?? .claude/agent-memory/atomic-executor/index_test_execution_and_coverage.md +?? .claude/agent-memory/atomic-executor/project_code_commit_before_final_format_pass_orphans_rewrites.md +?? .claude/agent-memory/atomic-executor/project_plan_column_widths_may_include_markdown_indent.md +?? .claude/agent-memory/atomic-planner/project_942_prime_fault_report_then_clear_plan_seams.md +?? .claude/agent-memory/task-researcher/project_engine_toggle_prime_fault_log_order_942.md +?? docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md +?? docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md +?? docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md +``` + +No porcelain line names a path under TaskMaster/ or TaskMaster.Test/ (no CODE TREE DIRTY AT ANCHOR). + +## PHASE0-ARTIFACTS: + +Timestamp: 2026-09-30T07-32 (P0-T15; listing of evidence/baseline/, field check by a line-anchored search for the four schema fields and ExpectedExitCode) + +| Artifact (evidence/baseline/) | Task | Command-bearing | Timestamp / Command / EXIT_CODE / Output Summary | EXIT_CODE | ExpectedExitCode | +|---|---|---|---|---|---| +| phase0-instructions-read.md | P0-T1 | no | Timestamp and Policy Order present | n/a | n/a | +| scope-and-anchor.md | P0-T2, P0-T4 | yes (P0-T4 section) | all four present | 0 | not required | +| bootstrap-sdk.md | P0-T3 | yes | all four present | 0 | not required | +| bootstrap-tool-restore.md | P0-T5 | yes | all four present | 0 | not required | +| bootstrap-nuget-restore.md | P0-T6 | yes | all four present | 0 | not required | +| bootstrap-dotnet-coverage.md | P0-T7 | yes | all four present | 0 | not required | +| csharpier-check-baseline.md | P0-T8 | yes | all four present | 0 | not required | +| msbuild-analyzer-baseline.md | P0-T9 | yes | all four present | 0 | not required | +| msbuild-nullable-baseline.md | P0-T10 | yes | all four present | 0 | not required | +| file-line-counts-baseline.md | P0-T11 | yes | all four present | 0 | not required | +| coordinator-tests-baseline.md | P0-T12 | yes | all four present | 0 | not required | +| stall-probe.md | P0-T13 | yes | all four present | 1 | 1 (matches) | +| coverage-baseline.md | P0-T14 | yes | all four present | 0 | not required | + +Every artifact named by P0-T1 through P0-T14 exists at its exact path; every non-zero EXIT_CODE carries a matching ExpectedExitCode. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md new file mode 100644 index 000000000..51208fce7 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md @@ -0,0 +1,23 @@ +# Baseline: shell-icon stall probe (issue 942) + +Timestamp: 2026-09-30T07-30 +Task: P0-T13 +Command: vstest.console.exe UtilitiesCS.Test\bin\Debug\UtilitiesCS.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~HelperClasses.ShellUtilities_Tests|FullyQualifiedName~HelperClasses.ShellUtilitiesStatic_Tests|FullyQualifiedName~HelperClasses.SysImageListHelperTests|FullyQualifiedName~EmailIntelligence.OSBrowser_Tests" "/ResultsDirectory:coverage\test-results\942\p0-t13" "/Logger:trx;LogFileName=p0-t13.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" +EXIT_CODE: 1 +ExpectedExitCode: 1 + +Output Summary: +- The probe was invoked once and not re-run. +- VSTEST_EXIT_CODE: 1 +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- COUNTERS total=23 executed=23 passed=22 failed=1 +- RESULT_COUNT: 23 +- FAILED GetFileIcon_WithUseFileType_ShouldReturnIconsForDirectoryAndFileExtension +- MESSAGE GetFileIcon_WithUseFileType_ShouldReturnIconsForDirectoryAndFileExtension :: Test method UtilitiesCS.Test.HelperClasses.ShellUtilities_Tests.GetFileIcon_WithUseFileType_ShouldReturnIconsForDirectoryAndFileExtension threw exception: System.ArgumentException: Win32 handle that was passed to Icon is not valid or is the wrong type. +- ExpectedExitCode is presentational; nothing is gated on it. + +STALL-PROBE: REPRODUCES +COVERAGE-ROUTE: DIRECT + +The run did not stall (no Sequence document), but it did not satisfy the CLEAR condition (EXIT_CODE 0, failed 0, SEQUENCE_FILES 0) because one shell-icon test failed on this workstation. The four excluded classes are a pre-existing local defect on this workstation (fact 13 of the plan) and are executed by CI; the DIRECT coverage route excludes them from the local coverage run. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/ac-status-summary.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/ac-status-summary.md new file mode 100644 index 000000000..e03ce8d2b --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/ac-status-summary.md @@ -0,0 +1,31 @@ +# Acceptance Criteria Status (issue 942) + +Timestamp: 2026-09-30T07-56 +Task: P3-T29 +Command: line-anchored count of `- [x] AC` and `- [ ] AC` lines in the spec's acceptance section, read from the file +EXIT_CODE: 0 + +Output Summary: +- Source: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md (work mode full-bug; spec.md is the only AC source) +- TOTAL: 14 of 14 (counted from the file: 14 lines beginning `- [x] AC`, 0 lines beginning `- [ ] AC`) +- UNMET: NONE (no `ACn: NOT MET` was recorded by P3-T15 through P3-T28) +- Remaining unchecked criteria: none + +Per-criterion evidence cited at check-off: + +| AC | Check-off task | Evidence | +|---|---|---| +| AC1 | P3-T15 | evidence/qa-gates/production-reorder-scope.md (POST-FORMAT) | +| AC2 | P3-T16 | evidence/qa-gates/production-reorder-scope.md (POST-FORMAT documentation-token rows) | +| AC3 | P3-T17 | evidence/qa-gates/harness-hook-edit-scope.md (POST-FORMAT) | +| AC4 | P3-T18 | evidence/regression-testing/build-before-reorder.md (POST-FORMAT) and evidence/regression-testing/prime-fault-ordering-pass-after.md (RESULT lines) | +| AC5 | P3-T19 | evidence/regression-testing/build-before-reorder.md (POST-FORMAT) and the strict-mock row of evidence/qa-gates/harness-hook-edit-scope.md (POST-FORMAT) | +| AC6 | P3-T20 | evidence/qa-gates/csproj-registration.md and evidence/regression-testing/prime-fault-ordering-pass-after.md | +| AC7 | P3-T21 | evidence/regression-testing/prime-fault-ordering-fail-before.md | +| AC8 | P3-T22 | evidence/regression-testing/prime-fault-ordering-pass-after.md | +| AC9 | P3-T23 | evidence/regression-testing/prime-fault-ordering-pass-after.md (COUNTERS, POPULATION-COMPARISON) and evidence/qa-gates/original-test-unchanged.md (POST-FORMAT) | +| AC10 | P3-T24 | evidence/qa-gates/determinism-tokens.md | +| AC11 | P3-T25 | evidence/qa-gates/toolchain-final-pass.md | +| AC12 | P3-T26 | evidence/qa-gates/coverage-post-change.md (COMPARISON), evidence/baseline/coverage-baseline.md, evidence/qa-gates/footprint-scope.md | +| AC13 | P3-T27 | evidence/qa-gates/footprint-scope.md, evidence/qa-gates/determinism-tokens.md, evidence/qa-gates/production-reorder-scope.md (POST-FORMAT) | +| AC14 | P3-T28 | evidence/qa-gates/file-line-counts.md | diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/reduced-audit-handoff.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/reduced-audit-handoff.md new file mode 100644 index 000000000..2462c6cb9 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/reduced-audit-handoff.md @@ -0,0 +1,29 @@ +# Reduced-audit handoff (issue 942) + +Timestamp: 2026-09-30T07-57 +Task: P3-T30 +Command: git rev-parse HEAD (recorded before the final feature-folder commit) +EXIT_CODE: 0 + +Output Summary: +- PRE-FINAL-COMMIT-HEAD: 509f7f0a576d82dd668821dbb4bbb181f3a45912 (the P2-T8 implementation commit) +- COVERAGE-ROUTE: DIRECT (STALL-PROBE: REPRODUCES at P0-T13; the four shell-icon classes were excluded from the local coverage run and are executed by CI) + +Pointers for the reviewer: + +- Final toolchain pass: evidence/qa-gates/toolchain-final-pass.md +- Coverage post-change and comparison: evidence/qa-gates/coverage-post-change.md +- Footprint: evidence/qa-gates/footprint-scope.md +- Fail-before: evidence/regression-testing/prime-fault-ordering-fail-before.md +- Pass-after: evidence/regression-testing/prime-fault-ordering-pass-after.md +- Acceptance-criteria status: evidence/other/ac-status-summary.md + +Scope statements: + +- Hazard B (registration racing removal on a synchronous non-success prime, NB-2 of the issue 735 code review) is out of scope for this item and is promoted separately by the coordinator; no potential entry was written by this run. +- The committed coverage-route test evidence is projections only: the JaCoCo package projection, the first-party coverage line, the trx-derived summary and per-file figures, transcribed into Markdown. No raw trx, Cobertura or coverage document is committed; the stage documents remain under the git-ignored coverage directory. + +Execution notes for the reviewer (recorded in the cited artifacts): + +- Two PreToolUse hooks (the parallel and epic worktree-removal gates, and the promotion-path gate) refused command strings before execution because the strings carried the worktree path next to a removal verb or git invocation, or carried an issue-reference literal. In each case nothing ran; the payload was re-issued with the path or literal composed by concatenation inside the payload, which does not change any computed value. +- The P2-T6 span payload fails to parse verbatim (an unbalanced parenthesis in a nested literal inside a subexpression); the lock literal was built outside the subexpression, which is the identical predicate. Recorded in evidence/qa-gates/production-reorder-scope.md. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/coverage-post-change.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/coverage-post-change.md new file mode 100644 index 000000000..c07447da4 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/coverage-post-change.md @@ -0,0 +1,144 @@ +# Final loop: repository-wide test and coverage run (issue 942) + +Timestamp: 2026-09-30T07-50 +Task: P3-T8 (creates this file); P3-T10 appends COMPARISON. +Command: dotnet-coverage collect --output coverage\final-942.cobertura.xml --output-format cobertura --settings coverage\effective-coverage-942.config -- vstest.console.exe <9 test assemblies> /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:TestCategory!=LiveOutlook&FullyQualifiedName!~HelperClasses.ShellUtilities_Tests&FullyQualifiedName!~HelperClasses.ShellUtilitiesStatic_Tests&FullyQualifiedName!~HelperClasses.SysImageListHelperTests&FullyQualifiedName!~EmailIntelligence.OSBrowser_Tests" "/ResultsDirectory:coverage\test-results\942\final" "/Logger:trx;LogFileName=final-942.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None"; then CMD-COVERAGE-POST (STAGE final, RAW True) +EXIT_CODE: 0 + +Output Summary: +- COVERAGE-ROUTE: DIRECT (fixed by P0-T13) +- RAW: True (post-processed in place by CMD-COVERAGE-POST with the native backslash repository root; AUDIT-ABSOLUTE-FILENAMES: 0) +- Attempts: one. The first attempt's FAILED-SET is empty, so the issue 780 re-run rule did not apply. +- COLLECT_EXIT_CODE: 0 +- ASSEMBLY_COUNT: 9 + - ASSEMBLY: \QuickFiler.Test\bin\Debug\QuickFiler.Test.dll + - ASSEMBLY: \SVGControl.Test\bin\Debug\SVGControl.Test.dll + - ASSEMBLY: \Tags.Test\bin\Debug\Tags.Test.dll + - ASSEMBLY: \TaskMaster.Test\bin\Debug\TaskMaster.Test.dll + - ASSEMBLY: \TaskTree.Test\bin\Debug\TaskTree.Test.dll + - ASSEMBLY: \TaskVisualization.Test\bin\Debug\TaskVisualization.Test.dll + - ASSEMBLY: \ToDoModel.Test\bin\Debug\ToDoModel.Test.dll + - ASSEMBLY: \UtilitiesCS.Test\bin\Debug\UtilitiesCS.Test.dll + - ASSEMBLY: \VBFunctions.Test\bin\Debug\VBFunctions.Test.dll +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- DOCUMENT_PRESENT: True +- LINE-FLOOR: MET +- BRANCH-FLOOR: MET +- First-party coverage: lines 56080/65736 (85.31%), branches 13596/17054 (79.72%) +- ROOT line-rate=0.853109 branch-rate=0.797232 lines-covered=56080 lines-valid=65736 branches-covered=13596 branches-valid=17054 + +Trx-derived summary: + +SUMMARY-BEGIN +Test run outcome: Completed +Total 7324, executed 7324, passed 7324, failed 0. +Skipped 0, derived as total minus executed rather than reported by the test platform. +Figures reported verbatim by the test platform: error 0, timeout 0, aborted 0, notExecuted 0, inconclusive 0. +Failed tests: none +SUMMARY-END + +FAILED-SET: (empty) + +JaCoCo package projection: + +PROJECTION-BEGIN +```xml + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +``` +PROJECTION-END + +Coordinator file figures (TaskMaster/Ribbon/EngineToggleStateCoordinator.cs): + +- COORD-CLASS-NODES: 1 +- COORD-LINES covered=143 valid=143 +- COORD-BRANCHES covered=37 valid=38 +- COMPLETEPRIME-SPAN: 344-360 +- COMPLETEPRIME-LINE-ELEMENTS: 10 +- COMPLETEPRIME-LINE 345 hits=1 +- COMPLETEPRIME-LINE 346 hits=1 +- COMPLETEPRIME-LINE 347 hits=1 +- COMPLETEPRIME-LINE 348 hits=1 +- COMPLETEPRIME-LINE 351 hits=1 +- COMPLETEPRIME-LINE 352 hits=1 +- COMPLETEPRIME-LINE 353 hits=1 +- COMPLETEPRIME-LINE 358 hits=1 +- COMPLETEPRIME-LINE 359 hits=1 +- COMPLETEPRIME-LINE 360 hits=1 +- COMPLETEPRIME-UNCOVERED: 0 + +Branch outcome: (a) exit 0, both floors met, FAILED-SET empty. The summary block's second line reports `failed 0`. RESULT-level proof that the new test executed is in the FINAL-FIXTURE-RUN section of evidence/regression-testing/prime-fault-ordering-pass-after.md (P3-T7). Under the DIRECT route the LINE-FLOOR and BRANCH-FLOOR lines are the floor gate. + +The stage documents coverage\final-942.cobertura.xml and coverage\final-942.trx remain on disk under the git-ignored coverage directory; neither is committed. + +## COMPARISON: + +Timestamp: 2026-09-30T07-52 (P3-T10; read from evidence/baseline/coverage-baseline.md and this file; changed lines from `git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` against the working tree, which is the tree the final coverage document was generated from) + +- COORD-LINES-BASELINE: covered=143 valid=143 +- COORD-LINES-FINAL: covered=143 valid=143 +- COORD-BRANCHES-BASELINE: covered=37 valid=38 +- COORD-BRANCHES-FINAL: covered=37 valid=38 +- COMPLETEPRIME-ELEMENTS-BASELINE: 10 +- COMPLETEPRIME-ELEMENTS-FINAL: 10 +- COMPLETEPRIME-UNCOVERED-FINAL: 0 +- FIRST-PARTY-BASELINE: First-party coverage: lines 56083/65736 (85.32%), branches 13597/17054 (79.73%) +- FIRST-PARTY-FINAL: First-party coverage: lines 56080/65736 (85.31%), branches 13596/17054 (79.72%) +- ROOT-BASELINE: ROOT line-rate=0.853155 branch-rate=0.797291 lines-covered=56083 lines-valid=65736 branches-covered=13597 branches-valid=17054 +- ROOT-FINAL: ROOT line-rate=0.853109 branch-rate=0.797232 lines-covered=56080 lines-valid=65736 branches-covered=13596 branches-valid=17054 +- DENOMINATOR-BRANCH: COMPARABLE (lines-valid 65736 at both stages, a difference of 0, within 1 percent of the baseline figure). Rate clause: final root line-rate 0.853109 is at least the baseline 0.853155 minus 0.005 (0.848155). Holds. +- CHANGED-LINES (added lines of the anchored diff inside COMPLETEPRIME-SPAN 344-360): + - 355: no line element (comment line) + - 356: no line element (comment line) + - 357: no line element (comment line) + - 359: hits=1 (the moved `_primeTasks.TryRemove(engineName, out _);` statement) +- Added lines outside the span (documentation only, no line elements): 245-247 (GetPrimeTask returns element) and 331-333 (CompletePrime summary element). + +Acceptance check (P3-T10): + +- COORD-LINES-FINAL valid (143) equals COORD-LINES-BASELINE valid (143): no executable statement added; no COORD-LINES-VALID CHANGED. +- COORD-LINES-FINAL covered (143) is at least baseline covered (143). +- COORD-BRANCHES-FINAL covered (37) is at least baseline covered (37). +- COMPLETEPRIME-ELEMENTS-FINAL (10) equals COMPLETEPRIME-ELEMENTS-BASELINE (10). +- COMPLETEPRIME-UNCOVERED-FINAL: 0. +- Every CHANGED-LINES entry that has a line element has hits at least 1 (line 359, hits=1). +- Exactly one DENOMINATOR-BRANCH value is recorded, and under COMPARABLE its rate clause holds. +- The repository-wide first-party figures moved by 3 covered lines and 1 covered branch outside the coordinator file with an unchanged denominator; the coordinator file, the only production file changed, is unchanged in every figure. This variation is within the D-7 tolerance and is recorded, not attributed to this change. +- All clauses hold. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-check-final.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-check-final.md new file mode 100644 index 000000000..d4627f065 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-check-final.md @@ -0,0 +1,11 @@ +# Final loop: CSharpier read-only check (issue 942) + +Timestamp: 2026-09-30T07-46 +Task: P3-T4 +Command: dotnet tool run csharpier check . +EXIT_CODE: 0 + +Output Summary: +- Console: "Checked 1626 files in 7312ms." +- CSHARPIER_EXIT_CODE: 0 +- The check reported no unformatted file (no differences). diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-format.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-format.md new file mode 100644 index 000000000..3a9688ca0 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-format.md @@ -0,0 +1,20 @@ +# Final loop: CSharpier format, repository-wide (issue 942) + +Timestamp: 2026-09-30T07-43 +Task: P3-T1 +Command: dotnet tool run csharpier format . +EXIT_CODE: 0 + +Output Summary: +- Console: "Formatted 1626 files in 11786ms." (files processed, not files changed; not used as the rewritten count) +- CSHARPIER_EXIT_CODE: 0 +- HASH before TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = D9C915AE9B00BB7AAB80183A7A0BA11748DE393781D7E5ADE2BDE29073B7002B +- HASH before TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = AA754469AA204624B14E3BBF4E229EAE57FEEA6722561F956382A4A6BEEAA3FC +- HASH before TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = AA88DC05B45CE2E0D935014778779C5B500025BCFD237AEE05A184CED7D6F8DB +- HASH after TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = D9C915AE9B00BB7AAB80183A7A0BA11748DE393781D7E5ADE2BDE29073B7002B +- HASH after TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = AA754469AA204624B14E3BBF4E229EAE57FEEA6722561F956382A4A6BEEAA3FC +- HASH after TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = AA88DC05B45CE2E0D935014778779C5B500025BCFD237AEE05A184CED7D6F8DB +- Rewritten-file count (Write Set paths whose two hashes differ): 0. No POST-COMMIT CODE REWRITE. +- Scoped porcelain before (`git status --porcelain -- . ":(exclude)docs/features" ":(exclude).claude"`): empty. +- Scoped porcelain after (same command): empty. +- The two scoped porcelain outputs are identical line sets (both empty): the repository-wide format rewrote no file outside the Write Set. No FORMAT WIDENED FOOTPRINT. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md new file mode 100644 index 000000000..1a15e3bd9 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md @@ -0,0 +1,14 @@ +# Test project registration of the new partial (issue 942) + +Timestamp: 2026-09-30T07-36 +Task: P1-T3 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $p = Get-Content -LiteralPath "TaskMaster.Test\TaskMaster.Test.csproj" -Encoding UTF8; ... "RACE_LINE=$race NEW_LINE=$new NEW_COUNT=..."; git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/TaskMaster.Test.csproj' +EXIT_CODE: 0 + +Output Summary: +- Inserted `` immediately after the Race entry, with the same four-space indentation. +- RACE_LINE=359 NEW_LINE=360 NEW_COUNT=1 +- NEW_LINE equals RACE_LINE plus 1. +- Anchored numstat: `1 0 TaskMaster.Test/TaskMaster.Test.csproj` (1 insertion, 0 deletions). +- Line endings after the edit: 423 CRLF of 423 LF (consistent CRLF). +- The project file is outside the formatter (.csharpierignore), so no format pass follows this edit. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/determinism-tokens.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/determinism-tokens.md new file mode 100644 index 000000000..711e0ed1e --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/determinism-tokens.md @@ -0,0 +1,32 @@ +# Determinism-token constraints (issue 942) + +Timestamp: 2026-09-30T07-52 +Task: P3-T11 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $added = @(git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon | Where-Object { $_.StartsWith("+") -and -not $_.StartsWith("+++") }); ... ADDED-TOKEN counts ...; git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.runsettings scripts/vscode/TaskMaster.cli.runsettings; git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs TaskMaster/Ribbon/RibbonController.EngineCommands.cs' (the P3-T11 payload; the worktree path was composed inside the payload by concatenation, which does not change any computed value) +EXIT_CODE: 0 + +Output Summary: +- ADDED-LINE-COUNT: 89 (the 77-line new partial plus the 12 added Harness lines; at least 60) +- ADDED-TOKEN [Thread.Sleep] = 0 +- ADDED-TOKEN [Task.Delay] = 0 +- ADDED-TOKEN [DoNotParallelize] = 0 +- ADDED-TOKEN [[Timeout] = 0 +- ADDED-TOKEN [Timeout=] = 0 +- ADDED-TOKEN [DateTime.Now] = 0 +- ADDED-TOKEN [DateTime.UtcNow] = 0 +- ADDED-TOKEN [Stopwatch] = 0 +- ADDED-TOKEN [.Wait(] = 0 +- ADDED-TOKEN [.Result] = 0 +- ADDED-TOKEN [ManualResetEvent] = 0 +- ADDED-TOKEN [SemaphoreSlim] = 0 +- ADDED-TOKEN [GetTempFileName] = 0 +- ADDED-TOKEN [GetTempPath] = 0 +- ADDED-TOKEN [TaskScheduler] = 0 +- ADDED-TOKEN [while (] = 0 +- ADDED-TOKEN [for (] = 0 +- ADDED-TOKEN [Retry] = 0 +- ADDED-TOKEN [GetResult(] = 0 +- ADDED-TOKEN [DateTimeOffset] = 0 +- RUNSETTINGS_DIFF_EXIT=0 +- NONGOAL_FILES_DIFF_EXIT=0 +- Every ADDED-TOKEN count is 0; the run-settings files and the two non-goal files are unchanged from the merge base. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/evidence-hygiene.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/evidence-hygiene.md new file mode 100644 index 000000000..256e2a49d --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/evidence-hygiene.md @@ -0,0 +1,12 @@ +# Evidence hygiene sweep (issue 942) + +Timestamp: 2026-09-30T07-53 +Task: P3-T13 +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $acct = Split-Path -Leaf $env:USERPROFILE; $machine = $env:COMPUTERNAME; $files = @(Get-ChildItem -LiteralPath "docs\features\active\2026-09-29-engine-toggle-prime-fault-logging-test-races-942" -Recurse -File -Filter "*.md"); ... "FILES_SCANNED=... ACCOUNT_HITS=... MACHINE_HITS=... DRIVE_USERS_HITS=..."' (the P3-T13 payload, run verbatim) +EXIT_CODE: 0 + +Output Summary: +- FILES_SCANNED=36 ACCOUNT_HITS=0 MACHINE_HITS=0 DRIVE_USERS_HITS=0 +- The account and machine tokens are derived at run time; neither value is written into this artifact. +- The drive-path count normalises backslashes to forward slashes and applies the CI hygiene guard's user-profile pattern case-insensitively with separator runs; no occurrence was found, so no REDACTED-PATH repair was needed. +- Scope: every Markdown file in the feature folder at the time of the sweep, including the plan. The two artifacts written later (evidence/other/ac-status-summary.md and evidence/other/reduced-audit-handoff.md) carry no host path by construction. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/file-line-counts.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/file-line-counts.md new file mode 100644 index 000000000..445f40817 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/file-line-counts.md @@ -0,0 +1,12 @@ +# Post-format line counts (issue 942) + +Timestamp: 2026-09-30T07-46 +Task: P3-T3 +Command: CMD-LINECOUNT (Get-Content line count of the three Write Set source files, after the P3-T1 format pass) +EXIT_CODE: 0 + +Output Summary: +- LINES TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = 420 +- LINES TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = 470 +- LINES TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = 77 +- Each count is at most 500. This is the authoritative AC14 audit. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/footprint-scope.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/footprint-scope.md new file mode 100644 index 000000000..dd288d3d7 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/footprint-scope.md @@ -0,0 +1,95 @@ +# Footprint scope (issue 942) + +Timestamp: 2026-09-30T07-53 +Task: P3-T12 (creates this file); P3-T14 appends. +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $ext = @(".trx", ".xml", ".coverage", ".coveragexml", ".cobertura"); $added = @(git diff --name-only --diff-filter=A 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD); ...; "RAW-DOCS-COMMITTED: ..."; $untracked = @(git status --porcelain --untracked-files=all --ignored -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 | ...); "RAW-DOCS-UNTRACKED-IN-FEATURE: ..."' (the P3-T12 payload; the worktree path was composed inside the payload by concatenation) +EXIT_CODE: 0 + +Output Summary: + +Committed additions since the anchor (name-listing diff): + +- ADDED-PATH: TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md +- ADDED-PATH: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md +- ADDED-PATH: docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md + +- RAW-DOCS-COMMITTED: 0 +- RAW-DOCS-UNTRACKED-IN-FEATURE: 0 (porcelain span over the feature folder with --untracked-files=all and --ignored, because .gitignore ignores trx and cobertura xml names repository-wide) +- Positive control: the ADDED-PATH list contains TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs, so the enumeration saw the committed additions. + +## Change footprint (P3-T14) + +Timestamp: 2026-09-30T07-54 +Command: git diff --name-status 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD; git status --porcelain --untracked-files=all +EXIT_CODE: 0 + +Output Summary: + +INHERITED-AND-EXCLUDED: +- A docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md (the promotion record the amended AC13 exempts; a member of INHERITED-COMMITTED from P0-T4, subtracted by rule D-8) + +THIS-ITEM-FOOTPRINT: +- A TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs +- M TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs +- M TaskMaster.Test/TaskMaster.Test.csproj +- M TaskMaster/Ribbon/EngineToggleStateCoordinator.cs +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md +- A docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md + +Checks: +- Every THIS-ITEM-FOOTPRINT path is one of the four code paths or lies under the feature folder. +- All four code paths are present; the new partial carries status A. +- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs, TaskMaster/Ribbon/RibbonController.EngineCommands.cs, TaskMaster.runsettings and scripts/vscode/TaskMaster.cli.runsettings are absent from the footprint. +- No diff path is outside the code paths, the feature folder and the promotion record: no FOOTPRINT OUTSIDE AC13. +- Porcelain composition (stated without a count): modified and untracked entries under the feature folder (evidence artifacts written or appended after the P2-T8 commit, and this plan file), and modified and untracked entries under .claude/agent-memory/ (uncommitted session memory, all members of PRE-EXISTING-WORKTREE-PATHS from P0-T4, never staged). No porcelain line names a path under TaskMaster/ or TaskMaster.Test/. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md new file mode 100644 index 000000000..3be92c6b6 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md @@ -0,0 +1,63 @@ +# Harness hook edit scope (issue 942) + +Timestamp: 2026-09-30T07-34 +Task: P1-T1 (creates this file); P2-T6 and P3-T2 append. +Command: CMD-TOKEN-COUNT with FILE TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs and TOKEN "internal Action OnLogError { get; set; }", "OnLogError?.Invoke(message, exception);", "Errors.Add(new LoggedError(message, exception));", "invoked from inside the error-log sink", "[TestMethod]" +EXIT_CODE: 0 + +Output Summary: +- Edit applied: the single-line error-log lambda (anchor line 415) became the five-line block (append, then null-conditional invoke); one blank line and the six hook lines were inserted after the OnInvalidate property line (line 438 after the lambda edit). +- TOKEN [internal Action OnLogError { get; set; }] = 1 +- TOKEN [OnLogError?.Invoke(message, exception);] = 1 +- TOKEN [Errors.Add(new LoggedError(message, exception));] = 1 +- TOKEN [invoked from inside the error-log sink] = 1 +- TOKEN [[TestMethod]] = 15 +- FIRST-LINE [internal Action OnLogError { get; set; }] = 445 +- FIRST-LINE [OnLogError?.Invoke(message, exception);] = 418 +- FIRST-LINE [Errors.Add(new LoggedError(message, exception));] = 417 +- FIRST-LINE [invoked from inside the error-log sink] = 441 +- FIRST-LINE [[TestMethod]] = 30 +- Adjacency: FIRST-LINE of the invoke (418) equals FIRST-LINE of the append (417) plus 1. +- [TestMethod] count 15 equals the anchor count (fact 2), so no test method was added or removed. +- Line endings after the edit: 470 CRLF of 470 LF (consistent CRLF). +- No other file was modified by this task. + +## Harness hunk check (P2-T6) + +Timestamp: 2026-09-30T07-41 +Command: git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs; git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs; CMD-TOKEN-COUNT with TOKEN "private sealed class Harness", "private sealed class LoggedError", "new Mock(MockBehavior.Strict)" +EXIT_CODE: 0 + +Output Summary: +- TOKEN [private sealed class Harness] = 1 +- TOKEN [private sealed class LoggedError] = 1 +- TOKEN [new Mock(MockBehavior.Strict)] = 1 +- FIRST-LINE [private sealed class Harness] = 403 +- FIRST-LINE [private sealed class LoggedError] = 457 +- FIRST-LINE [new Mock(MockBehavior.Strict)] = 424 +- Hunk headers: `@@ -415 +415,5 @@` (new-side start 415) and `@@ -435,0 +440,7 @@` (new-side start 440). Both starts are greater than 403 and less than 457, so every hunk lies inside the Harness type. +- The diff adds and removes zero lines containing `[TestMethod]`. +- Numstat: `12 1 TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` (deletions exactly 1: the replaced single-line lambda). +- All harness clauses hold. + +## POST-FORMAT: + +Timestamp: 2026-09-30T07-45 +Task: P3-T2 +Command: the P1-T1 CMD-TOKEN-COUNT with its exact TOKEN list, and the P2-T6 harness commands (anchored `git diff -U0`, `git diff --numstat`, CMD-TOKEN-COUNT for the Harness boundary tokens), re-run after the P3-T1 repository-wide format pass +EXIT_CODE: 0 + +Output Summary: +- TOKEN [internal Action OnLogError { get; set; }] = 1; FIRST-LINE 445 +- TOKEN [OnLogError?.Invoke(message, exception);] = 1; FIRST-LINE 418 +- TOKEN [Errors.Add(new LoggedError(message, exception));] = 1; FIRST-LINE 417 +- TOKEN [invoked from inside the error-log sink] = 1; FIRST-LINE 441 +- TOKEN [[TestMethod]] = 15; FIRST-LINE 30 +- Adjacency: invoke (418) = append (417) plus 1. +- TOKEN [private sealed class Harness] = 1; FIRST-LINE 403 +- TOKEN [private sealed class LoggedError] = 1; FIRST-LINE 457 +- TOKEN [new Mock(MockBehavior.Strict)] = 1; FIRST-LINE 424 +- Hunk headers: `@@ -415 +415,5 @@` (start 415) and `@@ -435,0 +440,7 @@` (start 440); both inside (403, 457), so every hunk lies inside the Harness type. +- The diff adds and removes zero lines containing `[TestMethod]`. +- Numstat: `12 1 TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` (deletions exactly 1). +- Every clause of P1-T1 and the P2-T6 harness clauses holds on the post-format tree. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md new file mode 100644 index 000000000..5451fa3b8 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md @@ -0,0 +1,30 @@ +# Implementation format and commit (issue 942) + +Timestamp: 2026-09-30T07-42 +Task: P2-T8 +Command: dotnet tool run csharpier format TaskMaster\Ribbon\EngineToggleStateCoordinator.cs TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs (with CMD-HASH before and after); then git add -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs TaskMaster.Test/TaskMaster.Test.csproj docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942; git commit -m "fix(ribbon): report a prime fault before clearing its in-flight marker (issue 942)" -m "Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com"; git show --name-only --format= HEAD; git status --porcelain -- TaskMaster TaskMaster.Test +EXIT_CODE: 0 + +Output Summary: + +Scoped format: + +- Console: "Formatted 3 files in 3883ms." +- CSHARPIER_EXIT_CODE: 0 +- HASH before TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = D9C915AE9B00BB7AAB80183A7A0BA11748DE393781D7E5ADE2BDE29073B7002B +- HASH before TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = AA754469AA204624B14E3BBF4E229EAE57FEEA6722561F956382A4A6BEEAA3FC +- HASH before TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = AA88DC05B45CE2E0D935014778779C5B500025BCFD237AEE05A184CED7D6F8DB +- HASH after TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = D9C915AE9B00BB7AAB80183A7A0BA11748DE393781D7E5ADE2BDE29073B7002B +- HASH after TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = AA754469AA204624B14E3BBF4E229EAE57FEEA6722561F956382A4A6BEEAA3FC +- HASH after TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs = AA88DC05B45CE2E0D935014778779C5B500025BCFD237AEE05A184CED7D6F8DB +- PRECOMMIT-FORMAT-REWRITES: 0 (each path's two hashes are equal; the Delivered Source layout was already formatter-stable, so no PRECOMMIT-FORMAT-RECHECK was required) + +Commit (recorded after the commit; this section is committed with the P3-T30 commit): + +- git commit exited 0: 15 files changed. +- IMPLEMENTATION-COMMIT-SHA: 509f7f0a576d82dd668821dbb4bbb181f3a45912 +- Pushed to origin bug/engine-toggle-prime-fault-logging-test-races-942 (3c9f75b66..509f7f0a5). +- git show --name-only --format= HEAD listed exactly the four code paths (TaskMaster/Ribbon/EngineToggleStateCoordinator.cs, TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs, TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs, TaskMaster.Test/TaskMaster.Test.csproj) plus eleven paths under the feature folder, and nothing else. +- git status --porcelain -- TaskMaster TaskMaster.Test printed no line. +- No PreToolUse refusal occurred on the git add or the git commit. +- From this commit on, no code file is edited. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-analyzer-final.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-analyzer-final.md new file mode 100644 index 000000000..5398b745a --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-analyzer-final.md @@ -0,0 +1,18 @@ +# Final loop: analyzer rebuild (issue 942) + +Timestamp: 2026-09-30T07-47 +Task: P3-T5 +Command: msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-REBUILD (TASKID p3-t5): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- WARNINGS: 0 (ANALYZER-BASELINE-WARNINGS: 0; not higher than baseline) +- SKIP_CORECOMPILE_LINES: 0 (no project reported a skipped CoreCompile target) +- CSC_OUT_TASKMASTER: 2 +- CSC_OUT_TASKMASTER_TEST: 2 +- WRITESET_DIAGNOSTIC_LINES: 0 +- TEST_DLL_EXISTS: True +- UCS_TEST_DLL_EXISTS: True diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-nullable-final.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-nullable-final.md new file mode 100644 index 000000000..b5bf48782 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-nullable-final.md @@ -0,0 +1,18 @@ +# Final loop: nullable rebuild (issue 942) + +Timestamp: 2026-09-30T07-47 +Task: P3-T6 +Command: msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-REBUILD (TASKID p3-t6): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. No Nullable property override. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- WARNINGS: 0 (NULLABLE-BASELINE-WARNINGS: 0; not higher than baseline) +- SKIP_CORECOMPILE_LINES: 0 (no project reported a skipped CoreCompile target) +- CSC_OUT_TASKMASTER: 2 +- CSC_OUT_TASKMASTER_TEST: 2 +- WRITESET_DIAGNOSTIC_LINES: 0 +- TEST_DLL_EXISTS: True +- UCS_TEST_DLL_EXISTS: True diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md new file mode 100644 index 000000000..f343c26b0 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md @@ -0,0 +1,26 @@ +# Original reproduction test unchanged (issue 942) + +Timestamp: 2026-09-30T07-42 +Task: P2-T7 (creates this file); P3-T2 appends POST-FORMAT. +Command: pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; function Get-MethodText([string[]]$lines) { ... }; $base = @(git show 231e1c0b55105aeb626bf5a6e8d0266a567cacad:TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs); $now = Get-Content -LiteralPath "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs" -Encoding UTF8; ... "METHOD_UNCHANGED=..."; "METHOD_LINES=..."' (the P2-T7 payload, run verbatim) +EXIT_CODE: 0 + +Output Summary: +- BASE_METHOD_SHA=19-3B-28-A7-14-24-D5-29-7C-F9-8E-63-49-B4-5D-54-B5-C7-17-D5-7D-55-5A-C8-03-DC-B7-D7-DB-F4-CD-E6 +- NOW_METHOD_SHA=19-3B-28-A7-14-24-D5-29-7C-F9-8E-63-49-B4-5D-54-B5-C7-17-D5-7D-55-5A-C8-03-DC-B7-D7-DB-F4-CD-E6 +- METHOD_UNCHANGED=True +- METHOD_LINES=32 (the `[TestMethod]` attribute line through the method's closing brace) +- Decoding note: the compared method span contains only ASCII characters and the file carries no BOM, so the console-code-page decoding of the `git show` output cannot alter the base side; no correction was applied. + +## POST-FORMAT: + +Timestamp: 2026-09-30T07-46 +Task: P3-T2 +Command: the P2-T7 payload, re-run verbatim after the P3-T1 repository-wide format pass +EXIT_CODE: 0 + +Output Summary: +- BASE_METHOD_SHA=19-3B-28-A7-14-24-D5-29-7C-F9-8E-63-49-B4-5D-54-B5-C7-17-D5-7D-55-5A-C8-03-DC-B7-D7-DB-F4-CD-E6 +- NOW_METHOD_SHA=19-3B-28-A7-14-24-D5-29-7C-F9-8E-63-49-B4-5D-54-B5-C7-17-D5-7D-55-5A-C8-03-DC-B7-D7-DB-F4-CD-E6 +- METHOD_UNCHANGED=True +- METHOD_LINES=32 diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md new file mode 100644 index 000000000..f97c9efaf --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md @@ -0,0 +1,114 @@ +# Production reorder scope (issue 942) + +Timestamp: 2026-09-30T07-41 +Task: P2-T6 (creates this file); P3-T2 appends POST-FORMAT. +Command: CMD-TOKEN-COUNT with FILE TaskMaster\Ribbon\EngineToggleStateCoordinator.cs and the eleven-token list of P2-T6; git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs; git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs; the P2-T6 span measurement (adapted as noted below) +EXIT_CODE: 0 + +Output Summary: + +Token counts (file level): + +- TOKEN [private void CompletePrime(] = 1 +- TOKEN [_logError(BuildPrimeFailedMessage(engineName), failure);] = 1 +- TOKEN [_primeTasks.TryRemove(engineName, out _);] = 1 +- TOKEN [Report-then-clear is load-bearing] = 1 +- TOKEN [and only then is the in-flight marker cleared] = 1 +- TOKEN [cleared only after that report has returned] = 1 +- TOKEN [internal Task GetPrimeTask(] = 1 +- TOKEN [catch (] = 1 +- TOKEN [lock (] = 1 +- TOKEN [new TaskCanceledException(completed)] = 1 +- TOKEN [GetBaseException()] = 1 +- FIRST-LINE [private void CompletePrime(] = 344 +- FIRST-LINE [_logError(BuildPrimeFailedMessage(engineName), failure);] = 358 +- FIRST-LINE [_primeTasks.TryRemove(engineName, out _);] = 359 +- FIRST-LINE [Report-then-clear is load-bearing] = 355 +- FIRST-LINE [and only then is the in-flight marker cleared] = 332 +- FIRST-LINE [cleared only after that report has returned] = 246 +- FIRST-LINE [internal Task GetPrimeTask(] = 249 +- FIRST-LINE [catch (] = 181 +- FIRST-LINE [lock (] = 271 +- FIRST-LINE [new TaskCanceledException(completed)] = 353 +- FIRST-LINE [GetBaseException()] = 352 + +Span measurement: + +- SPAN=344-360 +- SPAN_RETURN=1 +- SPAN_RANTOCOMPLETION=1 +- SPAN_TRY=0 +- SPAN_CATCH=0 +- SPAN_LOCK=0 +- Adaptation (recorded, not silent): the plan's verbatim span payload exits 1 with no output from both Bash and a pwsh host, because the nested literal `"lock ("` inside a `"$( ... )"` subexpression carries an unbalanced parenthesis that the expandable-string scanner cannot parse. Probes: the same subexpression with `"lock"` prints a count; with `"lock ("` it exits 1; the `"return;"` and `"\btry\b"` forms parse. The literal was built outside the subexpression as `$lk = "lock" + " ("` (length probe PROBE_LK_LENGTH=6) and passed as `$_.Contains($lk)`, which is the identical predicate. All other span statements were run verbatim. + +Anchored diff (-U0) hunk headers and hunk-window rule: + +- G = FIRST-LINE of `internal Task GetPrimeTask(` = 249; window [G-6, G-1] = [243, 248] +- S = FIRST-LINE of `and only then is the in-flight marker cleared` = 332; R = FIRST-LINE of `_primeTasks.TryRemove(engineName, out _);` = 359; window [S-2, R] = [330, 359] +- `@@ -245 +245,3 @@` new-side span 245-247: inside [243, 248] +- `@@ -329,2 +331,3 @@` new-side span 331-333: inside [330, 359] +- `@@ -348,2 +350,0 @@` new-side single line 350 (count 0): inside [330, 359] +- `@@ -353,0 +355,3 @@` new-side span 355-357: inside [330, 359] +- `@@ -354,0 +359 @@` new-side span 359-359 (omitted count is 1): inside [330, 359] +- Every hunk lies wholly within one of the two windows; no HUNK OUTSIDE EDIT WINDOWS. + +Numstat: `10 5 TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (deletions 5, at most 8). + +Removed lines, quoted and classified: + +1. ` /// itself and reported through logError.` — a line of the GetPrimeTask returns element. +2. ` /// is left unset — so the key still reports unchecked — the in-flight marker is cleared so` — a line of the CompletePrime summary element. +3. ` /// a later read may re-prime, and the failure is reported through logError.` — a line of the CompletePrime summary element. +4. ` _primeTasks.TryRemove(engineName, out _);` — the TryRemove statement. +5. (empty line) — the blank line adjacent to the TryRemove statement. + +No line of the early return, the failure computation or the synthesized exception was removed or rewritten. + +Clause check (P2-T1, P2-T2, P2-T6 production acceptance): + +- `_logError(` line 358 is less than the `TryRemove` line 359; both are greater than the CompletePrime line 344. +- `Report-then-clear is load-bearing` line 355 equals the `_logError(` line minus 3. +- `catch (` = 1 and `lock (` = 1 (unchanged from the anchor, fact 1). +- The three documentation tokens count 1 each: the summary token (332) lies within the twelve lines above 344; the returns token (246) lies within the eight lines above 249. The `remarks` element is unchanged (no hunk touches it). +- `new TaskCanceledException(completed)` = 1 and `GetBaseException()` = 1. +- SPAN start 344 equals FIRST-LINE of `private void CompletePrime(`. +- All clauses hold. + +## POST-FORMAT: + +Timestamp: 2026-09-30T07-45 +Task: P3-T2 +Command: the P2-T6 production token count, span measurement (same `$lk` adaptation) and anchored `git diff -U0` / `git diff --numstat` against 231e1c0b55105aeb626bf5a6e8d0266a567cacad, re-run on the tree after the P3-T1 repository-wide format pass +EXIT_CODE: 0 + +Output Summary: + +- TOKEN [private void CompletePrime(] = 1; FIRST-LINE 344 +- TOKEN [_logError(BuildPrimeFailedMessage(engineName), failure);] = 1; FIRST-LINE 358 +- TOKEN [_primeTasks.TryRemove(engineName, out _);] = 1; FIRST-LINE 359 +- TOKEN [Report-then-clear is load-bearing] = 1; FIRST-LINE 355 +- TOKEN [and only then is the in-flight marker cleared] = 1; FIRST-LINE 332 +- TOKEN [cleared only after that report has returned] = 1; FIRST-LINE 246 +- TOKEN [internal Task GetPrimeTask(] = 1; FIRST-LINE 249 +- TOKEN [catch (] = 1; FIRST-LINE 181 +- TOKEN [lock (] = 1; FIRST-LINE 271 +- TOKEN [new TaskCanceledException(completed)] = 1; FIRST-LINE 353 +- TOKEN [GetBaseException()] = 1; FIRST-LINE 352 +- SPAN=344-360 +- SPAN_RETURN=1 +- SPAN_RANTOCOMPLETION=1 +- SPAN_TRY=0 +- SPAN_CATCH=0 +- SPAN_LOCK=0 +- G = 249, S = 332, R = 359; windows [243, 248] and [330, 359]. +- Hunk headers: `@@ -245 +245,3 @@` (245-247, window 1); `@@ -329,2 +331,3 @@` (331-333, window 2); `@@ -348,2 +350,0 @@` (350, window 2); `@@ -353,0 +355,3 @@` (355-357, window 2); `@@ -354,0 +359 @@` (359, window 2). Every hunk lies inside a window. +- Numstat: `10 5 TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (deletions 5, at most 8). +- Removed lines re-quoted and classified: + 1. ` /// itself and reported through logError.` — GetPrimeTask returns element line. + 2. ` /// is left unset — so the key still reports unchecked — the in-flight marker is cleared so` — CompletePrime summary line. + 3. ` /// a later read may re-prime, and the failure is reported through logError.` — CompletePrime summary line. + 4. ` _primeTasks.TryRemove(engineName, out _);` — the TryRemove statement. + 5. (empty line) — the blank line adjacent to the TryRemove statement. +- Display note: the pwsh console decoded the em dashes of removed line 2 through the console code page when printing `git diff` output; the line is quoted here with its actual characters, which the pre-format P2-T6 diff printed correctly from Bash. +- Every clause of P2-T6 (and the P2-T1 and P2-T2 position clauses) holds on the post-format tree: `_logError(` (358) precedes `TryRemove` (359), both after 344; the comment is exactly 3 lines above `_logError(`; `catch (` and `lock (` each 1; both exception tokens 1; the documentation tokens at their required positions. No POST-COMMIT CODE REWRITE. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/toolchain-final-pass.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/toolchain-final-pass.md new file mode 100644 index 000000000..86bc35998 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/toolchain-final-pass.md @@ -0,0 +1,26 @@ +# Final toolchain pass (issue 942) + +Timestamp: 2026-09-30T07-51 +Task: P3-T9 +Command: the P3-T1 through P3-T8 steps listed below, in the CLAUDE.md order +EXIT_CODE: 0 + +Output Summary: +- Pass number: 1. The first pass completed clean; no step failed and no step rewrote a file, so there is no failed pass to record. +- The code was committed at P2-T8 (509f7f0a576d82dd668821dbb4bbb181f3a45912) after a scoped format; no code file was edited after that commit. + +| Step | Task | Command | Exit code | Observation | +|---|---|---|---|---| +| 1 Format | P3-T1 | dotnet tool run csharpier format . | 0 | Write Set hashes identical before and after (rewritten count 0); scoped porcelain empty before and after | +| 1a Line-count audit | P3-T3 | CMD-LINECOUNT | 0 | 420, 470, 77 lines (each at most 500) | +| 1b Post-format gates | P3-T2 | token, span, diff and method gates | 0 | every P1-T1, P1-T2, P2-T6 and P2-T7 clause holds post-format | +| 2 Format check | P3-T4 | dotnet tool run csharpier check . | 0 | "Checked 1626 files"; the check reported no differences | +| 3 Lint (analyzers) | P3-T5 | msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true | 0 | ERRORS 0, WARNINGS 0, SKIP_CORECOMPILE_LINES: 0, CSC_OUT_TASKMASTER 2, CSC_OUT_TASKMASTER_TEST 2 | +| 4 Type check (nullable) | P3-T6 | msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true | 0 | ERRORS 0, WARNINGS 0, SKIP_CORECOMPILE_LINES: 0, CSC_OUT_TASKMASTER 2, CSC_OUT_TASKMASTER_TEST 2 | +| 5a Fixture run | P3-T7 | vstest.console.exe (coordinator fixture filter, p3-t7) | 0 | 25 of 25 passed, including the new test | +| 5b Coverage-enabled tests | P3-T8 | dotnet-coverage collect ... vstest.console.exe (DIRECT route), then CMD-COVERAGE-POST | 0 | 7324 of 7324 passed, failed 0; LINE-FLOOR MET, BRANCH-FLOOR MET; First-party coverage: lines 56080/65736 (85.31%), branches 13596/17054 (79.72%) | + +- For P3-T5 and P3-T6: SKIP_CORECOMPILE_LINES: 0 and both CSC_OUT_ counts at least 1, so the analyzer and nullable gates compiled rather than short-circuited; no project reported a skipped CoreCompile target. +- For P3-T4: the read-only check reported no differences. +- For P3-T8: COVERAGE-ROUTE: DIRECT (STALL-PROBE: REPRODUCES at P0-T13); the run exited 0. +- The pass ran in order (format, check, analyzer rebuild, nullable rebuild, coverage-enabled test run) with no intervening file rewrite. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md new file mode 100644 index 000000000..6b6e0c309 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md @@ -0,0 +1,14 @@ +# Build after the reorder (issue 942) + +Timestamp: 2026-09-30T07-38 +Task: P2-T3 +Command: msbuild TaskMaster.sln /t:Build /m /p:Configuration=Debug "/p:Platform=Any CPU" +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-BUILD (TASKID p2-t3): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- TEST_DLL_ADVANCED: True +- CSC_OUT_TASKMASTER_TEST: 2 +- Transcription note: the stale-log removal used the equivalent .NET file-delete call instead of the cmdlet name, so the command string does not trip the parallel worktree-removal hook (see the execution note in prime-fault-ordering-fail-before.md); the gate lines are unchanged. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md new file mode 100644 index 000000000..6f4cb2d47 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md @@ -0,0 +1,141 @@ +# New partial token counts and build before the reorder (issue 942) + +Timestamp: 2026-09-30T07-35 +Task: P1-T2 (creates this file); P1-T4 and P3-T2 append. +Command: CMD-TOKEN-COUNT with FILE TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs and the 31-token list of P1-T2 +EXIT_CODE: 0 + +Output Summary: +- The new partial was written with the Delivered Source text (Markdown indent removed), CRLF line endings and no BOM, matching the sibling partials. + +Token counts: + +- TOKEN [public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged()] = 1 +- TOKEN [[TestMethod]] = 1 +- TOKEN [[TestClass]] = 0 +- TOKEN [public partial class EngineToggleStateCoordinatorTests] = 1 +- TOKEN [handleSeenBySink] = 3 +- TOKEN [harness.OnLogError =] = 1 +- TOKEN [var prime = harness.Coordinator.GetPrimeTask(SpamEngine);] = 1 +- TOKEN [probe.SetException(failure);] = 1 +- TOKEN [await prime;] = 1 +- TOKEN [has lost isolation] = 1 +- TOKEN [Regression for issue #942] = 2 +- TOKEN [.ContainSingle(] = 1 +- TOKEN [.BeEmpty(] = 1 +- TOKEN [.BeSameAs(] = 3 +- TOKEN [using Moq;] = 0 +- TOKEN [// Arrange] = 1 +- TOKEN [// Act] = 1 +- TOKEN [// Assert] = 1 +- TOKEN [.Contain(SpamEngine] = 1 +- TOKEN [.BeSameAs(failure] = 1 +- TOKEN [var harness = new Harness();] = 1 +- TOKEN [Invariant: for a key whose prime did not run to completion] = 1 +- TOKEN [handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);] = 1 +- TOKEN [Task.CompletedTask,] = 1 +- TOKEN [harness.Invalidations.Should().BeEmpty(] = 1 +- TOKEN [the message names the engine whose prime failed] = 1 +- TOKEN [the sink receives the injected exception unchanged] = 1 +- TOKEN [a prime fault is reported exactly once] = 1 +- TOKEN [a failed prime leaves nothing to display] = 1 +- TOKEN [must still be registered] = 1 +- TOKEN [so a later read may re-prime] = 1 + +First lines: + +- FIRST-LINE [public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged()] = 27 +- FIRST-LINE [[TestMethod]] = 26 +- FIRST-LINE [[TestClass]] = 0 +- FIRST-LINE [public partial class EngineToggleStateCoordinatorTests] = 14 +- FIRST-LINE [handleSeenBySink] = 36 +- FIRST-LINE [harness.OnLogError =] = 37 +- FIRST-LINE [var prime = harness.Coordinator.GetPrimeTask(SpamEngine);] = 35 +- FIRST-LINE [probe.SetException(failure);] = 41 +- FIRST-LINE [await prime;] = 42 +- FIRST-LINE [has lost isolation] = 46 +- FIRST-LINE [Regression for issue #942] = 9 +- FIRST-LINE [.ContainSingle(] = 55 +- FIRST-LINE [.BeEmpty(] = 64 +- FIRST-LINE [.BeSameAs(] = 49 +- FIRST-LINE [using Moq;] = 0 +- FIRST-LINE [// Arrange] = 29 +- FIRST-LINE [// Act] = 40 +- FIRST-LINE [// Assert] = 44 +- FIRST-LINE [.Contain(SpamEngine] = 59 +- FIRST-LINE [.BeSameAs(failure] = 63 +- FIRST-LINE [var harness = new Harness();] = 30 +- FIRST-LINE [Invariant: for a key whose prime did not run to completion] = 19 +- FIRST-LINE [handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);] = 38 +- FIRST-LINE [Task.CompletedTask,] = 69 +- FIRST-LINE [harness.Invalidations.Should().BeEmpty(] = 64 +- FIRST-LINE [the message names the engine whose prime failed] = 59 +- FIRST-LINE [the sink receives the injected exception unchanged] = 63 +- FIRST-LINE [a prime fault is reported exactly once] = 55 +- FIRST-LINE [a failed prime leaves nothing to display] = 64 +- FIRST-LINE [must still be registered] = 52 +- FIRST-LINE [so a later read may re-prime] = 71 + +Clause check (P1-T2 acceptance): every exactly-1 token counts 1; `.BeSameAs(` = 3; `[TestClass]` and `using Moq;` = 0; `handleSeenBySink` = 3 (at least 3); `Regression for issue #942` = 2 (at least 1); all nine round-2 tokens = 1; same-line pairs 59 = 59, 63 = 63, 55 = 55, 64 = 64; handle captured (35) and probe installed (37) before the trigger (41). All clauses hold. + +## Build before the reorder (P1-T4) + +Timestamp: 2026-09-30T07-36 +Command: msbuild TaskMaster.sln /t:Build /m /p:Configuration=Debug "/p:Platform=Any CPU" +EXIT_CODE: 0 + +Output Summary: +- Run as CMD-BUILD (TASKID p1-t4): MSBuild resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory. +- MSBUILD_EXIT_CODE: 0 +- ERRORS: 0 +- TEST_DLL_ADVANCED: True +- CSC_OUT_TASKMASTER_TEST: 2 +- The test assembly now carries the Harness hook and the new test; the production file is unchanged from the anchor at this point. + +## POST-FORMAT: + +Timestamp: 2026-09-30T07-45 +Task: P3-T2 +Command: the P1-T2 CMD-TOKEN-COUNT with its exact 31-token TOKEN list, re-run on the formatted partial after the P3-T1 repository-wide format pass +EXIT_CODE: 0 + +Output Summary: +- Transcription note: the token `Regression for issue #942` was composed inside the payload by string concatenation (runtime length probe 25, equal to the literal's length) because a PreToolUse hook refused the command string that carried the literal verbatim; the counted value is identical. + +Token counts and first lines (post-format): + +| Token | Count | FIRST-LINE | +|---|---|---| +| public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged() | 1 | 27 | +| [TestMethod] | 1 | 26 | +| [TestClass] | 0 | 0 | +| public partial class EngineToggleStateCoordinatorTests | 1 | 14 | +| handleSeenBySink | 3 | 36 | +| harness.OnLogError = | 1 | 37 | +| var prime = harness.Coordinator.GetPrimeTask(SpamEngine); | 1 | 35 | +| probe.SetException(failure); | 1 | 41 | +| await prime; | 1 | 42 | +| has lost isolation | 1 | 46 | +| Regression for issue #942 | 2 | 9 | +| .ContainSingle( | 1 | 55 | +| .BeEmpty( | 1 | 64 | +| .BeSameAs( | 3 | 49 | +| using Moq; | 0 | 0 | +| // Arrange | 1 | 29 | +| // Act | 1 | 40 | +| // Assert | 1 | 44 | +| .Contain(SpamEngine | 1 | 59 | +| .BeSameAs(failure | 1 | 63 | +| var harness = new Harness(); | 1 | 30 | +| Invariant: for a key whose prime did not run to completion | 1 | 19 | +| handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine); | 1 | 38 | +| Task.CompletedTask, | 1 | 69 | +| harness.Invalidations.Should().BeEmpty( | 1 | 64 | +| the message names the engine whose prime failed | 1 | 59 | +| the sink receives the injected exception unchanged | 1 | 63 | +| a prime fault is reported exactly once | 1 | 55 | +| a failed prime leaves nothing to display | 1 | 64 | +| must still be registered | 1 | 52 | +| so a later read may re-prime | 1 | 71 | + +Clause check (every P1-T2 clause on the post-format tree): every exactly-1 token counts 1; `.BeSameAs(` = 3; `[TestClass]` and `using Moq;` = 0; `handleSeenBySink` = 3; `Regression for issue #942` = 2; all nine round-2 tokens = 1; same-line pairs 59 = 59 (Contain), 63 = 63 (BeSameAs failure), 55 = 55 (ContainSingle), 64 = 64 (BeEmpty); capture (35) and probe install (37) precede the trigger (41). All clauses hold. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md new file mode 100644 index 000000000..d96080630 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md @@ -0,0 +1,32 @@ +# Fail-before: prime fault ordering (issue 942) + +Timestamp: 2026-09-30T07-37 +Task: P1-T5 [expect-fail] +Command: vstest.console.exe TaskMaster.Test\bin\Debug\TaskMaster.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests" "/ResultsDirectory:coverage\test-results\942\p1-t5" "/Logger:trx;LogFileName=p1-t5.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" +EXIT_CODE: 1 +ExpectedExitCode: 1 + +Output Summary: +- vstest.console.exe resolved through vswhere; the trx stays under the ignored coverage directory. +- VSTEST_EXIT_CODE: 1 +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- COUNTERS total=25 executed=25 passed=24 failed=1 +- RESULT_COUNT: 25 +- RESULT GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked = Passed +- RESULT GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker = Passed +- RESULT GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse = Passed +- RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Failed +- FAILED GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged +- MESSAGE GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged :: Expected handleSeenBySink to refer to System.Threading.Tasks.ContinuationTaskFromTask {Status=RanToCompletion} because while the fault is being reported the prime handle must still be registered, so a caller that fetches it after the trigger awaits the report, but found System.Threading.Tasks.Task {Status=RanToCompletion}. +- The message contains `to refer to` (the FluentAssertions BeSameAs failure fragment) and `must still be registered` (the sink-handle assertion's reason fragment), so the failing assertion is the same-instance assertion on the sink-observed handle. The handle observed from inside the sink was the completed static task, not the registered continuation. +- The COUNTERS total (25) equals BASELINE-TOTAL (24) plus 1. The only FAILED name is the new test. +- Execution note: the first launch of this payload was refused by a PreToolUse hook (PARALLEL_WORKTREE_REMOVAL_BLOCKED) before running, because the command string contained the backslash worktree path together with the payload's results-directory removal. Nothing ran and nothing was removed. The payload was re-issued unchanged except that the worktree path is composed by string concatenation inside the payload; this run is the single executed run. + +Environment of the control: + +- The production file TaskMaster/Ribbon/EngineToggleStateCoordinator.cs is byte-identical to the merge base 231e1c0b55105aeb626bf5a6e8d0266a567cacad: its CMD-HASH value equals BASE-HASH-PROD from evidence/baseline/file-line-counts-baseline.md. +- PROD-HASH-AT-CONTROL: F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0 +- BASE-HASH-PROD: F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0 +- The Harness OnLogError hook (P1-T1), the new partial TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs (P1-T2) and its csproj Compile entry (P1-T3) are present, and the test assembly was rebuilt with them (P1-T4). +- The run settings are unchanged: `git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.runsettings scripts/vscode/TaskMaster.cli.runsettings` exited 0 (RUNSETTINGS_DIFF_EXIT=0). diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md new file mode 100644 index 000000000..8ea60b933 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md @@ -0,0 +1,46 @@ +# Pass-after: prime fault ordering (issue 942) + +Timestamp: 2026-09-30T07-39 +Task: P2-T4 (creates this file); P2-T5 and P3-T7 append. +Command: vstest.console.exe TaskMaster.Test\bin\Debug\TaskMaster.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests" "/ResultsDirectory:coverage\test-results\942\p2-t4" "/Logger:trx;LogFileName=p2-t4.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" +EXIT_CODE: 0 + +Output Summary: +- Identical to the P1-T5 command except the task id segments (p2-t4); vstest.console.exe resolved through vswhere; the trx stays under the ignored coverage directory. +- VSTEST_EXIT_CODE: 0 +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- COUNTERS total=25 executed=25 passed=25 failed=0 +- RESULT_COUNT: 25 +- RESULT GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker = Passed +- RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Passed +- RESULT GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked = Passed +- RESULT GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse = Passed +- No FAILED line was printed. +- The only production difference between this run and the P1-T5 fail-before run is the statement reorder and documentation in `CompletePrime` and `GetPrimeTask` of `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`. The Harness hook, the new partial and its csproj entry are test-side and were present in both runs. +- PROD-HASH-AFTER: D9C915AE9B00BB7AAB80183A7A0BA11748DE393781D7E5ADE2BDE29073B7002B (differs from BASE-HASH-PROD F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0) + +## POPULATION-COMPARISON: + +Timestamp: 2026-09-30T07-39 (P2-T5; read from evidence/baseline/coordinator-tests-baseline.md and this file) + +- BASELINE-TOTAL: 24; pass-after total: 25 = BASELINE-TOTAL plus 1 (the new test). +- Pass-after failed: 0. +- BASELINE-FAILED: NONE, so no baseline failure needs to be re-checked; no test is recorded as still failing. +- Result: the pass-after population is the baseline population plus the new regression test, all passed. No PASS-AFTER NOT GREEN. + +## FINAL-FIXTURE-RUN: + +Timestamp: 2026-09-30T07-48 (P3-T7, on the assembly rebuilt by the P3-T5 and P3-T6 gates) +Command: vstest.console.exe TaskMaster.Test\bin\Debug\TaskMaster.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests" "/ResultsDirectory:coverage\test-results\942\p3-t7" "/Logger:trx;LogFileName=p3-t7.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" +EXIT_CODE: 0 + +- VSTEST_EXIT_CODE: 0 +- TRX_PRESENT: True +- SEQUENCE_FILES: 0 +- COUNTERS total=25 executed=25 passed=25 failed=0 (total = BASELINE-TOTAL plus 1) +- RESULT GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked = Passed +- RESULT GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse = Passed +- RESULT GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker = Passed +- RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Passed +- No FAILED line. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/feature-audit.2026-09-30T08-30.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/feature-audit.2026-09-30T08-30.md new file mode 100644 index 000000000..d495873c6 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/feature-audit.2026-09-30T08-30.md @@ -0,0 +1,73 @@ +# Feature Audit — engine-toggle-prime-fault-logging-test-races (Issue #942) + +- Review date label: 2026-09-30T08-30 (assigned without a clock; later than every executor evidence label) +- Work mode: `full-bug` (`issue.md` line 12) — acceptance-criteria source is `spec.md` only. `issue.md` carries no acceptance-criteria section and `user-story.md` does not exist; neither is consulted as an AC source. +- Companion artifacts: `policy-audit.2026-09-30T08-30.md`, `code-review.2026-09-30T08-30.md`. + +## Scope and Baseline + +- Base: `231e1c0b55105aeb626bf5a6e8d0266a567cacad` (origin/main as merged into the branch at `fadcb6417`; plan correction C1 re-anchored every gate from the preparation anchor `ddbab26a` to this commit). The executor's P0-T4 verified the anchor is an ancestor of HEAD and equals `git merge-base origin/main HEAD` (`evidence/baseline/scope-and-anchor.md`). +- Branch footprint against the base (executor P3-T14 `git diff --name-status`, corroborated by the caller's `git diff --stat` and by this review's Glob of the feature folder): `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (M), `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` (M), `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (A), `TaskMaster.Test/TaskMaster.Test.csproj` (M), the feature folder (issue, spec, research, plan, 37 evidence projections), and the inherited promotion record `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` (A). +- Baseline state of the production file: 415 lines, SHA-256 `F2A961DD50F2E4678B5CF8B7FAA3F0316AA22D2FB8FE904AE5D08057F26ACEF0`; `CompletePrime` at lines 341–355 with `TryRemove` (348) before `_logError` (354). Baseline fixture: 24 tests, all passing. Baseline suite with coverage: 7323/7323; first-party 85.32% lines / 79.73% branches; coordinator file 143/143 lines, 37/38 branches. +- Post-change state (direct read): 420 lines; `CompletePrime` at 344–360 with `_logError` (358) before `TryRemove` (359) and the why-comment at 355–357. Fixture: 25 tests. Suite: 7324/7324; first-party 85.31% / 79.72%; coordinator file unchanged at 143/143 and 37/38, line 359 hits=1 (read directly from `coverage/final-942.cobertura.xml`). +- Review tooling: Read, Grep and Glob only; no git or build tool was available. Where a criterion asserts a property of the diff or of a run, the evidence projection is cited and, where possible, cross-checked against the current file content or the raw Cobertura document. + +## Acceptance Criteria Inventory + +Source: `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, section `## Acceptance Criteria`, lines 228–241. Fourteen checkbox items, all `- [x]` at review start (executor check-offs P3-T15 to P3-T28). + +| ID | Criterion (abbreviated) | State at review start | +|---|---|---| +| AC1 | `CompletePrime` reports through the delegate before removing the key; early return, unwrap and synthesized `TaskCanceledException` unchanged; no `catch`, `try` or lock added | [x] | +| AC2 | Summary describes report-then-clear; adjacent comment states why; `GetPrimeTask` returns states the guarantee in one sentence | [x] | +| AC3 | `Harness.OnLogError` internal settable `Action`, documented, invoked null-conditionally right after `Errors.Add`; no existing test method modified (hunks only inside `Harness`) | [x] | +| AC4 | New partial with the named test capturing the handle before the trigger, sink-side `GetPrimeTask`, the six listed assertions | [x] | +| AC5 | MSTest, strict Moq, FluentAssertions with reasons, AAA, XML summary naming the issue and invariant, loss-of-isolation comment | [x] | +| AC6 | csproj `Compile Include` for the partial; pass-after lists the test as executed and passed | [x] | +| AC7 | Fail-before projection with Timestamp, Command, non-zero EXIT_CODE = ExpectedExitCode, merge-base commit, new test failing on the same-instance assertion | [x] | +| AC8 | Pass-after projection, EXIT_CODE 0, both tests passed, only production difference is the reorder | [x] | +| AC9 | Every fixture test passes in the pass-after run; original test byte-for-byte unchanged | [x] | +| AC10 | No sleep/delay/retry/wall-clock/timeout/`[DoNotParallelize]`/blocking wait/temp file/scheduler seam; run-settings unmodified | [x] | +| AC11 | Final toolchain pass: csharpier check clean, both rebuilds exit 0 with no skipped CoreCompile, MSTest-with-coverage exit 0, one uninterrupted pass | [x] | +| AC12 | Coverage baseline and post-change projections with the coordinator comparison; changed lines not decreased, method fully covered; no raw trx/xml/coverage file added | [x] | +| AC13 | Diff touches nothing outside the four code files, the feature folder and the promotion record; Race partial, wiring, run-settings, `StartPrimeIfNeeded`, prime gate unchanged | [x] | +| AC14 | Each of the three source files at or below 500 lines | [x] | + +## Acceptance Criteria Evaluation + +| ID | Verdict | Evidence and verification performed | +|---|---|---| +| AC1 | PASS | Direct read of lines 344–360: `if (completed.Status == TaskStatus.RanToCompletion) return;` (346–349), `var failure = (Exception)completed.Exception?.GetBaseException() ?? new TaskCanceledException(completed);` (351–353), `_logError(...)` at 358, `_primeTasks.TryRemove(engineName, out _);` at 359. Span tokens `SPAN_TRY=0`, `SPAN_CATCH=0`, `SPAN_LOCK=0`; file-level `catch (` = 1 (line 181) and `lock (` = 1 (line 271) unchanged from the anchor (`evidence/qa-gates/production-reorder-scope.md`, POST-FORMAT; both counts confirmed by Grep). | +| AC2 | PASS | Summary lines 330–333 contain "and only then is the in-flight marker cleared"; comment lines 355–357 begin "Report-then-clear is load-bearing" and state the caller guarantee; `GetPrimeTask` returns lines 245–247 carry the one-sentence guarantee "the marker is cleared only after that report has returned, so a caller that receives Task.CompletedTask can rely on the fault having been reported". Direct read. | +| AC3 | PASS | Lines 440–445: documented `internal Action OnLogError { get; set; }`; lambda lines 415–419: `Errors.Add(new LoggedError(message, exception));` then `OnLogError?.Invoke(message, exception);`. Hunks `@@ -415 +415,5 @@` and `@@ -435,0 +440,7 @@`, both inside `Harness` (403–452); `[TestMethod]` count 15 unchanged (`evidence/qa-gates/harness-hook-edit-scope.md`; count confirmed by direct read). File is 470 = 459 + 12 - 1 lines, consistent with the recorded numstat. | +| AC4 | PASS | Direct read of the 77-line partial: handle captured at 35 before `probe.SetException(failure)` at 41; `harness.OnLogError = (_, _) => handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);` at 37–38; `await prime;` at 42; assertions `BeSameAs(prime)` (47–54), `ContainSingle` (55), `Message.Contain(SpamEngine)` (56–59), `Exception.BeSameAs(failure)` (60–63), `Invalidations.BeEmpty` (64), `GetPrimeTask(...).BeSameAs(Task.CompletedTask)` (65–72). | +| AC5 | PASS | `[TestMethod]` line 26; `Harness.Engines` is `new Mock(MockBehavior.Strict)` (primary partial line 424); six FluentAssertions calls each with a reason; `// Arrange` / `// Act` / `// Assert` at 29 / 40 / 44; XML summary (18–25) names issue #942 and states the invariant; comment at 45–46 states that a pass without the reorder means the negative control has lost isolation. | +| AC6 | PASS | `TaskMaster.Test.csproj` line 360: `` immediately after the Race entry at 359 (Grep). Pass-after RESULT line `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Passed` at P2-T4 and again at P3-T7; total 25 = baseline 24 + 1. | +| AC7 | PASS | `evidence/regression-testing/prime-fault-ordering-fail-before.md`: `Timestamp: 2026-09-30T07-37`, full vstest `Command:`, `EXIT_CODE: 1`, `ExpectedExitCode: 1`, merge base `231e1c0b…` cited with `PROD-HASH-AT-CONTROL` equal to `BASE-HASH-PROD`, hook/partial/csproj present, run settings diff-clean; `MESSAGE` line is the `BeSameAs` failure ("Expected handleSeenBySink to refer to … ContinuationTaskFromTask … but found System.Threading.Tasks.Task") with the sink-handle reason fragment. 24 passed / 1 failed, the only failure being the new test. | +| AC8 | PASS | `evidence/regression-testing/prime-fault-ordering-pass-after.md`: identical command apart from the results segments, `EXIT_CODE: 0`, 25/25, RESULT lines for both named tests `= Passed`, explicit statement that the only production difference is the reorder and documentation in `CompletePrime`/`GetPrimeTask` with `PROD-HASH-AFTER` differing from the base hash. | +| AC9 | PASS | Pass-after and final fixture runs: 25 executed, 25 passed, no FAILED line; population = baseline 24 + 1 (independently recomputed: 15 + 3 data rows + 6 Race + 1 new = 25). Original test: `BASE_METHOD_SHA` = `NOW_METHOD_SHA`, `METHOD_UNCHANGED=True`, re-verified post-format (`evidence/qa-gates/original-test-unchanged.md`); the method text read at lines 212–243 still fetches the handle after the trigger (223–224), as the spec requires it to. | +| AC10 | PASS | `evidence/qa-gates/determinism-tokens.md`: 20 tokens at 0 over the 89 added test lines; `RUNSETTINGS_DIFF_EXIT=0`. Independent direct read of both test files finds no `Thread.Sleep`, `Task.Delay`, retry loop, `DateTime`, `[Timeout]`, `[DoNotParallelize]`, `.Wait(`/`.Result`, temp-file API or `TaskScheduler`. | +| AC11 | PASS (disclosed substitution) | `evidence/qa-gates/toolchain-final-pass.md`: pass 1, in CLAUDE.md order, no intervening rewrite; check "no differences" (1626 files); analyzer and nullable rebuilds exit 0 with `SKIP_CORECOMPILE_LINES: 0` and `CSC_OUT_*` ≥ 1; coverage-enabled run exit 0, 7324/7324. The MSTest-with-coverage step ran the DIRECT route (the runner's own inner collector invocation with four shell-icon classes excluded, plan D-6, after the stall probe recorded a workstation-local failure) rather than `Invoke-MSTestWithCoverage.ps1` verbatim; the substitution is recorded in the plan, the stall-probe artifact and the toolchain artifact, and the criterion's wording ("the MSTest-with-coverage run exiting zero") is met. Recorded as policy-audit PA-2, non-blocking. | +| AC12 | PASS | `evidence/baseline/coverage-baseline.md` and `evidence/qa-gates/coverage-post-change.md` (COMPARISON) exist with the coordinator comparison: lines 143/143 → 143/143, branches 37/38 → 37/38, `COMPLETEPRIME-UNCOVERED-FINAL: 0`, changed line 359 hits=1. Independently verified by reading the class element of `coverage/final-942.cobertura.xml` (document line 230308 onward): every `CompletePrime` line element hits=1, branch conditions 2/2 and 4/4. No trx/xml/coverage file added: `RAW-DOCS-COMMITTED: 0`, `RAW-DOCS-UNTRACKED-IN-FEATURE: 0` (with `--ignored`), the 29-path added list contains only `.cs` and `.md`; Glob of the feature folder lists Markdown only. | +| AC13 | PASS | `evidence/qa-gates/footprint-scope.md` (P3-T14): `THIS-ITEM-FOOTPRINT` is the four code paths plus feature-folder paths; `INHERITED-AND-EXCLUDED` is exactly the promotion record; `NONGOAL_FILES_DIFF_EXIT=0` for the Race partial and the ribbon wiring, `RUNSETTINGS_DIFF_EXIT=0`. The caller's independent `git diff --stat` reports the same footprint. `StartPrimeIfNeeded` (263–280) and `_primeGate` read unchanged from the plan's fact 1 description; the production numstat `10 5` is confined to the two documented hunk windows. This review could not run git; the criterion is credited on two independent git-based enumerations plus direct reads. | +| AC14 | PASS | Direct read: 420, 470, 77 lines (matches `evidence/qa-gates/file-line-counts.md`). | + +Summary: 14 PASS, 0 PARTIAL, 0 FAIL, 0 UNVERIFIED. + +## Acceptance Criteria Check-off + +All fourteen items were already `- [x]` in `spec.md` at review start and every one evaluates PASS above, so no check-off edit and no uncheck was made. No item was unchecked silently; no phantom criterion was added. + +### Acceptance Criteria Status +- Source: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md +- Total AC items: 14 +- Checked off (delivered): 14 +- Remaining (unchecked): 0 +- Items remaining: none + +## Summary + +- Verdict: **PASS** — 14/14 acceptance criteria verified against evidence; 0 blocking findings in this artifact. +- The defect's mechanism (statement order inside `CompletePrime`, not a missing await) is fixed by the two-statement reorder; the regression test is deterministic by construction and was observed failing against the base production file before the fix. +- Non-blocking follow-ups (owed outside this branch): hazard B (registration racing removal on a synchronous non-success prime; separately promoted), optional `try`/`finally` hardening should a throwing sink ever be injected, and the two other continuation-discarding production sites named in the spec. Policy-audit observations PA-1 to PA-5 are recorded there. +- Merge readiness: the local coverage run excluded four shell-icon test classes that CI executes; the PR's CI `mstest-coverage` check remains the repo-wide gate for those classes. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md new file mode 100644 index 000000000..6e79f104a --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md @@ -0,0 +1,62 @@ +# engine-toggle-prime-fault-logging-test-races (Issue #942) + +- Date captured: 2026-09-29 +- Author: Dan Moisan +- Status: Promoted -> docs/features/active/engine-toggle-prime-fault-logging-test-races/ (Issue #942) + +> Automation note: Keep the section headings below unchanged; the promotion tooling maps each of them into the GitHub bug issue template. + +- Issue: #942 +- Issue URL: https://github.com/drmoisan/TaskMaster/issues/942 +- Last Updated: 2026-09-30 +- Work Mode: full-bug + +## 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 #939 at head `9624376dc` (first attempt). + +## Impact / Severity + +- [ ] Blocker +- [x] High +- [ ] Medium +- [ ] Low + +## 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 that hits a required check. 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. + +## Next Step + +- [x] Promote to GitHub issue (bug-report template) +- [ ] Move to active fix folder / branch diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md new file mode 100644 index 000000000..54e31248b --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md @@ -0,0 +1,757 @@ +# 2026-09-29-engine-toggle-prime-fault-logging-test-races (Plan) + +- **Issue:** #942 +- **Parent (optional):** none +- **Owner:** drmoisan +- **Work Mode:** full-bug (acceptance criteria come from `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` only; no user story exists for this item and none is to be authored) +- **Last Updated:** 2026-09-30T12-00 +- **Status:** Ready for confirming preflight (orchestrator corrections C1 and C2) +- **Version:** 1.4 +- **Revision record:** preflight round 1 returned deltas D1 to D18; every one is applied, and the spec's AC13 was amended by the planner in that round (D7, option a) to name the inherited promotion record. Preflight round 2 returned deltas R1 to R6 and advisories A1, A2 and A4 (A3 not adopted); every one is applied in this version, and the spec header was advanced to version 0.3 to record the AC13 amendment. +- **Orchestrator correction C1 (2026-09-30, re-anchor to post-merge main; standing authority):** at execution start origin/main (`231e1c0b55105aeb626bf5a6e8d0266a567cacad`) was merged into this branch as merge commit `fadcb6417`, as the item notes require. After that merge the prepared anchor `ddbab26a0149bf2ca5d0256e60686ad79e74d90c` is no longer `git merge-base origin/main HEAD`, so P0-T4 would stop with `BASE-SHA MISMATCH`, and every anchored diff against it would list the 41 commits main gained, making the AC13 footprint gate fail for reasons unrelated to this item. Every gate anchor in fact 14, D-8 and the tasks P0-T4, P1-T3, P1-T5, P2-T6, P2-T7, P3-T10, P3-T11, P3-T12 and P3-T14 is therefore re-anchored to `231e1c0b55105aeb626bf5a6e8d0266a567cacad`. No acceptance criterion changes intent. Verified before the edit: `git diff --stat ddbab26a0149bf2ca5d0256e60686ad79e74d90c 231e1c0b55105aeb626bf5a6e8d0266a567cacad` over the four code files, TaskMaster/Ribbon, TaskMaster.Test/Ribbon, both run-settings files, scripts/vscode, .csharpierignore, global.json, dotnet-tools.json, coverage.config and the promotion record prints nothing, so every line number, line count and hash fact in this plan holds unchanged at the new anchor; and `git diff --name-status 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD` lists exactly the four feature-folder files and the promotion record. Negative control (the re-anchored gates can still fail): that same name-status diff reports the promotion record, a path outside the feature folder and the code files, so the P0-T4 and P3-T14 footprint gates observe out-of-scope paths at the new anchor; and `git merge-base --is-ancestor ddbab26a0149bf2ca5d0256e60686ad79e74d90c HEAD` still exits 0 while `git merge-base origin/main HEAD` no longer prints `ddbab26a0149bf2ca5d0256e60686ad79e74d90c`, which is the mismatch the P0-T4 equality clause exists to detect. The self-review entries below that record the branch cut are historical and keep the original anchor. If main is merged again before the pull request is opened, the orchestrator repeats this correction with the new merge base and re-runs the affected gates. +- **Orchestrator correction C2 (2026-09-30, confirming-preflight deltas D1 to D3; standing authority):** the confirming preflight on C1 returned REVISIONS REQUIRED with three merge-induced deltas, applied verbatim. D1 re-anchors the .gitignore citations (fact 10, the Write Set paragraph, fact 14 and the CITATION line) to the post-merge line numbers 146, 147, 150 and 151. D2 adds `--ignored` to the P3-T12 feature-folder porcelain span, because the merged .gitignore ignores `*.trx` and `*cobertura*.xml` repository-wide and a plain porcelain span would read 0 whatever is on disk; negative control: `git check-ignore -v --no-index` reports a feature-folder `probe.trx` and `probe.cobertura.xml` as ignored by lines 146 and 147. D3 aligns the P3-T13 drive-path pattern with the merged CI hygiene guard (scripts/hygiene/Test-RepositoryHygiene.Rules.ps1 line 21); negative control: the old pattern counts 0 for a lower-case drive-and-profile path and for a drive-and-profile path with doubled separators, where the new pattern counts 1 for each; both patterns count 1 for a single-separator upper-case drive-and-profile path; and both count 0 for a drive path outside the profile folder. The probe strings are not reproduced here because P3-T13 and the CI hygiene guard would count them in this file. No acceptance criterion changes intent; D2 and D3 make two existing checks stricter. +- **Plan path continuity:** this file is updated in place for every preflight revision round. No timestamped sibling plan file is created for this cycle. + +**Fail-closed evidence rule:** every command-bearing task writes one evidence artifact carrying `Timestamp:`, `Command:`, `EXIT_CODE:` and `Output Summary:`. A task whose artifact is missing or incomplete stays unchecked, and the plan outcome is BLOCKED or INCOMPLETE, never PASS. + +**Evidence accounting rule:** the artifact path is named in the task text. Do not mark an evidence-bearing task complete without the artifact on disk at that exact path. + +**Evidence location:** every artifact lives under `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/` in the canonical sub-kinds `baseline/`, `regression-testing/`, `qa-gates/` and `other/`. EVIDENCE_LOCATION_OVERRIDE_REJECTED: none supplied; no artifacts-tree evidence path appears in this plan. In task text the token FEATURE abbreviates `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942`; the Write Set below spells every path in full. + +## Requirement sources + +- Acceptance criteria: `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, section `## Acceptance Criteria`, lines 228 to 241: fourteen checkbox lines `- [ ] AC1 —` through `- [ ] AC14 —`, each on one line. The check-off edit changes only `- [ ] ACn —` to `- [x] ACn —`. +- Design record: `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md` (sections 2, 4, 6, 7, 8 and 9 govern this plan). +- Issue metadata: `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md` carries `- Work Mode: full-bug` at line 12 and no acceptance-criteria section. It is not an acceptance-criteria source for this cycle. + +## Write Set (every file this plan creates or modifies) + +Code files (the only paths outside the feature folder this plan may change): + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (new) +- `TaskMaster.Test/TaskMaster.Test.csproj` + +Feature documents: + +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` (the AC13 line was amended by the planner in preflight round 1 to name the inherited promotion record, and the header's Last Updated and Version fields were advanced by the planner in round 2 to record that amendment; the executor makes check-off edits only) +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/plan.2026-09-29T23-07.md` (task check-off edits only) + +Evidence files, all new, fixed names (the write time is the `Timestamp:` field): + +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-instructions-read.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/scope-and-anchor.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-sdk.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-tool-restore.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-nuget-restore.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/bootstrap-dotnet-coverage.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/csharpier-check-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-analyzer-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/msbuild-nullable-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/stall-probe.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coordinator-tests-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/file-line-counts-baseline.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/phase0-commit.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-before-reorder.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/build-after-reorder.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/harness-hook-edit-scope.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csproj-registration.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/production-reorder-scope.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/implementation-commit.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-format.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/file-line-counts.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/csharpier-check-final.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-analyzer-final.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/msbuild-nullable-final.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/coverage-post-change.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/toolchain-final-pass.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/original-test-unchanged.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/determinism-tokens.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/footprint-scope.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/evidence-hygiene.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/ac-status-summary.md` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/other/reduced-audit-handoff.md` + +Files this plan must not touch, stated so the executor fails closed rather than infers: TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs, TaskMaster/Ribbon/RibbonController.EngineCommands.cs, TaskMaster.runsettings, scripts/vscode/TaskMaster.cli.runsettings, every file under scripts/vscode/, every file under .claude/ except .claude/agent-memory/ (session memory, never staged by this plan), every file under config/, and every file under docs/features/potential/ (including the inherited promotion record docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md, which the branch already carries and this plan neither edits nor stages). No potential entry is written by this plan: the spec records hazard B as promoted separately by the coordinator, so that promotion is outside this plan's Write Set. No orchestration state file is written or named by any task. No raw test-result document (trx), raw coverage document (cobertura, coverage, coveragexml) or msbuild log is copied into the feature folder under any name; raw documents stay under the repository coverage directory, which .gitignore line 150 ignores (line 151 re-includes only its .gitkeep). + +## AC identity table + +Each ID names one checkbox in the spec's `## Acceptance Criteria` section, in document order (spec lines 228 to 241). + +| ID | Spec line | Opening words of the criterion | +|---|---|---| +| AC1 | 228 | CompletePrime invokes the injected error-log delegate before it removes the engine key | +| AC2 | 229 | The summary documentation on CompletePrime describes the report-then-clear order | +| AC3 | 230 | The private Harness exposes an internal settable OnLogError hook | +| AC4 | 231 | A new partial contains the test GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged | +| AC5 | 232 | The new test uses MSTest attributes, a strict Moq mock, FluentAssertions with reason strings | +| AC6 | 233 | The test project contains an explicit Compile Include item for the new partial | +| AC7 | 234 | Fail-before evidence exists as prime-fault-ordering-fail-before.md | +| AC8 | 235 | Pass-after evidence exists as prime-fault-ordering-pass-after.md | +| AC9 | 236 | Every test method in the coordinator fixture passes in the pass-after run, and the original test is byte-for-byte unchanged | +| AC10 | 237 | Neither the new partial nor the Harness change introduces a sleep, delay, retry, wall-clock read, timeout, parallelism attribute, blocking wait, temporary file or scheduler seam | +| AC11 | 238 | The final toolchain pass recorded in toolchain-final-pass.md | +| AC12 | 239 | Coverage evidence exists as coverage-baseline.md and coverage-post-change.md | +| AC13 | 240 | The diff against the merge base modifies no repository file outside the four code files and this feature folder | +| AC14 | 241 | Each of the three source files is at or below the five-hundred-line ceiling | + +## Verified tree facts (re-derived against this worktree while authoring) + +1. `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` is 415 content lines (416 with the trailing newline). No nullable directive. `GetPrimeTask` documentation 237 to 246 with the `returns` element at 242 to 246; body 247 to 255. `StartPrimeIfNeeded` 261 to 278 with `lock (_primeGate)` at 269 and the assignment at 276. `StartObservedPrime` 290 to 303. `CompletePrime` documentation 327 to 340 (`summary` 327 to 331, `remarks` 332 to 340); body 341 to 355: status test 343, `return;` 345, `_primeTasks.TryRemove(engineName, out _);` 348, `var failure =` 350 to 352, `_logError(BuildPrimeFailedMessage(engineName), failure);` 354. The file carries exactly one `catch (` (181) and exactly one `lock (` (269). `_primeTasks.TryRemove(engineName, out _);` occurs exactly once (348) and `internal Task GetPrimeTask(` exactly once (247), so both serve as single-line anchors for the hunk-window rule in P2-T6. The word `try` occurs as a code token only at 177 (the click boundary) and `catch` at 181; the `CompletePrime` span contains neither today. +2. `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` is 459 content lines. No nullable directive. `[TestClass]` at 22 on `public partial class EngineToggleStateCoordinatorTests` at 23. `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` 213 to 243 (`[TestMethod]` at 212), fetching the handle after the trigger at 223 to 224. `Harness` 403 to 441: constructor 405 to 417 with the error-log lambda on the single line 415; the strict engines mock is declared at 419 to 420 with the literal `new Mock(MockBehavior.Strict);` alone on line 420 (the only occurrence in the file); `OnInvalidate` 430 to 434; `Invalidations` 436; `Errors` 440. `LoggedError` 446 to 457. The file declares sixteen test methods: fifteen `[TestMethod]` lines and one `[DataTestMethod]` line (101, three data rows). The lambda `(_, _) => { }` at 35 proves discard parameters compile in this project. +3. `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs` is 277 content lines: a second partial with no `[TestClass]`, usings System, System.Threading.Tasks, FluentAssertions, Microsoft.VisualStudio.TestTools.UnitTesting and Moq. `GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker` captures the handle before the trigger at 211 to 215. +4. `TaskMaster.Test/TaskMaster.Test.csproj`: `` at 352 and `` at 359, the latter followed by the EngineTogglePressedStateCacheTests entry at 360 and the AssemblyInfo entry at 361. Explicit compile items; an unlisted file is not compiled. The project file is excluded from the formatter by .csharpierignore line 12. +5. `TaskMaster/Ribbon/RibbonController.EngineCommands.cs` 67 to 77 wires the production sink as `(message, exception) => logger.Error(message, exception)`; it does not re-enter the coordinator. Not modified. +6. TaskMaster.Test/packages.config: FluentAssertions 8.11.0 (line 7), Moq 4.21.0 (41), MSTest 4.4.1 (42 to 44). FluentAssertions renders a failed `BeSameAs` as `Expected ... to refer to ..., but found ...`; the phrase `to refer to` is the stable fragment. +7. TaskMaster.runsettings lines 4 to 7 and scripts/vscode/TaskMaster.cli.runsettings lines 4 to 7 both set Workers 0 and Scope ClassLevel; the root file additionally declares a Code Coverage collector (9 to 29). Neither is modified. +8. scripts/vscode/Invoke-MSTestWithCoverage.ps1 (462 lines): `Get-DotnetCoverageArgumentList` 41 to 95 hard-codes the LiveOutlook exclusion at 91 and the trx logger at 93 with no extension point; `Invoke-DotnetCoverageCollection` throws at 262 on a non-zero collector exit, after vstest has written the trx; `Invoke-MSTestWithCoverageMain` 274 to 457 with `-ResultsDirectory` default coverage\test-results and `-LogFileName` default mstest-coverage-run.trx (297, 298), assembly discovery 348 to 355 (filters bin\Debug, excludes obj, ref and a .claude segment relative to the search root), post-processing in place 399 to 402, scoped-run threshold gate 406 to 409, the `First-party coverage:` line 410, the JaCoCo projection written beside the document as coverage.cobertura.jacoco.xml 415 to 423, the trx summary 430 to 447, and the raw-document retention rule 449 to 453 (retained only when the output's parent directory is the repository coverage directory, so the default output path is retained). Entry guard at 459 to 461, so dot-sourcing is safe. It prints `Discovered N test assemblies.` (374) and never prints a discovered path. +9. scripts/vscode/Invoke-MSTestWithCoverage.Helpers.ps1: `Get-CoberturaClassLineSummary` 160 to 258 (LineMap keyed by line number, TotalLines, CoveredLines, TotalBranches, CoveredBranches); `Merge-CoberturaClassesByFilename` 260 to 405 merges every class element sharing a filename into one, so a post-processed document carries exactly one class element per source file per package; `ConvertTo-KoverageCoberturaXml` 407 to 471 rewrites filename attributes to workspace-relative paths with native separators (backslash on this host) and recomputes the root counters. Threshold functions in Invoke-MSTestWithCoverage.Threshold.ps1: 80 percent line (52 to 55), 75 percent branch (122 to 125). Scope gate in Invoke-MSTestWithCoverage.Scope.ps1 59 to 104. First-party line shape in Invoke-MSTestWithCoverage.FirstParty.ps1 117 to 120. Projection, reconciliation and retention predicate in Invoke-MSTestWithCoverage.Projection.ps1 14 to 81, 83 to 146 and 148 to 197. Trx summary in Invoke-MSTest.TrxSummary.ps1: `Get-TrxRunSummary` 12 to 101 (Skipped derived at 98), `Format-TrxRunSummary` 103 to 150 (five lines, the last `Failed tests: ` followed by names or `none`). +10. .gitignore: the coverage-directory ignore at 150 and the gitkeep re-include at 151 (the C1 merge inserted a comment block at 143 to 145 and the repository-wide ignore patterns *.trx at 146 and *cobertura*.xml at 147); coverage\.gitkeep is tracked so the coverage directory always exists after checkout. .csharpierignore excludes the evidence tree (4), cobertura, coverage, coveragexml and trx files (5 to 8), csproj, props and targets (12 to 14), packages.config (16) and app.config (18); CSharpier 1.2.6 (dotnet-tools.json, repository root) also processes xml files not so excluded. global.json pins SDK 8.0.205 with latestFeature roll-forward under .dotnet-sdk; scripts/vscode/Install-RepoDotNetSdk.ps1 installs it (default version parameter 8.0.205). scripts/vscode/Invoke-Restore.ps1 takes SolutionPath, Configuration and Platform (defaults TaskMaster.sln, Debug, Any CPU). coverage.config excludes third-party modules only (14 to 20). +11. .claude/hooks/validate-planner-output.ps1 line 95 requires a separator-bearing path token on every task's opening line; line 339 requires the final phase's title or task text to carry QA vocabulary. +12. The .dotnet-sdk, packages and TaskMaster.Test\bin\Debug trees were not observable from the planning session (they are git-ignored; a miss is inconclusive), so every bootstrap task is guarded and gated on its post-task marker. +13. Host constraint carried from the sibling item #931 plan in this same run: four UtilitiesCS.Test shell-icon test classes (`HelperClasses.ShellUtilities_Tests`, `HelperClasses.ShellUtilitiesStatic_Tests`, `HelperClasses.SysImageListHelperTests`, `EmailIntelligence.OSBrowser_Tests`) have stalled vstest on this workstation; CI executes them. Whether the stall reproduces today is unknown, so P0-T13 measures it and the result selects the coverage route (Decision D-6). +14. The worktree reflog records this branch cut at `ddbab26a0149bf2ca5d0256e60686ad79e74d90c` with documentation-only commits since (five at preflight round 2: active folder creation, research, spec, plan, plan revision; the count is not gated); at execution start origin/main `231e1c0b55105aeb626bf5a6e8d0266a567cacad` was merged in (merge commit `fadcb6417`, correction C1), which changed none of the files this plan cites except .gitignore (fact 10 carries its shifted line numbers), and that commit is the gate anchor from then on; and `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` exists on the tree with `Status: Promoted` at its line 5. The anchored name-status diff at P0-T4 is therefore expected to list only feature-folder paths and that promotion record; the amended AC13 names the record as its sole exemption outside the four code files and the feature folder. The harness-supplied git status at session start described a different worktree (its recent-commit list is not this branch's history), so it was not used as a fact source. + +## Design decisions (do not redesign) + +- **D-1 Fix shape.** In `CompletePrime`, `_primeTasks.TryRemove(engineName, out _);` moves to after `_logError(BuildPrimeFailedMessage(engineName), failure);`. The early return, the `failure` computation, the synthesized `TaskCanceledException`, the message builder and the `StartPrimeIfNeeded` lock are untouched. No `try`, `catch`, `finally` or lock is added to `CompletePrime`. The spec's non-goals (hazard B, try-finally hardening, the Race partial, the wiring, the run settings, the original test) are binding. +- **D-2 Documentation literals (quoted here so the presence gates are exonerated; each is written on ONE physical line).** The `summary` element on `CompletePrime` is replaced by four lines whose third line carries the token `and only then is the in-flight marker cleared`; the comment placed immediately above the `_logError` statement begins with the token `Report-then-clear is load-bearing`; the `returns` element on `GetPrimeTask` gains one sentence, one line of which carries the token `cleared only after that report has returned`. Each token has zero occurrences in the tree today. Written without code formatting so the quotation stands in plan prose outside every command span, the summary token reads: and only then is the in-flight marker cleared. The full texts are in the Delivered Source section. +- **D-3 Harness hook.** `internal Action OnLogError { get; set; }` is added after `OnInvalidate` (one blank line and then the six hook lines are inserted after line 434), documented with a summary whose first line carries the token `invoked from inside the error-log sink`. The single-line error-log lambda at line 415 becomes a block whose first statement is `Errors.Add(new LoggedError(message, exception));` and whose second is `OnLogError?.Invoke(message, exception);`. No other line of that file changes; no `[TestMethod]` line is added or removed. +- **D-4 New partial.** `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` follows the Race partial's shape: no `[TestClass]` (the attribute applies to the whole type from the primary file), usings System, System.Threading.Tasks, FluentAssertions and Microsoft.VisualStudio.TestTools.UnitTesting (Moq is not named directly, so its using is omitted to avoid an unnecessary-using diagnostic), one region, one `[TestMethod]`. The full text is in the Delivered Source section. Its csproj entry is inserted immediately after the Race entry (line 359). +- **D-5 Fail-before is a real run, not a dossier.** The hook, the partial and the csproj entry compile against the unchanged production file, so P1-T5 runs the coordinator class with the reorder absent and the new test fails on the `BeSameAs(prime)` assertion deterministically (research section 8.2). The failing assertion's message carries `to refer to`. +- **D-6 Coverage route is selected by a recorded observation.** P0-T13 runs the four shell-icon classes alone and records `STALL-PROBE: CLEAR` or `REPRODUCES`. `COVERAGE-ROUTE: RUNNER` (CLEAR) runs scripts/vscode/Invoke-MSTestWithCoverage.ps1 verbatim (CLAUDE.md step 4). `COVERAGE-ROUTE: DIRECT` (REPRODUCES) issues the runner's own inner collector invocation with the four-class exclusion appended and post-processes with the runner's own helpers, because the runner hard-codes its filter (fact 8). Both routes yield the same committed forms: the `First-party coverage:` line, the JaCoCo package projection text, the trx-derived summary text and the per-file coordinator figures, all transcribed into Markdown. Under DIRECT the CLAUDE.md floors are applied to the post-processed document by the runner's own threshold functions and a NOT MET result is treated exactly as a runner exit non-zero. +- **D-7 Repository-wide rate is recorded, per-file figures are gated.** The merged repository-wide line rate is not reproducible across runs of an identical tree, so P3-T10 compares it in two branches (comparable denominators within 1 percent of lines-valid: gate with 0.5 percentage points of tolerance; otherwise record and do not gate) and puts the no-regression weight on the coordinator file: lines-valid equal to baseline (the change adds no executable statement), covered lines not lower, covered branches not lower, and every Cobertura line element inside the `CompletePrime` span with hits at least 1. +- **D-8 Anchor.** The merge base is the commit `231e1c0b55105aeb626bf5a6e8d0266a567cacad` (origin/main as merged at execution start, correction C1; it replaces the caller-supplied preparation anchor `ddbab26a0149bf2ca5d0256e60686ad79e74d90c`). P0-T4 verifies it is an ancestor of HEAD and equals `git merge-base origin/main HEAD`; a mismatch is `BASE-SHA MISMATCH`: stop and report. Every diff gate carries that literal as its ref operand. Paths already changed between the anchor and HEAD at P0-T4 form the inherited set, recorded once as `INHERITED-COMMITTED:`; P0-T4 requires every member to be a feature-folder path or exactly the promotion record `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` (the amended AC13's sole exemption), and stops otherwise. P3-T14 subtracts the promotion record by rule and records the subtraction as `INHERITED-AND-EXCLUDED:` beside `THIS-ITEM-FOOTPRINT:`. Uncommitted paths under .claude/agent-memory/ are session memory: they may appear in porcelain output, are never staged, and are not part of the anchored diff. +- **D-9 Commits.** Three commits: P0-T16 (feature folder only, the exempt tree docs/features/active/), P2-T8 (the four code files plus the feature folder, staged only after a scoped CSharpier format of the three source files so the committed text is already formatter-stable and the Phase 3 format pass has nothing to rewrite), P3-T30 (feature folder). No code file is edited after the P2-T8 commit. Each is `git add -- ` then a separate `git commit`, one command per invocation, never chained, never `git add -A`. No `-m` paragraph contains an angle bracket, a dollar sign or a backtick; an attribution trailer, when the session requires one, is a second `-m` paragraph with the address written bare. A PreToolUse refusal of any `git add`, `git commit`, `.cs` edit or `.csproj` edit is reported verbatim as `PRE-IMPLEMENTATION GATE BLOCKED` and stops the run; the executor does not modify hooks, checkpoints or permission configuration. +- **D-10 Git gates are pathspec-scoped and anchored.** Every `git diff` carries the anchor as ref operand; every name-listing diff is paired with a porcelain span in the same task; no gate asserts an empty unscoped porcelain. Porcelain gates after a commit admit this plan file (its check-off mark is written after each commit) and assert scope (no entry under TaskMaster/ or TaskMaster.Test/), never membership or count. +- **D-11 Check-offs follow the loop.** Every acceptance criterion's evidence is either a post-format observation (line counts, hunks, tokens) or a final-run observation, and the format pass can reflow the edited files, so every check-off task sits in Phase 3 after P3-T9 and reads the artifact that survived the final pass. Each check-off task flips exactly one checkbox and completes with the box unchecked when its evidence does not hold, recording `ACn: NOT MET` in the summary. +- **D-12 Fixed artifact names.** No artifact name carries a timestamp; acceptance conditions name files exactly. The write time is the `Timestamp:` field (ISO yyyy-MM-ddTHH-mm). + +## Delivered source (the executor writes these texts; CSharpier output wins on any layout difference) + +Indentation rule: the production-file and primary-fixture blocks below are shown at their in-file indentation (eight, twelve or twenty leading spaces, matching the lines they replace or follow), so they are written exactly as shown. The new-partial block is shown with four leading spaces of Markdown indent on every line; those four spaces are removed on every line when the file is written, so `using` and `namespace` sit at column 0 and the class body keeps the four-space steps shown relative to them. + +**Production file `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`.** + +`GetPrimeTask` `returns` element (replaces lines 242 to 246): + + /// + /// The prime task, or when no prime has been started for + /// the key. The returned task never faults: a prime fault is observed inside the prime + /// itself and reported through logError. For a key whose prime did not run to + /// completion, the marker is cleared only after that report has returned, so a caller that + /// receives can rely on the fault having been reported. + /// + +`CompletePrime` `summary` element (replaces lines 327 to 331): + + /// + /// Observes the outcome of a prime. On any outcome other than ran-to-completion the cache + /// is left unset — so the key still reports unchecked — the failure is reported through + /// logError, and only then is the in-flight marker cleared so a later read may + /// re-prime. + /// + +`CompletePrime` body (replaces lines 341 to 355; the `remarks` element between is unchanged): + + private void CompletePrime(Task completed, string engineName) + { + if (completed.Status == TaskStatus.RanToCompletion) + { + return; + } + + var failure = + (Exception)completed.Exception?.GetBaseException() + ?? new TaskCanceledException(completed); + + // Report-then-clear is load-bearing: the marker stays registered until the report has + // returned, so a caller that observes the marker absent — including one that fetched the + // prime handle after the fault — is guaranteed the fault has already been reported. + _logError(BuildPrimeFailedMessage(engineName), failure); + _primeTasks.TryRemove(engineName, out _); + } + +**Primary fixture `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`.** + +Error-log lambda (replaces line 415): + + (message, exception) => + { + Errors.Add(new LoggedError(message, exception)); + OnLogError?.Invoke(message, exception); + } + +Hook property: insert one blank line and then the six hook lines after line 434 (the `OnInvalidate` property's line), so the existing blank line that preceded `Invalidations` now follows the hook: + + /// + /// An optional extra observer invoked from inside the error-log sink, immediately after + /// the error has been appended to , so a test can probe coordinator + /// state at the exact moment a fault is reported. + /// + internal Action OnLogError { get; set; } + +**New partial `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (whole text).** + + using System; + using System.Threading.Tasks; + using FluentAssertions; + using Microsoft.VisualStudio.TestTools.UnitTesting; + + namespace TaskMaster.Test.Ribbon + { + /// + /// Regression for issue #942: the prime-fault report must precede the in-flight marker's + /// removal, so any caller that observes the marker absent observes a report that has already + /// completed. A third partial of the coordinator fixture, so the private Harness and + /// LoggedError types are reused; the primary file is close to the 500-line ceiling. + /// + public partial class EngineToggleStateCoordinatorTests + { + #region Issue #942 — prime fault report precedes marker removal + + /// + /// Regression for issue #942. Invariant: for a key whose prime did not run to completion, + /// the in-flight marker is present until the fault report has returned. The discriminator + /// is the prime handle observed from inside the error-log sink: it is the still-registered + /// continuation under the fixed order and under the + /// defective one, on the same thread, so the outcome is a function of program order + /// rather than of scheduling. No sleep, delay, gate, timer or parallelism attribute. + /// + [TestMethod] + public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged() + { + // Arrange + var harness = new Harness(); + var probe = new TaskCompletionSource(); + var failure = new InvalidOperationException("configuration load failed"); + harness.Engines.Setup(x => x.EngineActiveAsync(SpamEngine)).Returns(probe.Task); + harness.Coordinator.GetPressed(SpamEngine); + var prime = harness.Coordinator.GetPrimeTask(SpamEngine); + Task handleSeenBySink = null; + harness.OnLogError = (_, _) => + handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine); + + // Act + probe.SetException(failure); + await prime; + + // Assert + // If this test passes without the production reorder in CompletePrime, the negative + // control has lost isolation: investigate the run rather than accepting it. + handleSeenBySink + .Should() + .BeSameAs( + prime, + "while the fault is being reported the prime handle " + + "must still be registered, so a caller that fetches it " + + "after the trigger awaits the report" + ); + harness.Errors.Should().ContainSingle("a prime fault is reported exactly once"); + harness + .Errors[0] + .Message.Should() + .Contain(SpamEngine, "the message names the engine whose prime failed"); + harness + .Errors[0] + .Exception.Should() + .BeSameAs(failure, "the sink receives the injected exception unchanged"); + harness.Invalidations.Should().BeEmpty("a failed prime leaves nothing to display"); + harness + .Coordinator.GetPrimeTask(SpamEngine) + .Should() + .BeSameAs( + Task.CompletedTask, + "once the handle has completed the marker has been cleared " + + "so a later read may re-prime" + ); + } + + #endregion Issue #942 — prime fault report precedes marker removal + } + } + +**Project file `TaskMaster.Test/TaskMaster.Test.csproj`.** + +One line inserted immediately after line 359: `` with the same four-space indentation as its neighbours. + +## Execution conventions + +- **Working directory and paths.** `WORKTREE` denotes the absolute path of the item worktree supplied in the delegation prompt; it is substituted into every payload's first line and is never written into an artifact. Every artifact records repository-relative paths only. No artifact, and no line of this plan, carries an absolute host path, an account name or a machine name. +- **Payload channel.** Each indented payload block below is executed as one PowerShell 7 invocation (`pwsh -NoProfile -Command` with the payload in single quotes, or the session's PowerShell tool), with the worktree as the current directory. Payloads use double quotes only, so the outer single quotes never conflict. The `Command:` field of the artifact records the canonical command the payload runs (named in each payload's note), not the payload text. +- **Exit codes.** `EXIT_CODE:` records the printed `_EXIT_CODE:` value of the payload's principal command. Deliberately failing runs carry `ExpectedExitCode:` equal to the observed non-zero value. A task that runs several commands names one as the row and records the others as named `Output Summary:` lines. +- **Tool resolution.** MSBuild and vstest.console.exe are resolved through vswhere inside each payload (`TOOLS` prelude below); the resolved paths are used, never printed into an artifact. +- **TOOLS prelude** (the first lines of every build and test payload after the `Set-Location`): + + $vswhere = Join-Path ${env:ProgramFiles(x86)} "Microsoft Visual Studio\Installer\vswhere.exe" + $msbuild = & $vswhere -latest -products * -find "MSBuild\**\Bin\MSBuild.exe" | Select-Object -First 1 + $vstest = & $vswhere -latest -products * -find "Common7\IDE\Extensions\TestPlatform\vstest.console.exe" | Select-Object -First 1 + New-Item -ItemType Directory -Path "coverage\logs" -Force | Out-Null + +- **Stall handling.** Every direct vstest run carries the hang-dump blame switch (CollectHangDump, TestTimeout 4min, HangDumpType None, spelled out in the CMD-VSTEST payload), so a stalled test is named in a Sequence document under the results directory; a run that produces one is recorded as failed with that test name. The RUNNER coverage route passes no blame argument (fixed argument list), so P0-T14 and P3-T8 bound it by wall clock: a run still in progress after 120 minutes is `COVERAGE RUN STALLED`: stop and report. +- **Long-running payloads.** `CMD-COVERAGE-RUNNER`, `CMD-COVERAGE-DIRECT` and any payload expected to exceed 8 minutes are started as background processes with the payload's own standard output redirected to `coverage\logs\.result.log`; completion is detected by polling that file for the payload's final line, `PAYLOAD-COMPLETE`, which both coverage payloads print last, rather than by waiting on a foreground tool call. The collector's or runner's own output is tee'd to the log the payload names. If a foreground attempt times out anyway, the executor runs `pwsh -NoProfile -Command '"STRAY_TEST_PROCESSES: " + @(Get-Process -ErrorAction SilentlyContinue | Where-Object { $_.ProcessName -like "vstest*" -or $_.ProcessName -like "testhost*" -or $_.ProcessName -like "dotnet-coverage*" }).Count'` and proceeds to any rerun only when it prints `STRAY_TEST_PROCESSES: 0`; a non-zero value is recorded and the executor waits for the processes to exit before rerunning, never runs two collections at once. +- **Restart rule (Phase 3).** No code file is edited after the P2-T8 commit. If any of P3-T1 through P3-T8 fails or rewrites a file, the run stops and reports the failing step with its artifact; there is no in-plan repair. The only admitted repeat is P3-T8's single same-command re-run for the issue 780 sporadic failure, after which P3-T9 records both attempts. + +## Command reference + +**CMD-REBUILD** (`GATEARGS` is either the analyzer pair, EnableNETAnalyzers true with EnforceCodeStyleInBuild true, or the nullable switch, TreatWarningsAsErrors true, each written in the msbuild property form the `Command:` field quotes; `TASKID` substituted; the `Command:` field records `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" GATEARGS`, resolved through vswhere, plus /nodeReuse:false and a normal-verbosity file logger under the ignored coverage directory): + + Set-Location -LiteralPath "WORKTREE" + TOOLS + $log = "coverage\logs\TASKID.msbuild.log" + if (Test-Path -LiteralPath $log) { Remove-Item -LiteralPath $log -Force } + $global:LASTEXITCODE = 0 + & $msbuild TaskMaster.sln /t:Rebuild /m /nodeReuse:false /p:Configuration=Debug "/p:Platform=Any CPU" GATEARGS "/flp:LogFile=$log;Verbosity=normal" | Out-Null + Write-Output ("MSBUILD_EXIT_CODE: " + $LASTEXITCODE) + $lines = Get-Content -LiteralPath $log -Encoding UTF8 + Write-Output ("ERRORS: " + (($lines | Select-String -Pattern "^\s*(\d+) Error\(s\)" | Select-Object -Last 1).Matches[0].Groups[1].Value)) + Write-Output ("WARNINGS: " + (($lines | Select-String -Pattern "^\s*(\d+) Warning\(s\)" | Select-Object -Last 1).Matches[0].Groups[1].Value)) + Write-Output ("SKIP_CORECOMPILE_LINES: " + @($lines | Where-Object { $_.Contains("Skipping target ""CoreCompile""") }).Count) + Write-Output ("CSC_OUT_TASKMASTER: " + @($lines | Where-Object { $_.Contains("/out:obj\Debug\TaskMaster.dll") }).Count) + Write-Output ("CSC_OUT_TASKMASTER_TEST: " + @($lines | Where-Object { $_.Contains("/out:obj\Debug\TaskMaster.Test.dll") }).Count) + Write-Output ("WRITESET_DIAGNOSTIC_LINES: " + @($lines | Where-Object { ($_.Contains("EngineToggleStateCoordinator")) -and ($_ -match "(error|warning) [A-Z]+\d+") }).Count) + Write-Output ("TEST_DLL_EXISTS: " + (Test-Path -LiteralPath "TaskMaster.Test\bin\Debug\TaskMaster.Test.dll")) + Write-Output ("UCS_TEST_DLL_EXISTS: " + (Test-Path -LiteralPath "UtilitiesCS.Test\bin\Debug\UtilitiesCS.Test.dll")) + +Under /t:Rebuild the `SKIP_CORECOMPILE_LINES` count is 0 by construction; the two `CSC_OUT_` counts are the non-vacuity observation that the compiler ran for the two Write Set projects (MSBuild echoes the csc command line under each project's CoreCompile heading at normal verbosity). `ERRORS:` is read from the summary line, so `0 Error(s)` is not mistaken for a substring of a larger count. + +**CMD-BUILD** (plain incremental build so that a scoped test run observes a fresh assembly; `TASKID` substituted; `Command:` records `msbuild TaskMaster.sln /t:Build /m /p:Configuration=Debug "/p:Platform=Any CPU"`): + + Set-Location -LiteralPath "WORKTREE" + TOOLS + $log = "coverage\logs\TASKID.msbuild.log" + if (Test-Path -LiteralPath $log) { Remove-Item -LiteralPath $log -Force } + $before = (Get-Item -LiteralPath "TaskMaster.Test\bin\Debug\TaskMaster.Test.dll" -ErrorAction SilentlyContinue).LastWriteTimeUtc + $global:LASTEXITCODE = 0 + & $msbuild TaskMaster.sln /t:Build /m /nodeReuse:false /p:Configuration=Debug "/p:Platform=Any CPU" "/flp:LogFile=$log;Verbosity=normal" | Out-Null + Write-Output ("MSBUILD_EXIT_CODE: " + $LASTEXITCODE) + $lines = Get-Content -LiteralPath $log -Encoding UTF8 + Write-Output ("ERRORS: " + (($lines | Select-String -Pattern "^\s*(\d+) Error\(s\)" | Select-Object -Last 1).Matches[0].Groups[1].Value)) + $after = (Get-Item -LiteralPath "TaskMaster.Test\bin\Debug\TaskMaster.Test.dll").LastWriteTimeUtc + Write-Output ("TEST_DLL_ADVANCED: " + ($null -eq $before -or $after -gt $before)) + Write-Output ("CSC_OUT_TASKMASTER_TEST: " + @($lines | Where-Object { $_.Contains("/out:obj\Debug\TaskMaster.Test.dll") }).Count) + +**CMD-VSTEST** (`ASSEMBLY`, `FILTER`, `TASKID` and the `NAMES` list substituted; `Command:` records `vstest.console.exe ASSEMBLY /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FILTER" "/ResultsDirectory:coverage\test-results\942\TASKID" "/Logger:trx;LogFileName=TASKID.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None"`, resolved through vswhere): + + Set-Location -LiteralPath "WORKTREE" + TOOLS + $results = "coverage\test-results\942\TASKID" + if (Test-Path -LiteralPath $results) { Remove-Item -LiteralPath $results -Recurse -Force } + $names = @(NAMES) + $global:LASTEXITCODE = 0 + & $vstest "ASSEMBLY" /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FILTER" "/ResultsDirectory:$results" "/Logger:trx;LogFileName=TASKID.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" 2>&1 | Tee-Object -FilePath "coverage\logs\TASKID.vstest.log" | Out-Null + Write-Output ("VSTEST_EXIT_CODE: " + $LASTEXITCODE) + $trxPath = Join-Path $results "TASKID.trx" + Write-Output ("TRX_PRESENT: " + (Test-Path -LiteralPath $trxPath)) + Write-Output ("SEQUENCE_FILES: " + @(Get-ChildItem -LiteralPath $results -Recurse -Filter "Sequence_*.xml" -ErrorAction SilentlyContinue).Count) + if (-not (Test-Path -LiteralPath $trxPath)) { exit 3 } + [xml]$trx = Get-Content -LiteralPath $trxPath -Raw -Encoding UTF8 + $ns = New-Object System.Xml.XmlNamespaceManager($trx.NameTable) + $ns.AddNamespace("t", "http://microsoft.com/schemas/VisualStudio/TeamTest/2010") + $counters = $trx.SelectSingleNode("//t:ResultSummary/t:Counters", $ns) + Write-Output ("COUNTERS total=" + $counters.GetAttribute("total") + " executed=" + $counters.GetAttribute("executed") + " passed=" + $counters.GetAttribute("passed") + " failed=" + $counters.GetAttribute("failed")) + $all = @($trx.SelectNodes("//t:UnitTestResult", $ns)) + Write-Output ("RESULT_COUNT: " + $all.Count) + foreach ($r in $all) { if ($names -contains $r.GetAttribute("testName")) { Write-Output ("RESULT " + $r.GetAttribute("testName") + " = " + $r.GetAttribute("outcome")) } } + foreach ($r in $all) { if ($r.GetAttribute("outcome") -eq "Failed") { Write-Output ("FAILED " + $r.GetAttribute("testName")); $msg = $r.SelectSingleNode("t:Output/t:ErrorInfo/t:Message", $ns); Write-Output ("MESSAGE " + $r.GetAttribute("testName") + " :: " + $(if ($msg) { $msg.InnerText } else { "(no message)" })) } } + +The trx stays under the ignored coverage directory. The artifact transcribes the `COUNTERS`, `RESULT_COUNT:`, `RESULT`, `FAILED` and `MESSAGE` lines (absolute paths inside a message are replaced by the placeholder REDACTED-PATH before transcription). + +Substitutions used by this plan: `ASSEMBLY-TM` is the built test assembly TaskMaster.Test\bin\Debug\TaskMaster.Test.dll; `ASSEMBLY-UCS` is the built test assembly UtilitiesCS.Test\bin\Debug\UtilitiesCS.Test.dll; `FILTER-COORD` is `FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests`; `FILTER-STALL` is `FullyQualifiedName~HelperClasses.ShellUtilities_Tests|FullyQualifiedName~HelperClasses.ShellUtilitiesStatic_Tests|FullyQualifiedName~HelperClasses.SysImageListHelperTests|FullyQualifiedName~EmailIntelligence.OSBrowser_Tests`; `NAMES-COORD` is `"GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse", "GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged", "GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker", "GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked"`; `NAMES-NONE` is empty. + +**CMD-COVERAGE-RUNNER** (CLAUDE.md step 4 route; `STAGE` is `baseline` or `final`; `Command:` records `pwsh -NoProfile -File scripts\vscode\Invoke-MSTestWithCoverage.ps1`): + + Set-Location -LiteralPath "WORKTREE" + New-Item -ItemType Directory -Path "coverage\logs" -Force | Out-Null + foreach ($f in @("coverage\coverage.cobertura.xml", "coverage\coverage.cobertura.jacoco.xml", "coverage\test-results\mstest-coverage-run.trx", "coverage\test-results\mstest-coverage-run.summary.txt")) { if (Test-Path -LiteralPath $f) { Remove-Item -LiteralPath $f -Force } } + foreach ($f in @("coverage\STAGE-942.cobertura.xml", "coverage\STAGE-942.trx")) { if (Test-Path -LiteralPath $f) { Remove-Item -LiteralPath $f -Force } } + $script = Join-Path (Get-Location).Path "scripts\vscode\Invoke-MSTestWithCoverage.ps1" + $global:LASTEXITCODE = 0 + & pwsh -NoProfile -File $script 2>&1 | Tee-Object -FilePath "coverage\logs\STAGE-942.runner.log" | Out-Null + Write-Output ("RUNNER_EXIT_CODE: " + $LASTEXITCODE) + $log = Get-Content -LiteralPath "coverage\logs\STAGE-942.runner.log" -Raw -Encoding UTF8 + Write-Output ("DISCOVERED_LINE: " + [regex]::Match($log, "Discovered \d+ test assemblies\.").Value) + Write-Output ("FIRST_PARTY_LINE: " + [regex]::Match($log, "First-party coverage: [^\r\n]*").Value) + Write-Output ("THRESHOLD_MESSAGE: " + [regex]::Match($log, "Cobertura (line|branch) coverage [^\r\n]*threshold\.").Value) + Write-Output ("COLLECT_FAILURE_MESSAGE: " + [regex]::Match($log, "MSTest with coverage failed with exit code \d+").Value) + Write-Output ("DOCUMENT_PRESENT: " + (Test-Path -LiteralPath "coverage\coverage.cobertura.xml")) + Write-Output ("TRX_PRESENT: " + (Test-Path -LiteralPath "coverage\test-results\mstest-coverage-run.trx")) + if (Test-Path -LiteralPath "coverage\coverage.cobertura.xml") { Copy-Item -LiteralPath "coverage\coverage.cobertura.xml" -Destination "coverage\STAGE-942.cobertura.xml" -Force } + if (Test-Path -LiteralPath "coverage\test-results\mstest-coverage-run.trx") { Copy-Item -LiteralPath "coverage\test-results\mstest-coverage-run.trx" -Destination "coverage\STAGE-942.trx" -Force } + Write-Output "PAYLOAD-COMPLETE" + +The runner's lines naming the resolved vstest path and the coverage output carry absolute paths and stay in the ignored log; only the named `_LINE`, `_MESSAGE` and `_PRESENT` values are transcribed. The stale-output removal makes every `_PRESENT` value an observation of this run. + +**CMD-COVERAGE-DIRECT** (the runner's inner invocation issued directly with the four-class exclusion; `STAGE` substituted; `Command:` records `dotnet-coverage collect --output coverage\STAGE-942.cobertura.xml --output-format cobertura --settings coverage\effective-coverage-942.config -- vstest.console.exe /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:" "/ResultsDirectory:coverage\test-results\942\STAGE" "/Logger:trx;LogFileName=STAGE-942.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None"`): + + Set-Location -LiteralPath "WORKTREE" + . (Join-Path (Get-Location).Path "scripts\vscode\Invoke-MSTestWithCoverage.ps1") + $ErrorActionPreference = "Continue" + $repo = (Get-Location).Path + New-Item -ItemType Directory -Path "coverage\logs" -Force | Out-Null + foreach ($f in @("coverage\STAGE-942.cobertura.xml", "coverage\STAGE-942.trx")) { if (Test-Path -LiteralPath $f) { Remove-Item -LiteralPath $f -Force } } + $canonical = Get-Content -LiteralPath "coverage.config" -Raw -Encoding UTF8 + $derived = ConvertTo-DerivedCoverageSettingsXml -CanonicalSettingsXml $canonical + $effective = Join-Path $repo "coverage\effective-coverage-942.config" + Set-Content -LiteralPath $effective -Value $derived -Encoding UTF8 -NoNewline + $vswhere = Join-Path ${env:ProgramFiles(x86)} "Microsoft Visual Studio\Installer\vswhere.exe" + $vstest = & $vswhere -latest -products * -find "Common7\IDE\Extensions\TestPlatform\vstest.console.exe" | Select-Object -First 1 + $rootLen = $repo.TrimEnd([char]92).Length + $asm = @(Get-ChildItem -Path $repo -Recurse -Filter "*.Test.dll" | Where-Object { $_.FullName -like "*\bin\Debug\*" -and $_.FullName -notlike "*\obj\*" -and $_.FullName -notlike "*\ref\*" -and $_.FullName.Substring($rootLen) -notlike "\.claude\*" } | Select-Object -ExpandProperty FullName) + $filter = "TestCategory!=LiveOutlook&FullyQualifiedName!~HelperClasses.ShellUtilities_Tests&FullyQualifiedName!~HelperClasses.ShellUtilitiesStatic_Tests&FullyQualifiedName!~HelperClasses.SysImageListHelperTests&FullyQualifiedName!~EmailIntelligence.OSBrowser_Tests" + $output = Join-Path $repo "coverage\STAGE-942.cobertura.xml" + $settings = Join-Path $repo "scripts\vscode\TaskMaster.cli.runsettings" + $results = Join-Path $repo "coverage\test-results\942\STAGE" + if (Test-Path -LiteralPath $results) { Remove-Item -LiteralPath $results -Recurse -Force } + $global:LASTEXITCODE = 0 + & dotnet-coverage collect --output $output --output-format cobertura --settings $effective -- $vstest @asm "/Settings:$settings" /InIsolation "/TestCaseFilter:$filter" "/ResultsDirectory:$results" "/Logger:trx;LogFileName=STAGE-942.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None" 2>&1 | Tee-Object -FilePath "coverage\logs\STAGE-942.collect.log" | Out-Null + Write-Output ("COLLECT_EXIT_CODE: " + $LASTEXITCODE) + Write-Output ("ASSEMBLY_COUNT: " + $asm.Count) + $asm | ForEach-Object { Write-Output ("ASSEMBLY: " + $_.Substring($rootLen)) } + Write-Output ("SEQUENCE_FILES: " + @(Get-ChildItem -LiteralPath $results -Recurse -Filter "Sequence_*.xml" -ErrorAction SilentlyContinue).Count) + if (Test-Path -LiteralPath (Join-Path $results "STAGE-942.trx")) { Copy-Item -LiteralPath (Join-Path $results "STAGE-942.trx") -Destination "coverage\STAGE-942.trx" -Force } + Write-Output ("TRX_PRESENT: " + (Test-Path -LiteralPath "coverage\STAGE-942.trx")) + Write-Output ("DOCUMENT_PRESENT: " + (Test-Path -LiteralPath $output)) + Write-Output "PAYLOAD-COMPLETE" + +**CMD-COVERAGE-POST** (post-process if raw, summarise the trx, apply the floors, print the first-party line, the projection, the root counters and the coordinator figures; `STAGE` substituted; `RAW` is `True` under DIRECT and under a RUNNER run whose `COLLECT_FAILURE_MESSAGE:` is non-empty, otherwise `False`, because a completed runner run has already post-processed the document in place): + + Set-Location -LiteralPath "WORKTREE" + . (Join-Path (Get-Location).Path "scripts\vscode\Invoke-MSTestWithCoverage.Helpers.ps1") + . (Join-Path (Get-Location).Path "scripts\vscode\Invoke-MSTest.TrxSummary.ps1") + $ErrorActionPreference = "Continue" + $repo = (Get-Location).Path + $summary = Get-TrxRunSummary -TrxContent (Get-Content -LiteralPath "coverage\STAGE-942.trx" -Raw -Encoding UTF8) + Write-Output "SUMMARY-BEGIN" + Write-Output (Format-TrxRunSummary -Summary $summary) + Write-Output "SUMMARY-END" + Write-Output ("FAILED-SET: " + (@($summary.FailedTestName) -join ", ")) + $doc = Get-Content -LiteralPath "coverage\STAGE-942.cobertura.xml" -Raw -Encoding UTF8 + if ("RAW" -eq "True") { $doc = ConvertTo-KoverageCoberturaXml -XmlContent $doc -RepoRoot $repo; Set-Content -LiteralPath "coverage\STAGE-942.cobertura.xml" -Value $doc -Encoding UTF8 -NoNewline } + try { Assert-CoberturaLineCoverageThreshold -CoberturaXml $doc; Write-Output "LINE-FLOOR: MET" } catch { Write-Output ("LINE-FLOOR: NOT MET " + $_.Exception.Message) } + try { Assert-CoberturaBranchCoverageThreshold -CoberturaXml $doc; Write-Output "BRANCH-FLOOR: MET" } catch { Write-Output ("BRANCH-FLOOR: NOT MET " + $_.Exception.Message) } + Write-Output (Get-CoberturaFirstPartyCoverageReport -CoberturaXml $doc) + [xml]$xml = $doc + $root = $xml.SelectSingleNode("/coverage") + Write-Output ("ROOT line-rate=" + $root.GetAttribute("line-rate") + " branch-rate=" + $root.GetAttribute("branch-rate") + " lines-covered=" + $root.GetAttribute("lines-covered") + " lines-valid=" + $root.GetAttribute("lines-valid") + " branches-covered=" + $root.GetAttribute("branches-covered") + " branches-valid=" + $root.GetAttribute("branches-valid")) + $projection = ConvertTo-JacocoPackageProjection -XmlDocument $xml + Assert-JacocoProjectionReconciliation -XmlDocument $xml -ProjectionXml $projection + Write-Output "PROJECTION-BEGIN" + Write-Output $projection + Write-Output "PROJECTION-END" + $target = "TaskMaster/Ribbon/EngineToggleStateCoordinator.cs" + $classes = @($xml.SelectNodes("//class[@filename]") | Where-Object { $_.GetAttribute("filename").Replace([string][char]92, "/").EndsWith($target) }) + Write-Output ("COORD-CLASS-NODES: " + $classes.Count) + if ($classes.Count -eq 1) { + $s = Get-CoberturaClassLineSummary -ClassNode $classes[0] + Write-Output ("COORD-LINES covered=" + $s.CoveredLines + " valid=" + $s.TotalLines) + Write-Output ("COORD-BRANCHES covered=" + $s.CoveredBranches + " valid=" + $s.TotalBranches) + $src = Get-Content -LiteralPath "TaskMaster\Ribbon\EngineToggleStateCoordinator.cs" -Encoding UTF8 + $start = 0; for ($i = 0; $i -lt $src.Count; $i++) { if ($src[$i].Contains("private void CompletePrime(")) { $start = $i + 1; break } } + $end = 0; for ($i = $start; $i -lt $src.Count; $i++) { if ($src[$i].TrimEnd() -eq " }") { $end = $i + 1; break } } + Write-Output ("COMPLETEPRIME-SPAN: " + $start + "-" + $end) + $inSpan = @($s.LineMap.Keys | Where-Object { $_ -ge $start -and $_ -le $end } | Sort-Object) + Write-Output ("COMPLETEPRIME-LINE-ELEMENTS: " + $inSpan.Count) + foreach ($n in $inSpan) { Write-Output ("COMPLETEPRIME-LINE " + $n + " hits=" + $s.LineMap[$n].Hits) } + Write-Output ("COMPLETEPRIME-UNCOVERED: " + @($inSpan | Where-Object { $s.LineMap[$_].Hits -lt 1 }).Count) + } + +The `COMPLETEPRIME-SPAN` is derived from the source file as it stands when the payload runs (the anchored file in Phase 0, the fixed file in Phase 3), so each stage measures its own statement set. The projection and the summary block are the two CLAUDE.md committed forms; the `COORD-` and `COMPLETEPRIME-` lines are figures, not documents. + +**CMD-HASH** (SHA-256 of the three formatter-visible Write Set files; hashes only, never the Path property): + + Set-Location -LiteralPath "WORKTREE" + foreach ($p in @("TaskMaster\Ribbon\EngineToggleStateCoordinator.cs", "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs", "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs")) { if (Test-Path -LiteralPath $p) { Write-Output ("HASH " + $p + " = " + (Get-FileHash -Algorithm SHA256 -LiteralPath $p).Hash) } else { Write-Output ("HASH " + $p + " = ABSENT") } } + +**CMD-LINECOUNT** (content line counts of the three source files): + + Set-Location -LiteralPath "WORKTREE" + foreach ($p in @("TaskMaster\Ribbon\EngineToggleStateCoordinator.cs", "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs", "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs")) { if (Test-Path -LiteralPath $p) { Write-Output ("LINES " + $p + " = " + @(Get-Content -LiteralPath $p -Encoding UTF8).Count) } else { Write-Output ("LINES " + $p + " = ABSENT") } } + +**CMD-TOKEN-COUNT** (`FILE` and the `TOKEN` list substituted; ordinal, case-sensitive substring counts per physical line, so a wrapped token reads 0 and the plan's single-line requirement is enforced by the count): + + Set-Location -LiteralPath "WORKTREE" + $src = Get-Content -LiteralPath "FILE" -Encoding UTF8 + foreach ($t in @(TOKEN)) { Write-Output ("TOKEN [" + $t + "] = " + @($src | Where-Object { $_.Contains($t) }).Count) } + foreach ($t in @(TOKEN)) { $idx = 0; for ($i = 0; $i -lt $src.Count; $i++) { if ($src[$i].Contains($t)) { $idx = $i + 1; break } }; Write-Output ("FIRST-LINE [" + $t + "] = " + $idx) } + +### Phase 0 — Policy Reads, Bootstrap, Anchor and Baseline Capture + +- [x] [P0-T1] Read the policy documents in the mandatory order — CLAUDE.md, then .claude/rules/general-code-change.md, then .claude/rules/general-unit-test.md, then .claude/rules/csharp.md — plus .claude/rules/plan-acceptance-gates.md and .claude/rules/tonality.md, and record the read in FEATURE/evidence/baseline/phase0-instructions-read.md. + - Acceptance: the artifact carries `Timestamp:`, a `Policy Order:` line naming the four mandatory documents in that order, and one line per document read recording its top-level heading count. No policy document is modified. +- [x] [P0-T2] Read the acceptance-criteria source `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, the issue document `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md` and the research record `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md` in full, and record the Write Set and the prohibited paths in FEATURE/evidence/baseline/scope-and-anchor.md (this task creates the file; P0-T4 appends to it). + - Acceptance: the artifact lists the four code paths of the Write Set verbatim, names the eight prohibited paths and trees from the Write Set section, records that issue.md line 12 reads `- Work Mode: full-bug`, and records that the spec's acceptance section holds exactly 14 lines beginning `- [ ] AC` and 0 lines beginning `- [x] AC`, counted from the file. +- [x] [P0-T3] Provision the repository .NET SDK with scripts/vscode/Install-RepoDotNetSdk.ps1 (guarded) and record it in FEATURE/evidence/baseline/bootstrap-sdk.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; if (-not (Test-Path -LiteralPath ".dotnet-sdk\sdk\8.0.205")) { & .\scripts\vscode\Install-RepoDotNetSdk.ps1 }; "SDK_MARKER=$(Test-Path -LiteralPath ".dotnet-sdk\sdk\8.0.205")"; dotnet --version'` + - Acceptance: `SDK_MARKER=True`, `dotnet --version` printed a version string rather than the global.json error message, `EXIT_CODE: 0`. The installer's filesystem marker is the gate; the version equality is not asserted because global.json rolls forward within the feature band. +- [x] [P0-T4] Record the anchor and the pre-change tree state by appending to FEATURE/evidence/baseline/scope-and-anchor.md. + - Command: `git rev-parse HEAD`; `git merge-base --is-ancestor 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD`; `git merge-base origin/main HEAD`; `git diff --name-status 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD`; `git status --porcelain --untracked-files=all`. + - Acceptance, all required: `HEAD-SHA:` records the first command's output as an observation (no expected value); the ancestor check exits 0 and `git merge-base origin/main HEAD` prints exactly `231e1c0b55105aeb626bf5a6e8d0266a567cacad`, otherwise the artifact records `BASE-SHA MISMATCH` with both values and the run stops; `INHERITED-COMMITTED:` lists every path the name-status diff prints with its status letter, or `NONE`, and every listed path is either under `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/` or is exactly `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` (the promotion record the amended AC13 exempts; fact 14 expects it and the documentation commits' feature-folder paths, nothing else) — any other path is `INHERITED SET EXCEEDS AC13 EXEMPTION`: the artifact records the offending paths and the run stops; `PRE-EXISTING-WORKTREE-PATHS:` lists every porcelain line verbatim or `NONE`; and no porcelain line names a path under TaskMaster/ or TaskMaster.Test/ (otherwise `CODE TREE DIRTY AT ANCHOR`: stop and report). The porcelain output is not asserted empty: this feature folder and uncommitted .claude/agent-memory/ files are expected in it. +- [x] [P0-T5] Restore the manifest tools with `dotnet tool restore` at the repository root (the manifest is dotnet-tools.json at the root) and record it in FEATURE/evidence/baseline/bootstrap-tool-restore.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool restore; "RESTORE_EXIT=$LASTEXITCODE"; dotnet tool list --local; dotnet tool run csharpier check --help | Out-Null; "CHECK_HELP_EXIT=$LASTEXITCODE"'` + - Acceptance: `RESTORE_EXIT=0`, the local tool list contains a row whose Package Id is `csharpier` and whose Version is `1.2.6`, and `CHECK_HELP_EXIT=0`. The artifact transcribes only the Package Id and Version columns of the `dotnet tool list --local` rows (the Manifest column carries an absolute path and is not transcribed). The version is read from the list row, not from a version switch, because the pinned tool's version switch has not been observed on this manifest. +- [x] [P0-T6] Restore NuGet packages with scripts/vscode/Invoke-Restore.ps1 and record it in FEATURE/evidence/baseline/bootstrap-nuget-restore.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $env:MSBUILDDISABLENODEREUSE = "1"; & .\scripts\vscode\Invoke-Restore.ps1; "RESTORE_EXIT=$LASTEXITCODE"; "PACKAGE_DIRS=$(@(Get-ChildItem -LiteralPath packages -Directory -ErrorAction SilentlyContinue).Count)"; foreach ($proj in @("TaskMaster\TaskMaster.csproj", "TaskMaster.Test\TaskMaster.Test.csproj")) { $dir = Split-Path -Parent $proj; [xml]$x = Get-Content -LiteralPath $proj -Raw; $missing = @($x.SelectNodes("//*[local-name()=""Analyzer""]") | Where-Object { -not (Test-Path -LiteralPath (Join-Path $dir $_.GetAttribute("Include"))) }).Count; "ANALYZER_MISSING $proj = $missing" }'` + - Acceptance: `RESTORE_EXIT=0`, `PACKAGE_DIRS=` at least 1, and both `ANALYZER_MISSING` values are 0 (every analyzer Include of the two Write Set projects resolves relative to its own project directory). A non-zero `ANALYZER_MISSING` is `ANALYZER PATH SKEW`: stop and report the unresolved Include values; it is an environment defect, not a plan defect. +- [x] [P0-T7] Provision the dotnet-coverage global tool (guarded) and record it in FEATURE/evidence/baseline/bootstrap-dotnet-coverage.md. + - Command: `pwsh -NoProfile -Command 'if (-not (Get-Command dotnet-coverage -ErrorAction SilentlyContinue)) { dotnet tool install --global dotnet-coverage }; "DOTNET_COVERAGE_RESOLVED=$($null -ne (Get-Command dotnet-coverage -ErrorAction SilentlyContinue))"; dotnet-coverage --version'` + - Acceptance: `DOTNET_COVERAGE_RESOLVED=True`, a version line is printed, `EXIT_CODE: 0`. +- [x] [P0-T8] Capture the read-only formatter baseline with `dotnet tool run csharpier check .` and record the verbatim unformatted-file set in FEATURE/evidence/baseline/csharpier-check-baseline.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool run csharpier check .; "CSHARPIER_EXIT_CODE: $LASTEXITCODE"'` + - Acceptance: the artifact records the printed `CSHARPIER_EXIT_CODE:` value as `EXIT_CODE:` and, when non-zero, every path CSharpier reported as unformatted, one per line. If the set is non-empty the run stops with `FORMAT BASELINE NOT CLEAN`: AC11 requires the repository-wide check to report no differences, and repairing pre-existing drift would widen the footprint beyond the Write Set, so the decision belongs to the orchestrator. `EXIT_CODE: 0` is therefore the gate. +- [x] [P0-T9] Capture the analyzer baseline with `CMD-REBUILD` (`GATEARGS` analyzers, `TASKID` p0-t9) and record it in FEATURE/evidence/baseline/msbuild-analyzer-baseline.md (`Command:` records `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true`). + - Acceptance, all required: `EXIT_CODE: 0`; `ERRORS: 0`; `SKIP_CORECOMPILE_LINES: 0`; `CSC_OUT_TASKMASTER:` and `CSC_OUT_TASKMASTER_TEST:` each at least 1; `WRITESET_DIAGNOSTIC_LINES: 0`; `WARNINGS:` recorded as `ANALYZER-BASELINE-WARNINGS:` (the comparison basis for P3-T5); `TEST_DLL_EXISTS: True` and `UCS_TEST_DLL_EXISTS: True`. A non-zero exit is `ANALYZER BASELINE NOT CLEAN`: stop and report. +- [x] [P0-T10] Capture the nullable baseline with `CMD-REBUILD` (`GATEARGS` nullable, `TASKID` p0-t10) and record it in FEATURE/evidence/baseline/msbuild-nullable-baseline.md (`Command:` records `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true`; no Nullable property override). + - Acceptance, all required: `EXIT_CODE: 0`; `ERRORS: 0`; `SKIP_CORECOMPILE_LINES: 0`; both `CSC_OUT_` counts at least 1; `WRITESET_DIAGNOSTIC_LINES: 0`; `WARNINGS:` recorded as `NULLABLE-BASELINE-WARNINGS:`; both `-DLL-EXISTS:` values `True`. A non-zero exit is `NULLABLE BASELINE NOT CLEAN`: stop and report. +- [x] [P0-T11] Record the pre-change line counts and hashes of the Write Set source files with `CMD-LINECOUNT` and `CMD-HASH` in FEATURE/evidence/baseline/file-line-counts-baseline.md. + - Acceptance: `LINES TaskMaster\Ribbon\EngineToggleStateCoordinator.cs = 415`, `LINES TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs = 459`, the PrimeFaultOrdering path reads `ABSENT` for both commands, and two `HASH` values are recorded as `BASE-HASH-PROD:` and `BASE-HASH-TEST:`. These counts are advisory; the authoritative audit is P3-T3 after the format pass. +- [x] [P0-T12] Capture the pre-change coordinator fixture run with `CMD-VSTEST` (`ASSEMBLY-TM`, `FILTER-COORD`, `TASKID` p0-t12, `NAMES-COORD`) and record it in FEATURE/evidence/baseline/coordinator-tests-baseline.md. + - Acceptance: `TRX_PRESENT: True`; `SEQUENCE_FILES: 0`; the `COUNTERS` line is present with `executed` at least 1 and is recorded as `BASELINE-COUNTERS:` with its total as `BASELINE-TOTAL:`; a `RESULT` line for `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` appears with its outcome recorded (Passed is expected; Failed is admissible here because that test is the racy original), and no `RESULT` line names `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged` (the test does not exist yet); `BASELINE-FAILED:` lists every `FAILED` name or `NONE`. `EXIT_CODE:` is recorded without a gate; when it is non-zero, `ExpectedExitCode:` carries the observed value (presentational). A non-empty `BASELINE-FAILED:` is recorded, not repaired: it is the population P2-T5 is compared against, and a failure of the original test here is consistent with the defect under repair. +- [x] [P0-T13] Run the stall probe with `CMD-VSTEST` (`ASSEMBLY-UCS`, `FILTER-STALL`, `TASKID` p0-t13, `NAMES-NONE`) and record FEATURE/evidence/baseline/stall-probe.md. + - Acceptance: the artifact records `EXIT_CODE:` (the printed `VSTEST_EXIT_CODE:`, or 3 when the trx is absent), `ExpectedExitCode:` equal to the observed value when non-zero (presentational; nothing is gated on it), `TRX_PRESENT:`, `SEQUENCE_FILES:`, the `COUNTERS` line when present and every `MESSAGE` line; then exactly one `STALL-PROBE:` line — `CLEAR` when `EXIT_CODE: 0`, `failed` is 0 and `SEQUENCE_FILES: 0`, otherwise `REPRODUCES` — and exactly one `COVERAGE-ROUTE:` line — `RUNNER` under `CLEAR`, `DIRECT` under `REPRODUCES` — with the sentence that the four excluded classes are a pre-existing local stall executed by CI (fact 13). The probe is invoked once and never re-run. Both `STALL-PROBE:` values complete this task. +- [x] [P0-T14] Capture the baseline repository-wide test-and-coverage run by the route P0-T13 fixed and record FEATURE/evidence/baseline/coverage-baseline.md (fixed name per the spec). Under `RUNNER` run `CMD-COVERAGE-RUNNER` with `STAGE` baseline; under `DIRECT` run `CMD-COVERAGE-DIRECT` with `STAGE` baseline; then, unless branch (d0) applies, run `CMD-COVERAGE-POST` with `STAGE` baseline and `RAW` per the Command Reference rule. + - Artifact: `Timestamp:`, `Command:` (the route's canonical command and the filter it applied), `EXIT_CODE:` (`RUNNER_EXIT_CODE:` or `COLLECT_EXIT_CODE:`), `ExpectedExitCode:` equal to the observed value when non-zero, and an `Output Summary:` recording `COVERAGE-ROUTE:`, `RAW:`, `DISCOVERED_LINE:` or `ASSEMBLY_COUNT:` with every `ASSEMBLY:` line, `TRX_PRESENT:`, `SEQUENCE_FILES:` (DIRECT), `THRESHOLD_MESSAGE:` and `COLLECT_FAILURE_MESSAGE:` (RUNNER), `LINE-FLOOR:`, `BRANCH-FLOOR:`, the `First-party coverage:` line, the `ROOT` line, the projection verbatim between `PROJECTION-BEGIN` and `PROJECTION-END`, the five summary lines verbatim between `SUMMARY-BEGIN` and `SUMMARY-END`, `FAILED-SET:`, `COORD-CLASS-NODES:`, `COORD-LINES`, `COORD-BRANCHES`, `COMPLETEPRIME-SPAN:`, `COMPLETEPRIME-LINE-ELEMENTS:`, every `COMPLETEPRIME-LINE` row and `COMPLETEPRIME-UNCOVERED:`. The `First-party coverage:` line is the numeric baseline headline (lines covered over valid with the percentage, branches likewise). A prospective sentence states that the planned change adds no executable statement to the coordinator, so `COORD-LINES valid=` is expected to be unchanged at P3-T10. + - Branches, checked in order: (d0) `SEQUENCE_FILES:` greater than 0 (DIRECT) or `TRX_PRESENT: False` is `COVERAGE RUN ABORTED`: stop, report the last lines of the collector log with absolute paths replaced, do not run `CMD-COVERAGE-POST`, do not re-run. (c) a `THRESHOLD_MESSAGE:` (RUNNER) or a `LINE-FLOOR: NOT MET` or `BRANCH-FLOOR: NOT MET` line is `COVERAGE FLOOR BASELINE NOT MET`: the projection is recorded and the run stops, because AC11 requires the coverage route to pass and this item changes no floor-relevant line. (b) a non-zero exit with no floor failure and a non-empty `FAILED-SET:` that is exactly the single name `TryAddValuesAsync_UpdatesExistingValue` (UtilitiesCS.Test/Extensions/DictionaryExtensions_Tests.cs line 237; a known sporadic failure tracked as issue 780, unrelated to this change) is recorded as `BASELINE-ADMISSIBLE-FAILURE:` and completes this task with `ExpectedExitCode:` equal to the observed value; any other non-empty `FAILED-SET:` is `BASELINE NOT GREEN`: stop and report the `Failed tests:` summary line (the baseline population must be green for the AC11 exit-zero requirement to be reachable). (a) exit 0 with both floors met: complete. (d) anything else, in particular a non-zero exit with an empty `FAILED-SET:`, is `COVERAGE RUN ABORTED` with the same handling as (d0). + - Acceptance, all required: branch (a) or branch (b); `COORD-CLASS-NODES: 1`; `COMPLETEPRIME-SPAN: 341-355`; `COMPLETEPRIME-LINE-ELEMENTS:` at least 4; `COMPLETEPRIME-UNCOVERED: 0`; the projection block contains a `package` element named `TaskMaster` with a `LINE` and a `BRANCH` counter; the summary block's first line begins `Test run outcome:`; the artifact contains no absolute path. coverage\baseline-942.cobertura.xml and coverage\baseline-942.trx remain on disk, git-ignored, for P3-T10. +- [x] [P0-T15] Verify the evidence completeness of Phase 0 by listing FEATURE/evidence/baseline/ and record the listing at the end of FEATURE/evidence/baseline/scope-and-anchor.md under a `PHASE0-ARTIFACTS:` heading. + - Acceptance: every artifact named by P0-T1 through P0-T14 exists at its exact path; every command-bearing artifact among them carries `Timestamp:`, `Command:`, `EXIT_CODE:` and `Output Summary:`; and every artifact whose recorded `EXIT_CODE:` is non-zero carries `ExpectedExitCode:` with that same value. +- [x] [P0-T16] Commit the feature folder — `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md`, the research record, this plan and every Phase 0 artifact — and record FEATURE/evidence/baseline/phase0-commit.md. + - Command: `git add -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942` then `git commit -m "docs(942): plan and Phase 0 baseline evidence for the prime-fault ordering fix" -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942` then `git status --porcelain -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942`. + - Acceptance: the commit exits 0; `PHASE0-COMMIT-SHA:` records `git rev-parse HEAD` as an observation; the porcelain span scoped to the feature folder lists no entry other than this plan file and the artifact this task itself writes (both are written after the commit); no path outside the feature folder is staged by this task. The pathspec form keeps the commit within the exempt tree. + +### Phase 1 — Regression Test First (fails against the unchanged production file) + +- [x] [P1-T1] Add the `OnLogError` hook to the `Harness` in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`: replace line 415 with the five-line lambda block (the file grows by four lines, so the `OnInvalidate` property line moves from 434 to 438), then insert one blank line and then the six hook lines immediately after the `OnInvalidate` property line (line 434 of the unedited file, line 438 once the lambda edit is in place), exactly as the Delivered Source section states. Run `CMD-TOKEN-COUNT` with `FILE` `TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs` and `TOKEN` `"internal Action OnLogError { get; set; }", "OnLogError?.Invoke(message, exception);", "Errors.Add(new LoggedError(message, exception));", "invoked from inside the error-log sink", "[TestMethod]"` and record it in FEATURE/evidence/qa-gates/harness-hook-edit-scope.md (this task creates the file; P2-T6 appends to it). + - Acceptance: the first four tokens each count exactly 1 (each was 0 at the anchor); `FIRST-LINE` of `OnLogError?.Invoke(message, exception);` equals `FIRST-LINE` of `Errors.Add(new LoggedError(message, exception));` plus 1 (the invoke immediately follows the append); the `[TestMethod]` count equals its count at the anchor, which is 15 (fact 2: sixteen test methods, of which one is a `[DataTestMethod]`, and the substring `[TestMethod]` does not occur inside `[DataTestMethod]`), so no test method was added or removed. No other file is modified by this task. +- [x] [P1-T2] Create `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` with the whole text given in the Delivered Source section, then run `CMD-TOKEN-COUNT` with `FILE` `TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` and `TOKEN` `"public async Task GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged()", "[TestMethod]", "[TestClass]", "public partial class EngineToggleStateCoordinatorTests", "handleSeenBySink", "harness.OnLogError =", "var prime = harness.Coordinator.GetPrimeTask(SpamEngine);", "probe.SetException(failure);", "await prime;", "has lost isolation", "Regression for issue #942", ".ContainSingle(", ".BeEmpty(", ".BeSameAs(", "using Moq;", "// Arrange", "// Act", "// Assert", ".Contain(SpamEngine", ".BeSameAs(failure", "var harness = new Harness();", "Invariant: for a key whose prime did not run to completion", "handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);", "Task.CompletedTask,", "harness.Invalidations.Should().BeEmpty(", "the message names the engine whose prime failed", "the sink receives the injected exception unchanged", "a prime fault is reported exactly once", "a failed prime leaves nothing to display", "must still be registered", "so a later read may re-prime"` and record the counts in FEATURE/evidence/regression-testing/build-before-reorder.md (this task creates the file; P1-T4 appends to it; P3-T2 re-runs this exact token list on the formatted file). The last nine tokens were appended in preflight round 2; each is a single physical line of the Delivered Source at its in-file indentation within the 100-column formatter width (the two reason strings that exceeded it were re-split at a word boundary so that `must still be registered` and `so a later read may re-prime` each sit whole on one concatenation operand, and the `BeEmpty` reason was shortened to `a failed prime leaves nothing to display` so that its call fits on one line), so the count of each is exactly 1 before and after the format pass. + - Acceptance: the method line, `[TestMethod]`, `public partial class EngineToggleStateCoordinatorTests`, `harness.OnLogError =`, `var prime = harness.Coordinator.GetPrimeTask(SpamEngine);`, `probe.SetException(failure);`, `await prime;`, `has lost isolation`, `// Arrange`, `// Act`, `// Assert`, `.Contain(SpamEngine`, `.BeSameAs(failure`, `var harness = new Harness();` and `Invariant: for a key whose prime did not run to completion` each count exactly 1; `.BeSameAs(` counts exactly 3 (the sink-handle assertion, the injected-exception assertion and the cleared-marker assertion); `[TestClass]` and `using Moq;` count 0; `handleSeenBySink` at least 3; `Regression for issue #942` at least 1; every one of the nine tokens appended in preflight round 2 (`handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);`, `Task.CompletedTask,`, `harness.Invalidations.Should().BeEmpty(`, `the message names the engine whose prime failed`, `the sink receives the injected exception unchanged`, `a prime fault is reported exactly once`, `a failed prime leaves nothing to display`, `must still be registered`, `so a later read may re-prime`) counts exactly 1; `FIRST-LINE` of `the message names the engine whose prime failed` equals `FIRST-LINE` of `.Contain(SpamEngine`; `FIRST-LINE` of `the sink receives the injected exception unchanged` equals `FIRST-LINE` of `.BeSameAs(failure`; `FIRST-LINE` of `a prime fault is reported exactly once` equals `FIRST-LINE` of `.ContainSingle(`; `FIRST-LINE` of `a failed prime leaves nothing to display` equals `FIRST-LINE` of `.BeEmpty(` (each reason string sits on the same physical line as the call it belongs to); `.ContainSingle(` and `.BeEmpty(` at least 1 each; and `FIRST-LINE` of `var prime = harness.Coordinator.GetPrimeTask(SpamEngine);` and of `harness.OnLogError =` are both less than `FIRST-LINE` of `probe.SetException(failure);` (the handle is captured and the probe installed before the trigger). +- [x] [P1-T3] Register the new partial in `TaskMaster.Test/TaskMaster.Test.csproj` by inserting `` immediately after the Race entry at line 359, and record FEATURE/evidence/qa-gates/csproj-registration.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $p = Get-Content -LiteralPath "TaskMaster.Test\TaskMaster.Test.csproj" -Encoding UTF8; $race = 0; $new = 0; for ($i = 0; $i -lt $p.Count; $i++) { if ($p[$i].Contains("EngineToggleStateCoordinatorTests.Race.cs")) { $race = $i + 1 }; if ($p[$i].Contains("EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs")) { $new = $i + 1 } }; "RACE_LINE=$race NEW_LINE=$new NEW_COUNT=$(@($p | Where-Object { $_.Contains("EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs") }).Count)"; git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/TaskMaster.Test.csproj'` + - Acceptance: `NEW_COUNT=1`, `NEW_LINE` equals `RACE_LINE` plus 1, and the anchored numstat line for the project file reports 1 insertion and 0 deletions. The project file is outside the formatter (fact 10), so no format pass follows this edit. +- [x] [P1-T4] Build with `CMD-BUILD` (`TASKID` p1-t4) so the test assembly carries the hook and the new test, appending the result to FEATURE/evidence/regression-testing/build-before-reorder.md. + - Acceptance: `MSBUILD_EXIT_CODE: 0`, `ERRORS: 0`, `TEST_DLL_ADVANCED: True`, `CSC_OUT_TASKMASTER_TEST:` at least 1. A green build here is what makes the next task's failure a genuine assertion failure rather than a compile error; the production file is unchanged from the anchor at this point (P2-T6 proves it retrospectively through `BASE-HASH-PROD:`). +- [x] [P1-T5] [expect-fail] Run the coordinator fixture against the unchanged production file with `CMD-VSTEST` (`ASSEMBLY-TM`, `FILTER-COORD`, `TASKID` p1-t5, `NAMES-COORD`) and record FEATURE/evidence/regression-testing/prime-fault-ordering-fail-before.md (fixed name per the spec). + - Artifact: `Timestamp:`, `Command:`, `EXIT_CODE:` (non-zero), `ExpectedExitCode:` equal to the observed value, an `Output Summary:` with the `COUNTERS` line, every `RESULT`, `FAILED` and `MESSAGE` line, plus an `Environment of the control:` paragraph stating that the production file is byte-identical to the merge base `231e1c0b55105aeb626bf5a6e8d0266a567cacad` (its `CMD-HASH` value equals `BASE-HASH-PROD:` from P0-T11, recorded here as `PROD-HASH-AT-CONTROL:`), that the hook, the new partial and its csproj entry are present, and that the run settings are unchanged (their anchored diff, `git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.runsettings scripts/vscode/TaskMaster.cli.runsettings`, exits 0). + - Acceptance, all required: `TRX_PRESENT: True`; `SEQUENCE_FILES: 0`; `EXIT_CODE:` non-zero; `RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Failed`; the transcribed `MESSAGE` text for that test contains both `to refer to` (the FluentAssertions `BeSameAs` failure fragment, fact 6) and `must still be registered` (a fragment of the first assertion's reason string, which appears in no other assertion of the fixture, so the two fragments together identify the sink-handle assertion as the failing one); the `COUNTERS` total equals `BASELINE-TOTAL:` plus 1; every `FAILED` name is either the new test or `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` (the racy original may fail here, which is consistent with the defect); `PROD-HASH-AT-CONTROL:` equals `BASE-HASH-PROD:`. If the new test is reported `Passed`, the negative control has lost isolation: stop and report `FAIL-BEFORE NOT REPRODUCED`; do not proceed to Phase 2. + +### Phase 2 — Minimal Production Fix and Green Flip + +- [x] [P2-T1] Reorder `CompletePrime` in `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` so that `_logError(BuildPrimeFailedMessage(engineName), failure);` precedes `_primeTasks.TryRemove(engineName, out _);`, inserting the three-line why-comment above the `_logError` statement exactly as the Delivered Source section states. No other line of the method changes. + - Acceptance (verified by P2-T6): within the `CompletePrime` span the `_logError(` line index is less than the `_primeTasks.TryRemove(engineName, out _);` line index; the span measurement prints `SPAN_RETURN=1`, `SPAN_RANTOCOMPLETION=1`, `SPAN_TRY=0`, `SPAN_CATCH=0` and `SPAN_LOCK=0`; the file-level tokens `GetBaseException()` and `new TaskCanceledException(completed)` each count exactly 1; and the line carrying `Report-then-clear is load-bearing` is exactly 3 lines above the `_logError(` line. +- [x] [P2-T2] Replace the `summary` element on `CompletePrime` and extend the `returns` element on `GetPrimeTask` in `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` with the texts given in the Delivered Source section, each documented token on one physical line. + - Acceptance (verified by P2-T6): `and only then is the in-flight marker cleared` counts exactly 1 and its line lies within the twelve lines above the `private void CompletePrime(` line; `cleared only after that report has returned` counts exactly 1 and its line lies within the eight lines above the `internal Task GetPrimeTask(` line. The `remarks` element on `CompletePrime` is unchanged. +- [x] [P2-T3] Build with `CMD-BUILD` (`TASKID` p2-t3) and record FEATURE/evidence/regression-testing/build-after-reorder.md. + - Acceptance: `MSBUILD_EXIT_CODE: 0`, `ERRORS: 0`, `TEST_DLL_ADVANCED: True`, `CSC_OUT_TASKMASTER_TEST:` at least 1. +- [x] [P2-T4] Re-run the original reproduction and the regression test together: run `CMD-VSTEST` (`ASSEMBLY-TM`, `FILTER-COORD`, `TASKID` p2-t4, `NAMES-COORD`) and record FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md (fixed name per the spec). + - Artifact: `Timestamp:`, `Command:` (identical to P1-T5's except the task id segments), `EXIT_CODE: 0`, an `Output Summary:` with the `COUNTERS` line, every `RESULT` line and the sentence that the only production difference between this run and P1-T5 is the statement reorder and documentation in `CompletePrime` and `GetPrimeTask` of `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (the hook, the partial and the csproj entry were present in both runs), with `PROD-HASH-AFTER:` from `CMD-HASH` differing from `BASE-HASH-PROD:`. + - Acceptance, all required: `EXIT_CODE: 0`; `TRX_PRESENT: True`; `SEQUENCE_FILES: 0`; the `COUNTERS` line has `failed` 0 and total equal to `BASELINE-TOTAL:` plus 1; `RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Passed`; `RESULT GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse = Passed`; `RESULT GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker = Passed`; `RESULT GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked = Passed`; no `FAILED` line. +- [x] [P2-T5] Compare the pass-after population against the baseline population by reading FEATURE/evidence/baseline/coordinator-tests-baseline.md and FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md, appending a `POPULATION-COMPARISON:` paragraph to the latter. + - Acceptance: the pass-after total equals `BASELINE-TOTAL:` plus 1; the pass-after `failed` is 0; every name in `BASELINE-FAILED:` (when not `NONE`) appears as `Passed` in the pass-after `RESULT` lines or is recorded by name as still failing (in which case the run stops with `PASS-AFTER NOT GREEN`). +- [x] [P2-T6] Verify the production edit scope and the documentation tokens of `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` and record FEATURE/evidence/qa-gates/production-reorder-scope.md; append the harness-file hunk check to FEATURE/evidence/qa-gates/harness-hook-edit-scope.md. + - Command, production: `CMD-TOKEN-COUNT` with `FILE` `TaskMaster\Ribbon\EngineToggleStateCoordinator.cs` and `TOKEN` `"private void CompletePrime(", "_logError(BuildPrimeFailedMessage(engineName), failure);", "_primeTasks.TryRemove(engineName, out _);", "Report-then-clear is load-bearing", "and only then is the in-flight marker cleared", "cleared only after that report has returned", "internal Task GetPrimeTask(", "catch (", "lock (", "new TaskCanceledException(completed)", "GetBaseException()"`; then `git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` and `git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`. + - Command, span: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $src = Get-Content -LiteralPath "TaskMaster\Ribbon\EngineToggleStateCoordinator.cs" -Encoding UTF8; $s = -1; for ($i = 0; $i -lt $src.Count; $i++) { if ($src[$i].Contains("private void CompletePrime(")) { $s = $i; break } }; $e = -1; for ($i = $s + 1; $i -lt $src.Count; $i++) { if ($src[$i].TrimEnd() -eq " }") { $e = $i; break } }; $span = @($src[$s..$e]); "SPAN=$($s + 1)-$($e + 1)"; "SPAN_RETURN=$(@($span | Where-Object { $_.Trim() -eq "return;" }).Count)"; "SPAN_RANTOCOMPLETION=$(@($span | Where-Object { $_.Contains("completed.Status == TaskStatus.RanToCompletion") }).Count)"; "SPAN_TRY=$(@($span | Where-Object { $_ -cmatch "\btry\b" }).Count)"; "SPAN_CATCH=$(@($span | Where-Object { $_ -cmatch "\bcatch\b" }).Count)"; "SPAN_LOCK=$(@($span | Where-Object { $_.Contains("lock (") }).Count)"'` (the span runs from the `CompletePrime` signature to the first following line that is exactly eight spaces and a closing brace, the method's own closing brace; the `if` block's brace sits at twelve spaces and does not end it). + - Command, harness: `git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` and `git diff --numstat 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, together with `CMD-TOKEN-COUNT` on that file with `TOKEN` `"private sealed class Harness", "private sealed class LoggedError", "new Mock(MockBehavior.Strict)"`. + - Hunk-window rule (production): let G be `FIRST-LINE` of `internal Task GetPrimeTask(`, S be `FIRST-LINE` of `and only then is the in-flight marker cleared`, and R be `FIRST-LINE` of `_primeTasks.TryRemove(engineName, out _);`, all read from the `CMD-TOKEN-COUNT` output of this task (each token occurs exactly once, so each `FIRST-LINE` is the only line). For each hunk header `@@ -a,b +c,d @@` of the anchored diff (an omitted count in a hunk header is 1) the new-side span is c through c+d-1, or the single line c when d is 0; every span must lie wholly within [G-6, G-1] (the `returns` element of `GetPrimeTask`, whose changed lines are the last content line and the two inserted lines) or wholly within [S-2, R] (the `CompletePrime` summary and body: the summary's two changed lines start no earlier than S-1, the pure deletion of the moved statement and its blank line reports a new-side position between the `return` block and the comment, and the last insertion ends at R). A span outside both windows is `HUNK OUTSIDE EDIT WINDOWS`: the artifact records the header and the run stops. The rule replaces any single lower bound: the constructor, `GetPressed`, `HandleToggleClickAsync`, `ExecuteToggleAsync`, `StartPrimeIfNeeded`, `StartObservedPrime`, `ApplyPrimeAsync` and the `remarks` element all lie outside both windows. + - Acceptance, production, all required: `FIRST-LINE` of `_logError(BuildPrimeFailedMessage(engineName), failure);` is less than `FIRST-LINE` of `_primeTasks.TryRemove(engineName, out _);` and both are greater than `FIRST-LINE` of `private void CompletePrime(`; `FIRST-LINE` of `Report-then-clear is load-bearing` equals `FIRST-LINE` of the `_logError(` token minus 3; `catch (` counts exactly 1 and `lock (` counts exactly 1 (both unchanged from the anchor, fact 1); the three documentation tokens count exactly 1 each and satisfy the position clauses of P2-T2; `new TaskCanceledException(completed)` and `GetBaseException()` count exactly 1 each; the span measurement prints `SPAN_RETURN=1`, `SPAN_RANTOCOMPLETION=1`, `SPAN_TRY=0`, `SPAN_CATCH=0` and `SPAN_LOCK=0`, and its `SPAN=` start equals `FIRST-LINE` of `private void CompletePrime(`; every hunk satisfies the hunk-window rule, with G, S, R and every header transcribed; the numstat deletion count is at most 8, and the artifact quotes every removed line to show each is one of: a line of the `CompletePrime` summary element, a line of the `GetPrimeTask` returns element, the `TryRemove` statement, or a blank line adjacent to that statement. + - Acceptance, harness, all required: every `@@` hunk of the anchored diff has a new-side start line greater than `FIRST-LINE` of `private sealed class Harness` and less than `FIRST-LINE` of `private sealed class LoggedError`; the diff adds and removes zero lines containing `[TestMethod]`; `new Mock(MockBehavior.Strict)` counts exactly 1 (fact 2: the strict engines mock the new test consumes through `harness.Engines`); the numstat deletion count is exactly 1 (the replaced single-line lambda). +- [x] [P2-T7] Verify that the original reproduction test is byte-for-byte unchanged and record FEATURE/evidence/qa-gates/original-test-unchanged.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; function Get-MethodText([string[]]$lines) { $s = 0; for ($i = 0; $i -lt $lines.Count; $i++) { if ($lines[$i].Contains("GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse()")) { $s = $i; break } }; $e = 0; for ($i = $s; $i -lt $lines.Count; $i++) { if ($lines[$i].TrimEnd([char]13) -eq " }") { $e = $i; break } }; return ((@($lines[($s - 1)..$e]) | ForEach-Object { $_.TrimEnd([char]13) }) -join ([string][char]10)) }; $base = @(git show 231e1c0b55105aeb626bf5a6e8d0266a567cacad:TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs); $now = Get-Content -LiteralPath "TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs" -Encoding UTF8; $sha = [System.Security.Cryptography.SHA256]::Create(); $h1 = [BitConverter]::ToString($sha.ComputeHash([Text.Encoding]::UTF8.GetBytes((Get-MethodText $base)))); $h2 = [BitConverter]::ToString($sha.ComputeHash([Text.Encoding]::UTF8.GetBytes((Get-MethodText $now)))); "BASE_METHOD_SHA=$h1"; "NOW_METHOD_SHA=$h2"; "METHOD_UNCHANGED=$($h1 -eq $h2)"; "METHOD_LINES=$(((Get-MethodText $now) -split ([string][char]10)).Count)"'` + - Acceptance: `METHOD_UNCHANGED=True` and `METHOD_LINES=32` (the `[TestMethod]` line 212 through the closing brace at 243, fact 2). The extraction starts one line above the method signature (its attribute) and ends at the first eight-space closing brace, so the whole method body is compared; only a trailing carriage return is trimmed on each side (`TrimEnd([char]13)`), so a CR difference between the blob and the checkout cannot produce a false mismatch while any other change to the method's text, including trailing whitespace, still does. The line separator is built as `[string][char]10` rather than an escape sequence so the payload carries no backtick. +- [x] [P2-T8] Format the three source files with CSharpier, then commit the implementation — `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`, `TaskMaster.Test/TaskMaster.Test.csproj` and the feature folder — and record FEATURE/evidence/qa-gates/implementation-commit.md. + - Command, format (before any `git add`): `CMD-HASH` before; then `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool run csharpier format TaskMaster\Ribbon\EngineToggleStateCoordinator.cs TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs; "CSHARPIER_EXIT_CODE: $LASTEXITCODE"'`; then `CMD-HASH` after. The artifact records the six hashes and `PRECOMMIT-FORMAT-REWRITES:` as the number of the three paths whose two hashes differ. When that number is non-zero, the P2-T6 production, span and harness commands and the P2-T7 comparison, the P1-T1 `CMD-TOKEN-COUNT` with its exact `TOKEN` list, and the P1-T2 `CMD-TOKEN-COUNT` with its exact `TOKEN` list are re-run on the formatted text before staging and recorded under `PRECOMMIT-FORMAT-RECHECK:` in this artifact; every clause of P1-T1, P1-T2, P2-T6 and P2-T7 must hold there. A failing recheck clause is repaired by editing the affected file (a Write Set edit, still before the commit), re-running the format command and the recheck, and only then staging; the artifact records each repair. + - Command, commit: `git add -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs TaskMaster.Test/TaskMaster.Test.csproj docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942` then `git commit -m "fix(ribbon): report a prime fault before clearing its in-flight marker (issue 942)"` then `git show --name-only --format= HEAD` then `git status --porcelain -- TaskMaster TaskMaster.Test`. + - Acceptance: `CSHARPIER_EXIT_CODE: 0`; `PRECOMMIT-FORMAT-REWRITES:` is recorded (0 is expected when the Delivered Source layout is already formatter-stable; a non-zero value is admissible only with a passing `PRECOMMIT-FORMAT-RECHECK:`); the commit exits 0; `IMPLEMENTATION-COMMIT-SHA:` records `git rev-parse HEAD` as an observation; the `git show` name list contains exactly the four code paths plus paths under the feature folder and nothing else; the porcelain span scoped to the two code trees prints no line. The format precedes the staging so that the committed text is the formatter's own output and the Phase 3 repository-wide format pass can rewrite nothing; from this commit on, no code file is edited. This commit stages paths outside every exempt tree, so it runs only in a session whose pre-implementation checkpoint is ready; a hook refusal is `PRE-IMPLEMENTATION GATE BLOCKED`. + +### Phase 3 — Final QA Toolchain Loop, Coverage Delta, Footprint and Acceptance + +The loop is format, then the read-only format check, then the analyzer rebuild, then the nullable rebuild, then the coverage-enabled test run, in the CLAUDE.md order. The code was committed at P2-T8 after a scoped format, so no step of this loop may rewrite a code file and no code file is edited after P2-T8: a failing or rewriting step is stop and report (Execution conventions, restart rule), except the single P3-T8 re-run the plan admits. P3-T9 records that a single pass completed clean. + +- [x] [P3-T1] Run the formatter repository-wide with `dotnet tool run csharpier format .` and record FEATURE/evidence/qa-gates/csharpier-format.md with a before-and-after observation. + - Command: `CMD-HASH` before; `git status --porcelain -- . ":(exclude)docs/features" ":(exclude).claude"` before; `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool run csharpier format .; "CSHARPIER_EXIT_CODE: $LASTEXITCODE"'`; `CMD-HASH` after; the same scoped porcelain span after. + - Acceptance: `CSHARPIER_EXIT_CODE: 0` recorded as `EXIT_CODE:`; the artifact records six hashes (three before, three after) and defines the rewritten-file count as the number of Write Set paths whose two hashes differ; it records both scoped porcelain outputs verbatim and requires them to be identical line sets (the repository-wide format rewrote no file outside the Write Set; a difference is `FORMAT WIDENED FOOTPRINT`: stop and report, because P0-T8 established a clean drift baseline). The console line `Formatted N files` is not used as the rewritten count: CSharpier reports files processed, not files changed. The rewritten count must be 0; a non-zero count is POST-COMMIT CODE REWRITE: stop and report. +- [x] [P3-T2] Re-run the scope and token gates on the formatted files: repeat the P2-T6 production, span and harness checks (including the hunk-window rule with G, S and R re-read from this run's `CMD-TOKEN-COUNT` output), the P2-T7 method comparison, the P1-T1 token count on the primary fixture with P1-T1's exact `TOKEN` list (appended to the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/harness-hook-edit-scope.md) and the P1-T2 token count on the new partial with P1-T2's exact `TOKEN` list, appending `POST-FORMAT:` sections to FEATURE/evidence/qa-gates/production-reorder-scope.md, FEATURE/evidence/qa-gates/harness-hook-edit-scope.md, FEATURE/evidence/qa-gates/original-test-unchanged.md and FEATURE/evidence/regression-testing/build-before-reorder.md. + - Acceptance: every clause of P1-T1, P2-T6, P2-T7 and P1-T2 holds on the post-format tree, with the numstat deletion counts re-recorded, every removed production line re-quoted and classified as P2-T6 requires, and the `SPAN_` rows re-printed (the `Report-then-clear` comment's three-line offset and the `_logError` before `TryRemove` order are format-stable; a failing clause is POST-COMMIT CODE REWRITE: stop and report; no code edit is made after P2-T8). The `POST-FORMAT:` section of build-before-reorder.md is the section P3-T18 and P3-T19 cite. +- [x] [P3-T3] Audit the post-format line counts of the three source files with `CMD-LINECOUNT` and record FEATURE/evidence/qa-gates/file-line-counts.md. + - Acceptance: `LINES` for `TaskMaster\Ribbon\EngineToggleStateCoordinator.cs`, `TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.cs` and `TaskMaster.Test\Ribbon\EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` are each at most 500 (expected roughly 420, 470 and 75). This is the authoritative AC14 audit; P0-T11 was advisory. +- [x] [P3-T4] Verify formatting repository-wide, read-only, with `dotnet tool run csharpier check .` and record FEATURE/evidence/qa-gates/csharpier-check-final.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; dotnet tool run csharpier check .; "CSHARPIER_EXIT_CODE: $LASTEXITCODE"'` + - Acceptance: `CSHARPIER_EXIT_CODE: 0` recorded as `EXIT_CODE:` and the output reports no unformatted file. A non-zero exit is a failing step: stop and report. +- [x] [P3-T5] Run the analyzer gate with `CMD-REBUILD` (`GATEARGS` analyzers, `TASKID` p3-t5) and record FEATURE/evidence/qa-gates/msbuild-analyzer-final.md (`Command:` records `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true`). + - Acceptance, all required: `EXIT_CODE: 0`; `ERRORS: 0`; `SKIP_CORECOMPILE_LINES: 0` (no project reported a skipped CoreCompile target); `CSC_OUT_TASKMASTER:` and `CSC_OUT_TASKMASTER_TEST:` each at least 1; `WRITESET_DIAGNOSTIC_LINES: 0`; `WARNINGS:` at most `ANALYZER-BASELINE-WARNINGS:` from P0-T9. +- [x] [P3-T6] Run the nullable type-check gate with `CMD-REBUILD` (`GATEARGS` nullable, `TASKID` p3-t6) and record FEATURE/evidence/qa-gates/msbuild-nullable-final.md (`Command:` records `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true`; no Nullable property override). + - Acceptance, all required: `EXIT_CODE: 0`; `ERRORS: 0`; `SKIP_CORECOMPILE_LINES: 0`; both `CSC_OUT_` counts at least 1; `WRITESET_DIAGNOSTIC_LINES: 0`; `WARNINGS:` at most `NULLABLE-BASELINE-WARNINGS:` from P0-T10; `TEST_DLL_EXISTS: True`. +- [x] [P3-T7] Re-run the coordinator fixture on the rebuilt assembly with `CMD-VSTEST` (`ASSEMBLY-TM`, `FILTER-COORD`, `TASKID` p3-t7, `NAMES-COORD`) and append a `FINAL-FIXTURE-RUN:` section to FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md. + - Acceptance: `EXIT_CODE: 0`; `SEQUENCE_FILES: 0`; `COUNTERS` `failed` 0 and total equal to `BASELINE-TOTAL:` plus 1; all four `NAMES-COORD` names `Passed`. This confirms the fixture on the exact assembly the coverage run measures. +- [x] [P3-T8] Run the coverage-enabled test gate by the route P0-T13 fixed and record FEATURE/evidence/qa-gates/coverage-post-change.md (fixed name per the spec): under `RUNNER` run `CMD-COVERAGE-RUNNER` with `STAGE` final, under `DIRECT` run `CMD-COVERAGE-DIRECT` with `STAGE` final, then `CMD-COVERAGE-POST` with `STAGE` final and `RAW` per the rule, with the same artifact fields as P0-T14. + - Re-run rule: when the first attempt's `FAILED-SET:` is exactly the single name `TryAddValuesAsync_UpdatesExistingValue` (the issue 780 sporadic failure P0-T14 admits), the same command is run once more with the first attempt recorded as `FIRST-ATTEMPT-FAILED-SET:` and the second attempt's values recorded as the run; no other failure and no second re-run is admitted. + - Acceptance, all required: branch (a) of P0-T14's branch rule (exit 0, both floors met, `FAILED-SET:` empty) on the recorded attempt; `SEQUENCE_FILES: 0` (DIRECT) and `TRX_PRESENT: True`; the summary block's second line reports `failed 0`; `COORD-CLASS-NODES: 1`; `COMPLETEPRIME-UNCOVERED: 0`; the projection block contains the `TaskMaster` package with both counters; `RESULT`-level proof that the new test executed is taken from P3-T7, because the runner's summary carries counts and failed names only. Any other outcome is a failing step: stop and report (a stall is `COVERAGE RUN STALLED` or `ABORTED`: stop and report; no code edit is made after P2-T8). Under `RUNNER` the runner's own 80 percent line and 75 percent branch assertions are the floor gate; under `DIRECT` the `LINE-FLOOR:` and `BRANCH-FLOOR:` lines are. +- [x] [P3-T9] Record the loop closure in FEATURE/evidence/qa-gates/toolchain-final-pass.md (fixed name per the spec). + - Acceptance: the artifact lists P3-T1 through P3-T8 with each step's `Command:` and exit code, states the pass number, states for P3-T5 and P3-T6 that `SKIP_CORECOMPILE_LINES: 0` and both `CSC_OUT_` counts are at least 1 (the analyzer and nullable gates compiled rather than short-circuited), states for P3-T4 that the check reported no differences, and states for P3-T8 the route and that the run exited 0. If any pass failed, both the failed pass and the clean pass are recorded, and the clean pass is the one whose steps ran in order with no intervening file rewrite. +- [x] [P3-T10] Compute the coverage comparison from FEATURE/evidence/baseline/coverage-baseline.md and FEATURE/evidence/qa-gates/coverage-post-change.md and append a `COMPARISON:` section to FEATURE/evidence/qa-gates/coverage-post-change.md. + - Rows, all required: `COORD-LINES-BASELINE:` and `COORD-LINES-FINAL:` (covered over valid, both stages); `COORD-BRANCHES-BASELINE:` and `COORD-BRANCHES-FINAL:`; `COMPLETEPRIME-ELEMENTS-BASELINE:` and `-FINAL:`; `COMPLETEPRIME-UNCOVERED-FINAL:`; `FIRST-PARTY-BASELINE:` and `FIRST-PARTY-FINAL:` (the two `First-party coverage:` lines verbatim); `ROOT-BASELINE:` and `ROOT-FINAL:` (the two `ROOT` lines); `DENOMINATOR-BRANCH:` either `COMPARABLE` (the two root lines-valid figures differ by at most 1 percent of the baseline figure; then the final root line-rate must be at least the baseline root line-rate minus 0.005) or `INCOMPARABLE` (recorded, not gated, with the one-sentence reason from D-7); `CHANGED-LINES:` the added line numbers of `git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` that fall inside `COMPLETEPRIME-SPAN:` (final), each with its `hits` value from the final `COMPLETEPRIME-LINE` rows or `no line element` for a comment or brace line. + - Acceptance, all required: `COORD-LINES-FINAL` valid equals `COORD-LINES-BASELINE` valid (no executable statement added; a mismatch is `COORD-LINES-VALID CHANGED`: record both and treat AC12 as NOT MET); `COORD-LINES-FINAL` covered at least `COORD-LINES-BASELINE` covered; `COORD-BRANCHES-FINAL` covered at least baseline covered; `COMPLETEPRIME-ELEMENTS-FINAL` equals `COMPLETEPRIME-ELEMENTS-BASELINE`; `COMPLETEPRIME-UNCOVERED-FINAL: 0`; every `CHANGED-LINES` entry that has a line element has hits at least 1; exactly one `DENOMINATOR-BRANCH:` value is recorded and, under `COMPARABLE`, its rate clause holds. The anchored diff against the working tree is used so its line numbers align with the coverage document generated from the same tree. +- [x] [P3-T11] Verify the determinism-token constraints of AC10 over the anchored diff of the test directory and record FEATURE/evidence/qa-gates/determinism-tokens.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $added = @(git diff -U0 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon | Where-Object { $_.StartsWith("+") -and -not $_.StartsWith("+++") }); "ADDED-LINE-COUNT: $($added.Count)"; foreach ($t in @("Thread.Sleep", "Task.Delay", "DoNotParallelize", "[Timeout", "Timeout=", "DateTime.Now", "DateTime.UtcNow", "Stopwatch", ".Wait(", ".Result", "ManualResetEvent", "SemaphoreSlim", "GetTempFileName", "GetTempPath", "TaskScheduler", "while (", "for (", "Retry", "GetResult(", "DateTimeOffset")) { "ADDED-TOKEN [$t] = $(@($added | Where-Object { $_.Contains($t) }).Count)" }; git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.runsettings scripts/vscode/TaskMaster.cli.runsettings | Out-Null; "RUNSETTINGS_DIFF_EXIT=$LASTEXITCODE"; git diff --exit-code 231e1c0b55105aeb626bf5a6e8d0266a567cacad -- TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs TaskMaster/Ribbon/RibbonController.EngineCommands.cs | Out-Null; "NONGOAL_FILES_DIFF_EXIT=$LASTEXITCODE"'` + - Acceptance: `ADDED-LINE-COUNT:` at least 60 (the new partial plus the hook; a smaller figure means the diff missed the new file); every `ADDED-TOKEN` count is 0, including the three added in preflight round 1 (`GetResult(`, a blocking awaiter read; `GetTempPath`, a temporary-file seam; `DateTimeOffset`, a wall-clock type); `RUNSETTINGS_DIFF_EXIT=0`; `NONGOAL_FILES_DIFF_EXIT=0`. The token list is applied to added lines of the test directory only, so this plan's own prose cannot trip it. +- [x] [P3-T12] Verify that no raw test-result or coverage document entered the repository and record it in FEATURE/evidence/qa-gates/footprint-scope.md (this task creates the file; P3-T14 appends to it). + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $ext = @(".trx", ".xml", ".coverage", ".coveragexml", ".cobertura"); $added = @(git diff --name-only --diff-filter=A 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD); $added | ForEach-Object { "ADDED-PATH: $_" }; "RAW-DOCS-COMMITTED: $(@($added | Where-Object { $ext -contains [IO.Path]::GetExtension($_).ToLowerInvariant() }).Count)"; $untracked = @(git status --porcelain --untracked-files=all --ignored -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 | ForEach-Object { $_.Substring(3) }); "RAW-DOCS-UNTRACKED-IN-FEATURE: $(@($untracked | Where-Object { $ext -contains [IO.Path]::GetExtension($_).ToLowerInvariant() }).Count)"'` (the name-listing diff enumerates committed additions since the anchor; the porcelain span covers untracked and ignored files in the feature folder; --ignored is required because .gitignore lines 146 to 147 ignore trx and cobertura xml names repository-wide). + - Acceptance: `RAW-DOCS-COMMITTED: 0` and `RAW-DOCS-UNTRACKED-IN-FEATURE: 0`; the artifact lists every `ADDED-PATH:` line so the classification is auditable, and that list contains `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (the positive control that the enumeration saw the committed additions). The porcelain companion is required because a name-listing diff cannot see an untracked file. +- [x] [P3-T13] Sweep the feature folder, including this plan, for host identifiers and record FEATURE/evidence/qa-gates/evidence-hygiene.md. + - Command: `pwsh -NoProfile -Command 'Set-Location -LiteralPath "WORKTREE"; $acct = Split-Path -Leaf $env:USERPROFILE; $machine = $env:COMPUTERNAME; $files = @(Get-ChildItem -LiteralPath "docs\features\active\2026-09-29-engine-toggle-prime-fault-logging-test-races-942" -Recurse -File -Filter "*.md"); $a = 0; $m = 0; $d = 0; foreach ($f in $files) { $c = Get-Content -LiteralPath $f.FullName -Raw -Encoding UTF8; $a += ([regex]::Matches($c, [regex]::Escape($acct), "IgnoreCase")).Count; $m += ([regex]::Matches($c, [regex]::Escape($machine), "IgnoreCase")).Count; $n = $c.Replace([string][char]92, "/"); $d += ([regex]::Matches($n, "[a-z]:/+users/+[a-z0-9_.~-]", "IgnoreCase")).Count }; "FILES_SCANNED=$($files.Count) ACCOUNT_HITS=$a MACHINE_HITS=$m DRIVE_USERS_HITS=$d"'` + - Acceptance: `ACCOUNT_HITS=0`, `MACHINE_HITS=0`, `DRIVE_USERS_HITS=0`, `FILES_SCANNED=` at least 36. The two tokens are derived at run time and neither value is written into the artifact. The drive-path check normalises every backslash to a forward slash (`[string][char]92` is the backslash, spelled that way so the payload carries no backslash escape) and then counts matches of the CI hygiene guard's user-profile pattern (scripts/hygiene/Test-RepositoryHygiene.Rules.ps1 line 21) in its forward-slash form, case-insensitively and with a separator run on either side of the profile folder, so every path the CI guard would reject is also counted here. A non-zero count is repaired by replacing the occurrence with the placeholder REDACTED-PATH and re-running this task. +- [x] [P3-T14] Verify the change footprint against the anchor and append it to FEATURE/evidence/qa-gates/footprint-scope.md. + - Command: `git diff --name-status 231e1c0b55105aeb626bf5a6e8d0266a567cacad HEAD` and `git status --porcelain --untracked-files=all`. + - Acceptance, all required: `INHERITED-AND-EXCLUDED:` is either `NONE` or exactly the single path `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` with its status letter (the promotion record the amended AC13 exempts; it is a member of `INHERITED-COMMITTED:` from P0-T4 and D-8 subtracts it by rule); `THIS-ITEM-FOOTPRINT:` lists every remaining path, and every one of them is one of the four code paths or lies under `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/`; the four code paths are all present in the footprint (the new partial with status A); TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs, TaskMaster/Ribbon/RibbonController.EngineCommands.cs, TaskMaster.runsettings and scripts/vscode/TaskMaster.cli.runsettings are absent from the footprint; no porcelain line names a path under TaskMaster/ or TaskMaster.Test/; every other porcelain line is under the feature folder, under .claude/agent-memory/ (uncommitted session memory, never staged), or a member of `PRE-EXISTING-WORKTREE-PATHS:` from P0-T4, and the composition is stated without a count. Any diff path that is neither a code path, nor under the feature folder, nor the promotion record is `FOOTPRINT OUTSIDE AC13`: recorded by path, and AC13 is NOT MET at P3-T27. The porcelain companion is required beside the name-listing diff. +- [x] [P3-T15] Check off AC1 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/production-reorder-scope.md. + - Acceptance: either exactly one checkbox changes from `- [ ] AC1 —` to `- [x] AC1 —` because that section shows `_logError(` before `TryRemove`, `catch (` and `lock (` each 1 at file level, the two exception tokens each 1, and the span rows `SPAN_RETURN=1`, `SPAN_RANTOCOMPLETION=1`, `SPAN_TRY=0`, `SPAN_CATCH=0` and `SPAN_LOCK=0`, and the section's quoted removed lines are each a `CompletePrime` summary line, a `GetPrimeTask` returns line, the `TryRemove` statement or a blank line adjacent to it, so no line of the early return, the failure computation or the synthesized exception was removed or rewritten (the early return and its status test are present inside `CompletePrime` and no `try`, `catch` or lock was added to it); or the box stays unchecked and `AC1: NOT MET` is recorded with the failing values in P3-T29. The criterion text is unmodified. This task completes in either case. +- [x] [P3-T16] Check off AC2 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the three documentation-token rows of the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/production-reorder-scope.md. + - Acceptance: exactly one checkbox flips, or `AC2: NOT MET` is recorded; the cited rows show each token counting 1 at its required position. +- [x] [P3-T17] Check off AC3 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/harness-hook-edit-scope.md. + - Acceptance: exactly one checkbox flips, or `AC3: NOT MET` is recorded; the cited section shows the four hook tokens at 1, the append-then-invoke adjacency, every hunk inside the `Harness` span and zero `[TestMethod]` lines added or removed. +- [x] [P3-T18] Check off AC4 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `POST-FORMAT:` section of FEATURE/evidence/regression-testing/build-before-reorder.md (the token counts on the formatted partial, written by P3-T2) and the `RESULT` lines of FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md. + - Acceptance: exactly one checkbox flips, or `AC4: NOT MET` is recorded; the cited counts show the capture-before-trigger order (`FIRST-LINE` clauses), the sink probe (`harness.OnLogError =` 1), `.BeSameAs(` 3 with `.BeSameAs(failure` 1, `.Contain(SpamEngine` 1, `.ContainSingle(` and `.BeEmpty(` at least 1, `handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);` 1 (the sink records `GetPrimeTask` from inside the hook), `Task.CompletedTask,` 1 and `harness.Invalidations.Should().BeEmpty(` 1 (the post-await cleared-marker assertion and the empty-invalidations assertion), and the test is `Passed` in the pass-after run. +- [x] [P3-T19] Check off AC5 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `POST-FORMAT:` section of FEATURE/evidence/regression-testing/build-before-reorder.md and the `new Mock(MockBehavior.Strict)` row of the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/harness-hook-edit-scope.md. + - Acceptance: exactly one checkbox flips, or `AC5: NOT MET` is recorded; the cited rows show `[TestMethod]` 1 (MSTest attribute), `new Mock(MockBehavior.Strict)` 1 (the strict Moq mock the test consumes through `harness.Engines`), `.Contain(SpamEngine` 1 and `.BeSameAs(failure` 1, `FIRST-LINE` of `the message names the engine whose prime failed` equal to `FIRST-LINE` of `.Contain(SpamEngine`, `FIRST-LINE` of `the sink receives the injected exception unchanged` equal to `FIRST-LINE` of `.BeSameAs(failure`, `FIRST-LINE` of `a prime fault is reported exactly once` equal to `FIRST-LINE` of `.ContainSingle(`, `FIRST-LINE` of `a failed prime leaves nothing to display` equal to `FIRST-LINE` of `.BeEmpty(`, `must still be registered` 1 and `so a later read may re-prime` 1 (every one of the six FluentAssertions calls carries a reason string: the four same-line equalities pin a reason to the `ContainSingle`, `Contain`, `BeSameAs(failure` and `BeEmpty` calls, and the two remaining reason fragments belong to the two `BeSameAs` calls on the prime handle), `// Arrange`, `// Act` and `// Assert` 1 each, `Regression for issue #942` at least 1 with `Invariant: for a key whose prime did not run to completion` 1 (the summary names the issue and the invariant), and `has lost isolation` 1 (the negative-control comment). +- [x] [P3-T20] Check off AC6 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/qa-gates/csproj-registration.md and the `RESULT GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged = Passed` line of FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md. + - Acceptance: exactly one checkbox flips, or `AC6: NOT MET` is recorded. +- [x] [P3-T21] Check off AC7 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/regression-testing/prime-fault-ordering-fail-before.md. + - Acceptance: exactly one checkbox flips, or `AC7: NOT MET` is recorded; the artifact carries `Timestamp:`, `Command:`, a non-zero `EXIT_CODE:`, a matching `ExpectedExitCode:`, the merge-base commit with `PROD-HASH-AT-CONTROL:` equal to `BASE-HASH-PROD:`, and the new test `Failed` with a transcribed message containing both `to refer to` and `must still be registered` (the same-instance assertion on the sink-observed handle is the one that failed). +- [x] [P3-T22] Check off AC8 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md. + - Acceptance: exactly one checkbox flips, or `AC8: NOT MET` is recorded; the artifact shows `EXIT_CODE: 0`, both named tests `Passed`, and the only-production-difference statement. +- [x] [P3-T23] Check off AC9 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `COUNTERS` line and `POPULATION-COMPARISON:` of FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md and the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/original-test-unchanged.md. + - Acceptance: exactly one checkbox flips, or `AC9: NOT MET` is recorded; `failed` is 0 with total `BASELINE-TOTAL:` plus 1 and `METHOD_UNCHANGED=True`. +- [x] [P3-T24] Check off AC10 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/qa-gates/determinism-tokens.md. + - Acceptance: exactly one checkbox flips, or `AC10: NOT MET` is recorded; every `ADDED-TOKEN` count is 0 and `RUNSETTINGS_DIFF_EXIT=0`. +- [x] [P3-T25] Check off AC11 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/qa-gates/toolchain-final-pass.md. + - Acceptance: exactly one checkbox flips, or `AC11: NOT MET` is recorded; the artifact records one clean pass in order with the check reporting no differences, both rebuilds at exit 0 with `SKIP_CORECOMPILE_LINES: 0`, and the coverage run at exit 0. +- [x] [P3-T26] Check off AC12 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing the `COMPARISON:` section of FEATURE/evidence/qa-gates/coverage-post-change.md, FEATURE/evidence/baseline/coverage-baseline.md and FEATURE/evidence/qa-gates/footprint-scope.md. + - Acceptance: exactly one checkbox flips, or `AC12: NOT MET` is recorded; the comparison shows the coordinator file's covered lines and covered branches not lower than baseline with equal lines-valid, `COMPLETEPRIME-UNCOVERED-FINAL: 0`, every changed line with a line element covered, and `RAW-DOCS-COMMITTED: 0` with `RAW-DOCS-UNTRACKED-IN-FEATURE: 0`. +- [x] [P3-T27] Check off AC13 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/qa-gates/footprint-scope.md, the `NONGOAL_FILES_DIFF_EXIT=0` row of FEATURE/evidence/qa-gates/determinism-tokens.md, and the `POST-FORMAT:` section of FEATURE/evidence/qa-gates/production-reorder-scope.md. + - Acceptance: exactly one checkbox flips, or `AC13: NOT MET` is recorded; `THIS-ITEM-FOOTPRINT:` holds only the four code paths and feature-folder paths, the four non-goal files are absent, no `FOOTPRINT OUTSIDE AC13` path is recorded, `INHERITED-AND-EXCLUDED:` is `NONE` or exactly the promotion record `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md`, which the amended AC13 names as its sole exemption, so the artifact's footprint and the criterion text agree without a divergence note; and, from the production-reorder-scope section, every production hunk lies within [G-6, G-1] or [S-2, R], neither of which contains `StartPrimeIfNeeded`, and `lock (` counts exactly 1 (the prime gate lock is unchanged, as the criterion's final clause requires). +- [x] [P3-T28] Check off AC14 in `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, citing FEATURE/evidence/qa-gates/file-line-counts.md. + - Acceptance: exactly one checkbox flips, or `AC14: NOT MET` is recorded; all three counts are at most 500. +- [x] [P3-T29] Write the acceptance-criteria status summary to FEATURE/evidence/other/ac-status-summary.md. + - Required contents: the source file path, `TOTAL: of 14` counted from the `- [x] AC` lines actually present in the spec's acceptance section, `UNMET:` listing every `ACn: NOT MET` recorded by P3-T15 through P3-T28 with its failing values, or `NONE`, and the text of every remaining unchecked criterion. + - Acceptance: the checked count equals the number of `- [x] AC` lines in the spec, counted from the file rather than summed from this plan's claims; the `UNMET:` line is present. +- [x] [P3-T30] Record the reduced-audit handoff in FEATURE/evidence/other/reduced-audit-handoff.md and commit the feature folder — `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md`, this plan and every Phase 1 to Phase 3 artifact. + - Handoff contents: pointers to FEATURE/evidence/qa-gates/toolchain-final-pass.md, FEATURE/evidence/qa-gates/coverage-post-change.md, FEATURE/evidence/qa-gates/footprint-scope.md, FEATURE/evidence/regression-testing/prime-fault-ordering-fail-before.md, FEATURE/evidence/regression-testing/prime-fault-ordering-pass-after.md and FEATURE/evidence/other/ac-status-summary.md; the `COVERAGE-ROUTE:` used; the statement that hazard B (registration racing removal on a synchronous non-success prime, NB-2 of the issue 735 code review) is out of scope and is promoted separately by the coordinator, so no potential entry was written by this run; and the statement that the coverage-route test evidence committed is projections only. + - Command: `git add -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942` then `git commit -m "docs(942): final QA, coverage comparison, footprint and acceptance evidence"` then `git status --porcelain -- docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942 TaskMaster TaskMaster.Test`. + - Acceptance: the handoff artifact is written before the commit and records `PRE-FINAL-COMMIT-HEAD:` (the P2-T8 commit id, from `git rev-parse HEAD` at write time) as an observation; the commit exits 0 and its id is transcribed into the executor's completion message rather than into any artifact (an artifact written after this commit would be left untracked); the porcelain span lists no entry other than this plan file (its own check-off mark for this task lands after the commit and is committed by the orchestrator) and no entry under TaskMaster/ or TaskMaster.Test/. The pathspec form keeps the commit within the exempt tree. + +## Planner Adversarial Self-Review + +SELF-REVIEW: RE-DERIVED THIS PASS + +Revision round 2, 2026-09-30. Every citation the round-2 deltas touched was re-derived directly against this worktree in this pass, with its sibling region re-read: + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` lines 341 to 355 — the `CompletePrime` body as it stands: status test 343, `return;` 345, the `if` block's brace 346, blank 347, `_primeTasks.TryRemove(engineName, out _);` 348, blank 349, `var failure =` 350 to 352, blank 353, `_logError(...)` 354, method brace 355. The R4 removed-line classification (summary line, returns line, the `TryRemove` statement, or a blank line adjacent to it) therefore leaves 343 to 346 and 350 to 354 as lines the delivered text retains verbatim; the delivered body deletes 348 and one adjacent blank and inserts the comment and the moved statement after 354. Sibling region: the `summary` element 327 to 331 (five lines, replaced by the six-line element of the Delivered Source) and the `remarks` element 332 to 340 (unchanged). +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` lines 242 to 247 and 261 to 278 — the `returns` element 242 to 246 with `internal Task GetPrimeTask(` at 247; `StartPrimeIfNeeded` 261 to 278 with `lock (_primeGate)` at 269, which is greater than G-1 (246) and less than S-2 (330 in the delivered text, where S, the summary token, is 332 and R is 359), so the whole member is outside both hunk windows [G-6, G-1] and [S-2, R] as the R5 clause on P3-T27 states. +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` lines 403 to 457 — `private sealed class Harness` at 403, the single-line error-log lambda at 415, `new Mock(MockBehavior.Strict);` alone at 420, `OnInvalidate` at 434, `private sealed class LoggedError` at 446; the R1 numstat command's expected single deletion is line 415 (replaced by the five-line block), and the hook insertion after 434 deletes nothing. +- New-partial delivered text (R2) — the nine appended tokens each occur exactly once in the Delivered Source, each on one physical line: `handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);` (the lambda body line, 16-space indent, 81 columns); `Task.CompletedTask,` (24-space indent; the doc-comment occurrence reads `Task.CompletedTask"` and does not match); `harness.Invalidations.Should().BeEmpty(` with `a failed prime leaves nothing to display` on the same 99-column line (the previous reason made that line 101 columns, which CSharpier would have broken into a member chain and the call token would then have counted 0); `a prime fault is reported exactly once` on the 96-column `ContainSingle` line; `the message names the engine whose prime failed` on the 92-column `.Contain(SpamEngine, ...)` line; `the sink receives the injected exception unchanged` on the 93-column `.BeSameAs(failure, ...)` line; `must still be registered` on the 86-column second operand of the re-split three-operand reason of the sink-handle assertion (the previous single operand was 103 columns); `so a later read may re-prime` on the 60-column second operand of the re-split cleared-marker reason (the previous single operand was 112 columns). Both re-split concatenations exceed 100 columns when joined, so CSharpier keeps them broken one operand per line. Sibling region: the P1-T5 and P3-T21 clauses read `must still be registered` from the failure message, which FluentAssertions renders from the concatenated reason, so the split does not change the transcribed message; the fragment still appears in no other assertion of the fixture. +- Worktree reflog (R6) — `.git/worktrees//logs/HEAD` records the branch cut at `ddbab26a0149bf2ca5d0256e60686ad79e74d90c` and five commits since, every one a `docs(942):` commit (active folder, research, spec, plan, plan revision). The count is stated as an observation and no gate reads it. +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` (A4) — header Last Updated at line 6 and Version at line 8 advanced to `2026-09-30T01-30` and `0.3`; the acceptance section is unchanged at 228 to 241, fourteen unchecked single-line checkboxes, AC13 at 240 as amended in round 1. + +Revision round 1, 2026-09-30. Every citation the round-1 deltas touched was re-derived directly against this worktree in this pass, with its sibling region re-read; the authoring-pass entries below that no delta touched were re-read where a delta's sibling region overlaps them. Round-1 re-derivations: + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` — `_primeTasks.TryRemove(engineName, out _);` occurs once (348) and `internal Task GetPrimeTask(` once (247), so G and R of the hunk-window rule are single-line anchors; the `try` keyword occurs as code only at 177 and `catch (` at 181, both in the click boundary outside the `CompletePrime` span (341 to 355), so `SPAN_TRY=0`, `SPAN_CATCH=0` and `SPAN_LOCK=0` are false-before only in the sense that the D6 payload did not exist before and true-after by construction, while `SPAN_RETURN=1` and `SPAN_RANTOCOMPLETION=1` pin the early return (345) and its status test (343) as retained. Sibling region: the `if` block's closing brace at 346 sits at twelve spaces and the method's at 355 at eight, so the span-end rule stops at the method brace. Window arithmetic on the delivered text: the `returns` hunk's new-side span is 245 to 247 with G at 249 (window 243 to 248); the summary hunk's span is 331 to 333 with S at 332; the moved statement's pure-deletion hunk reports position 350; the comment insertion spans 355 to 357 and the `TryRemove` insertion is R at 359 (window 330 to 359). +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` — `new Mock(MockBehavior.Strict);` alone on line 420 (declaration `internal Mock Engines { get; } =` at 419), the only occurrence in the file; the other three occurrences of `MockBehavior.Strict` under TaskMaster.Test/Ribbon are in RibbonControllerTests.cs (194, 223, 365) and are outside every gate of this plan. `OnInvalidate` property at 434 moves to 438 after the four-line growth of the lambda edit; the hook insertion is stated relative to both numberings. +- New-partial delivered text — `.BeSameAs(` occurs three times (the sink-handle, injected-exception and cleared-marker assertions); `.BeSameAs(failure` and `.Contain(SpamEngine` once each on their own chained lines; `// Arrange`, `// Act`, `// Assert`, `var harness = new Harness();` and `Invariant: for a key whose prime did not run to completion` once each on one physical line; `must still be registered` appears only in the first assertion's reason string. The reason-bearing calls are written in the member-chain layout CSharpier 1.2.6 produces for the neighbouring `Task.CompletedTask` assertion, so the single-line tokens survive the format pass (round 2 re-split two reason strings and shortened one so that every gated line fits the 100-column width; see the round-2 entry above); `GetResult(`, `GetTempPath` and `DateTimeOffset` do not occur in the delivered text. +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` — AC13 at 240 amended in this round to name the promotion record; the section still holds fourteen single-line checkboxes at 228 to 241, all unchecked, and the AC identity table's opening words for AC13 are unchanged. Sibling region: AC9 (236) and the Test Strategy fail-before paragraph (220) are the only other lines naming the merge base and neither restates AC13's file scope, so no sibling sentence needed the same amendment. +- `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` — exists; `Status: Promoted` at line 5, `Issue: #942` at 9. It is the only non-feature-folder path expected in the anchored diff. +- Worktree reflog — the branch was created at `ddbab26a0149bf2ca5d0256e60686ad79e74d90c` and carries only documentation commits (five at round 2); no code path is touched between the anchor and HEAD, which is the state P0-T4 gates. +- P2-T7 payload — after the D3 and D15 edits the payload contains no backtick and uses `TrimEnd([char]13)` at both call sites; `-split ([string][char]10)` is a one-character regular-expression pattern that splits on LF only. +- P3-T13 payload — after the D4 edit the drive-path count runs on a backslash-normalised copy with the pattern `[A-Za-z]:/Users/`; the payload carries no backslash escape. + +Authoring pass, 2026-09-29. Every citation below was read directly against this worktree in that pass; sibling regions were re-read where an edit lands. + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` — length 415 content lines; no nullable directive; `GetPrimeTask` doc 237 to 246 with `returns` 242 to 246, body 247 to 255; `StartPrimeIfNeeded` 261 to 278 with the lock at 269 and the assignment at 276; `StartObservedPrime` 290 to 303 (continuation options 299 to 301); `CompletePrime` doc 327 to 340 with `summary` 327 to 331 and `remarks` 332 to 340; body 341 to 355 with `TryRemove` at 348, `var failure` 350 to 352, `_logError` at 354; exactly one `catch (` at 181 and one `lock (` at 269; `Task.CompletedTask` at 243, 251, 254. Sibling region re-read: the only other `_logError` call is at 183 inside the click boundary and is untouched; the `remarks` element between the replaced `summary` and the body stays as the D-1 fix leaves it. +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` — length 459 content lines; `[TestClass]` 22, class 23 already `partial`; `[TestMethod]` and `[DataTestMethod]` counts 15 and 1 (16 methods: 4 constructor, 6 GetPressed including the data-row method and the original fault test, 3 ExecuteToggleAsync, 3 HandleToggleClickAsync); the original fault test 212 to 243 fetching the handle post-trigger at 223 to 224; `Harness` 403 to 441, constructor 405 to 417, error-log lambda on the single line 415, `OnInvalidate` 430 to 434, `Invalidations` 436, `Errors` 440; `LoggedError` 446 to 457; discard lambda `(_, _) => { }` at 35. Sibling region re-read: `OnInvalidate?.Invoke(controlId);` at 412 is the null-conditional precedent the hook mirrors. +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs` — length 277 content lines; header usings 1 to 5 (includes Moq); no `[TestClass]`; `partial` at 28; capture-before-trigger 211 to 215; post-trigger fetch at 263; `#endregion` at 275. Not modified; P3-T11 gates its diff to exit 0. +- `TaskMaster.Test/TaskMaster.Test.csproj` — the primary fixture entry at 352, the Race entry at 359 followed by the EngineTogglePressedStateCacheTests entry at 360 and the AssemblyInfo entry at 361; the item group closes at 362. Insertion after 359 lands between two existing entries. +- `TaskMaster/Ribbon/RibbonController.EngineCommands.cs` — the lazily built coordinator 67 to 77, sink at 76. Not modified. +- TaskMaster.Test/packages.config — FluentAssertions 8.11.0 at 7, Moq 4.21.0 at 41, MSTest 4.4.1 at 42 to 44. +- TaskMaster.runsettings — Workers and Scope at 5 and 6, collector 9 to 29; scripts/vscode/TaskMaster.cli.runsettings — Workers and Scope at 5 and 6, no collector. +- scripts/vscode/Invoke-MSTestWithCoverage.ps1 — 462 lines; argument list 41 to 95 (filter 91, trx logger 93); collection throw 262; main 274 to 457 (defaults 296 to 298, discovery 348 to 355, output 361 to 365, results directory 367 to 371, `Discovered` 374, post-process 399 to 402, scope gate 406 to 409, first-party line 410, projection 415 to 423, summary 430 to 447, retention 449 to 453); entry guard 459 to 461. +- scripts/vscode/Invoke-MSTestWithCoverage.Helpers.ps1 — `Get-CoberturaClassLineSummary` 160 to 258 (LineMap 192, axes 195 to 196, outputs 251 to 257); `Merge-CoberturaClassesByFilename` 260 to 405; `ConvertTo-KoverageCoberturaXml` 407 to 471 (filename rewrite 437 to 439, merge 442, root counters 455 to 461). +- scripts/vscode/Invoke-MSTestWithCoverage.Threshold.ps1 — line floor 52 to 55, branch floor 122 to 125. scripts/vscode/Invoke-MSTestWithCoverage.Scope.ps1 — gate 59 to 104. scripts/vscode/Invoke-MSTestWithCoverage.FirstParty.ps1 — line shape 117 to 120, report 123 to 162. scripts/vscode/Invoke-MSTestWithCoverage.Projection.ps1 — projection 14 to 81, reconciliation 83 to 146, retention 148 to 197. scripts/vscode/Invoke-MSTest.TrxSummary.ps1 — `Get-TrxRunSummary` 12 to 101, `Format-TrxRunSummary` 103 to 150. +- scripts/vscode/Invoke-Restore.ps1 — parameters SolutionPath, Configuration, Platform at 1 to 10. scripts/vscode/Install-RepoDotNetSdk.ps1 — version default 8.0.205 at 3 and 45. global.json — 8.0.205, latestFeature, .dotnet-sdk path. dotnet-tools.json — csharpier 1.2.6 at the repository root. +- .gitignore — 140 to 141 coverage extensions, 144 the coverage-directory ignore, 145 the gitkeep re-include. .csharpierignore — 4 to 8, 12 to 14, 16, 18. coverage.config — excludes 14 to 20. coverage\.gitkeep present. +- .claude/hooks/validate-planner-output.ps1 — path regex at 95 applied at 304 to 306; final-phase QA vocabulary at 339 to 340. +- .claude/rules/plan-acceptance-gates.md — rule table 31 to 44, attribution window 52 to 54; applied to every command span in this plan (every `git diff` carries the anchor literal; every name-listing diff has a porcelain companion; the formatter task records hashes and a scoped porcelain before-and-after; no coverage argument of the Python form appears). +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` — status and version at 7 and 8; Write Set 64 to 71; acceptance checkboxes 228 to 241 (fourteen, all unchecked); the fixed evidence names at 211, 212, 218, 220, 221. +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md` — work mode at 12; no acceptance section. +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md` — sections 2.1, 4, 6 (Approach 1 and rejected alternatives), 7, 8.1 to 8.4, 9. +- Literal absence checks — `OnLogError`, `handleSeenBySink`, `PrimeHandleStaysRegisteredUntilFaultIsLogged`, `PrimeFaultOrdering`, `has lost isolation`, `error-log sink`, `only then is the in-flight marker cleared`, `Report-then-clear is load-bearing`, `cleared only after that report has returned`: 0 occurrences across every .cs and .csproj file. Banned determinism tokens across the three coordinator test files: 0. + +Sibling-region findings that shaped this plan: + +1. The coverage runner now deletes the raw Cobertura document unless it was written to the repository coverage directory itself (Projection.ps1 148 to 197), and writes a JaCoCo projection plus a trx summary beside it. The plan therefore leaves the runner's default output path in place, copies stage documents under the ignored coverage directory, and transcribes the projection, the first-party line, the summary and the per-file figures into Markdown; no xml file is written into the feature folder (AC12 forbids any). +2. The runner's post-processor merges every class element sharing a filename into one (Helpers.ps1 260 to 405) and rewrites filenames to backslash-relative paths, so the coordinator figures are read from exactly one class node after a separator normalisation, and `COORD-CLASS-NODES: 1` is a hard gate. +3. The runner prints a count of discovered assemblies and never a path; the DIRECT route prints relative `ASSEMBLY:` lines from its own enumeration, and the RUNNER route records only the count line. +4. The pre-fix `catch (` and `lock (` counts are both exactly 1, so AC1's no-catch, no-lock clause is gated on those counts staying at 1 rather than on a zero that could never be false. +5. The `GetPrimeTask` `returns` element sits at 242 to 246 and the `CompletePrime` summary and body at 327 to 355, with `StartPrimeIfNeeded`, `StartObservedPrime`, `ApplyPrimeAsync` and the `remarks` element between or around them. A single lower bound would admit a hunk in any of those members, so the production hunk-position clause is the two-window rule of P2-T6: every hunk's new-side span lies wholly within [G-6, G-1] or within [S-2, R], with G, S and R read from single-occurrence tokens (revised in round 1, delta D5). +6. The main fixture's `[TestMethod]` count is 15, not 16: the null-or-whitespace key test is a `[DataTestMethod]`, whose text does not contain the `[TestMethod]` substring. P1-T1 gates on 15. +7. The branch was cut at the anchor and carries only documentation commits (five at round 2), one of which wrote the promotion record under docs/features/potential/promoted/ (fact 14), so an anchored footprint gate must account for that path. In round 1 the spec's AC13 was amended to name the record as its sole exemption (delta D7, option a), P0-T4 stops if the inherited set holds anything else, and P3-T14 and P3-T27 subtract exactly that path; the earlier divergence note is no longer needed. Uncommitted .claude/agent-memory/ files appear only in porcelain output and are never staged. + +## Planner Internal Review Record + +PLANNER-INTERNAL-REVIEW: PASS + +CITATION-TO-TREE: PASS +AC-TRACEABILITY: PASS +SCOPE-BOUNDARY: PASS + +CITATION: TaskMaster/Ribbon/EngineToggleStateCoordinator.cs | length 415; lines 177, 181, 237-255, 247, 261-278, 269, 276, 290-303, 327-340, 341-355, 343, 345, 346, 348, 350-352, 354 +CITATION: TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs | length 459; lines 22-23, 35, 212-243, 223-224, 403-441, 405-417, 412, 415, 419-420, 430-434, 436, 440, 446-457 +CITATION: TaskMaster.Test/Ribbon/RibbonControllerTests.cs | lines 194, 223, 365 (the only other MockBehavior.Strict occurrences under TaskMaster.Test/Ribbon; outside every gate) +CITATION: docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md | lines 5, 9 +CITATION: TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs | length 277; lines 1-5, 28, 211-215, 263, 275 +CITATION: TaskMaster.Test/TaskMaster.Test.csproj | lines 352, 359, 360, 361, 362 +CITATION: TaskMaster/Ribbon/RibbonController.EngineCommands.cs | lines 67-77 +CITATION: TaskMaster.Test/packages.config | lines 7, 41, 42-44 +CITATION: TaskMaster.runsettings | lines 5-6, 9-29 +CITATION: scripts/vscode/TaskMaster.cli.runsettings | lines 5-6 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.ps1 | length 462; lines 41-95, 91, 93, 262, 274-457, 296-298, 348-355, 374, 399-402, 406-410, 415-423, 430-453, 459-461 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.Helpers.ps1 | lines 160-258, 260-405, 407-471 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.Threshold.ps1 | lines 52-55, 122-125 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.Scope.ps1 | lines 59-104 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.FirstParty.ps1 | lines 117-120, 123-162 +CITATION: scripts/vscode/Invoke-MSTestWithCoverage.Projection.ps1 | lines 14-81, 83-146, 148-197 +CITATION: scripts/vscode/Invoke-MSTest.TrxSummary.ps1 | lines 12-101, 103-150 +CITATION: scripts/vscode/Invoke-Restore.ps1 | lines 1-10 +CITATION: scripts/vscode/Install-RepoDotNetSdk.ps1 | lines 3, 45 +CITATION: global.json | sdk version, rollForward and paths +CITATION: dotnet-tools.json | csharpier 1.2.6 +CITATION: .gitignore | lines 140-141, 146-147, 150, 151 +CITATION: .csharpierignore | lines 4-8, 12-14, 16, 18 +CITATION: coverage.config | lines 14-20 +CITATION: .claude/hooks/validate-planner-output.ps1 | lines 95, 304-306, 339-340 +CITATION: .claude/rules/plan-acceptance-gates.md | lines 31-44, 52-54 +CITATION: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md | lines 6-8 (header advanced to version 0.3 in round 2), 64-71, 211-221, 228-241, 240 (AC13 as amended in round 1) +CITATION: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/issue.md | line 12 +CITATION: docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md | sections 2.1, 4, 6, 7, 8.1-8.4, 9 + +AC-INVENTORY: AC1, AC2, AC3, AC4, AC5, AC6, AC7, AC8, AC9, AC10, AC11, AC12, AC13, AC14 + +AC-MAPPING: AC1 | IMPLEMENTATION: P2-T1 | TESTS: P2-T6, P3-T2 | EVIDENCE: evidence/qa-gates/production-reorder-scope.md +AC-MAPPING: AC2 | IMPLEMENTATION: P2-T2 | TESTS: P2-T6, P3-T2 | EVIDENCE: evidence/qa-gates/production-reorder-scope.md +AC-MAPPING: AC3 | IMPLEMENTATION: P1-T1 | TESTS: P2-T6, P3-T2 | EVIDENCE: evidence/qa-gates/harness-hook-edit-scope.md +AC-MAPPING: AC4 | IMPLEMENTATION: P1-T2 | TESTS: P1-T2, P2-T4, P3-T2 | EVIDENCE: evidence/regression-testing/build-before-reorder.md, evidence/regression-testing/prime-fault-ordering-pass-after.md +AC-MAPPING: AC5 | IMPLEMENTATION: P1-T2 | TESTS: P1-T2, P2-T6, P3-T2 | EVIDENCE: evidence/regression-testing/build-before-reorder.md, evidence/qa-gates/harness-hook-edit-scope.md +AC-MAPPING: AC6 | IMPLEMENTATION: P1-T3 | TESTS: P1-T3, P2-T4 | EVIDENCE: evidence/qa-gates/csproj-registration.md, evidence/regression-testing/prime-fault-ordering-pass-after.md +AC-MAPPING: AC7 | IMPLEMENTATION: P1-T1, P1-T2, P1-T3 | TESTS: P1-T5 | EVIDENCE: evidence/regression-testing/prime-fault-ordering-fail-before.md +AC-MAPPING: AC8 | IMPLEMENTATION: P2-T1 | TESTS: P2-T4 | EVIDENCE: evidence/regression-testing/prime-fault-ordering-pass-after.md +AC-MAPPING: AC9 | IMPLEMENTATION: N/A no-change requirement on the original test | TESTS: P2-T4, P2-T5, P2-T7, P3-T2 | EVIDENCE: evidence/regression-testing/prime-fault-ordering-pass-after.md, evidence/qa-gates/original-test-unchanged.md +AC-MAPPING: AC10 | IMPLEMENTATION: N/A prohibition on the test-side change | TESTS: P3-T11 | EVIDENCE: evidence/qa-gates/determinism-tokens.md +AC-MAPPING: AC11 | IMPLEMENTATION: N/A toolchain requirement | TESTS: P3-T1, P3-T4, P3-T5, P3-T6, P3-T8 | EVIDENCE: evidence/qa-gates/toolchain-final-pass.md +AC-MAPPING: AC12 | IMPLEMENTATION: N/A coverage-evidence requirement | TESTS: P0-T14, P3-T8, P3-T10, P3-T12 | EVIDENCE: evidence/baseline/coverage-baseline.md, evidence/qa-gates/coverage-post-change.md, evidence/qa-gates/footprint-scope.md +AC-MAPPING: AC13 | IMPLEMENTATION: N/A scope-boundary requirement | TESTS: P2-T6, P3-T2, P3-T11, P3-T14 | EVIDENCE: evidence/qa-gates/footprint-scope.md, evidence/qa-gates/determinism-tokens.md, evidence/qa-gates/production-reorder-scope.md +AC-MAPPING: AC14 | IMPLEMENTATION: P1-T2 (separate partial keeps the primary fixture under the ceiling) | TESTS: P3-T3 | EVIDENCE: evidence/qa-gates/file-line-counts.md + +UNRESOLVED-GAPS: NONE + +DIRECTIVE: PREFLIGHT VALIDATION ONLY diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/policy-audit.2026-09-30T08-30.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/policy-audit.2026-09-30T08-30.md new file mode 100644 index 000000000..9834fff96 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/policy-audit.2026-09-30T08-30.md @@ -0,0 +1,235 @@ +# Policy Compliance Audit — engine-toggle-prime-fault-logging-test-races (Issue #942) + +- Component: `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (`CompletePrime` report-then-clear ordering) and its MSTest fixture +- Branch: `bug/engine-toggle-prime-fault-logging-test-races-942` +- Base (merge base with origin/main): `231e1c0b55105aeb626bf5a6e8d0266a567cacad` (origin/main as merged into the branch at `fadcb6417`; plan correction C1) +- Work mode: `full-bug` (issue.md line 12); acceptance-criteria source: `spec.md` only (14 items) +- Review date label: 2026-09-30T08-30. No clock was available to the review session (no shell tool); the label was assigned as the first quarter-hour later than every executor evidence label (latest 2026-09-30T07-57) and is disclosed as such. The executor's labels are local time: the final Cobertura document carries epoch `timestamp="1790769007"` (2026-09-30T11:50:07Z) against the executor's `07-50` label, an offset of UTC-4. +- Reviewer tooling: Read, Grep and Glob only. No `git`, `msbuild`, `vstest`, `dotnet` or MCP invocation was possible, so no toolchain step was re-run; every gate below is verified from the committed evidence projections, the raw post-processed Cobertura documents the executor left under the git-ignored `coverage/` directory, and direct reads of the four changed code files. + +## Template Resolution Deviation + +The MCP tools `mcp__drm-copilot__resolve_policy_audit_template_asset` and `mcp__drm-copilot__validate_orchestration_artifacts` were not exposed to this review session. This artifact is hand-authored preserving all twelve canonical major headings listed in `.claude/skills/policy-audit-template-usage/SKILL.md` step 5, plus the per-language coverage comparison block, the coverage evidence checklist and the coverage metrics table in the shape the orchestrator's validator expects. The audit is not marked BLOCKED on that basis: every section below is evidence-backed and the template's instruction block is not present. + +## Executive Summary + +- Overall verdict: **PASS**. 0 FAIL findings, 0 blocking PARTIAL findings, 4 non-blocking observations (PA-1 to PA-4 in section 8). +- The change is a two-statement reorder inside a private method plus documentation, a test-fixture observer hook, one new deterministic regression test in a new partial, and one `Compile Include` line. No public API, dependency, configuration, run-settings or run-regime change. +- Bugfix workflow satisfied in order: regression test written first and observed failing against the byte-identical base production file (SHA-256 `F2A961DD…CEF0` at control equals the baseline hash), then the minimal fix, then the full CLAUDE.md toolchain in one uninterrupted pass (format, check, analyzer rebuild, nullable rebuild, coverage-enabled tests; exit 0 at every step; `SKIP_CORECOMPILE_LINES: 0` on both rebuilds). +- C# coverage verdict: PASS. First-party processed Cobertura 85.31% lines (56080/65736) and 79.72% branches (13596/17054) after the change against 85.32%/79.73% before, both above the 85%/75% floors of `.claude/rules/quality-tiers.md` and the 80%/75% floors of CLAUDE.md; the only changed production file is unchanged at 100% lines (143/143) and 97.37% branches (37/38), and the one moved executable line (359) has hits=1. +- No language other than C# has changed files on the branch (`.cs` x3, `.csproj` x1, Markdown under the feature folder and the inherited promotion record). +- Scope narrowing attempts: none. Evidence-location violations: none. + +## Rejected Scope Narrowing + +None detected. The caller's prompt supplied the resolved base, the full branch footprint (four code files, the feature folder, the inherited promotion record) and asked that the spec's declared non-goal (hazard B) be listed as a follow-up rather than treated as blocking. Listing a spec non-goal as non-blocking is the spec's own scope decision, not a narrowing of the audit scope; the full branch diff against `231e1c0b` was audited. + +## Evidence Location Compliance + +- Every evidence artifact on the branch lives under `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/{baseline,regression-testing,qa-gates,other}/`, the canonical `/evidence//` layout (plan "Evidence location" paragraph; 37 Markdown files enumerated by Glob). +- Branch diff scan for `artifacts/baselines/`, `artifacts/qa/`, `artifacts/evidence/`, `artifacts/coverage/`: the executor's anchored `--diff-filter=A` enumeration (`evidence/qa-gates/footprint-scope.md`) lists 29 added paths, none under `artifacts/`; a Grep of the feature folder for those four prefixes returns no match; the item worktree contains no `artifacts/` directory at all. Zero FAIL-level findings. +- `validate_evidence_locations.py --root .` could not be executed (no shell); the manual scan above substitutes for it and is recorded as such. +- `EVIDENCE_LOCATION_OVERRIDE_REJECTED`: none required; no caller instruction supplied a non-canonical evidence path. + +## 1. General Unit Test Policy Compliance + +| Requirement | Verdict | Evidence | +|---|---|---| +| Independence and isolation | PASS | The new test builds its own `Harness`, one strict mock, one held `TaskCompletionSource`; it shares no static state; it targets one behavior (marker registered at report time). | +| Fast, deterministic | PASS | Outcome is a function of program order on the pool thread (the sink reads `GetPrimeTask` on the same thread that will run `TryRemove`); no sleep, delay, timer, retry, gate or parallelism attribute (`evidence/qa-gates/determinism-tokens.md`, 20 tokens at 0; confirmed by direct read of both test files). | +| Readability, AAA, intent documented | PASS | `// Arrange`, `// Act`, `// Assert` markers; XML summary names issue #942 and states the invariant; every FluentAssertions call carries a reason string. | +| Scenario completeness | PASS (bug-fix scope) | Positive fault path (new test + existing fault test), cancellation path (two existing Race tests), success early-return (existing prime-success tests), null/unmapped keys (existing). The single `CompletePrime` path covers faulted and canceled outcomes together. | +| No external dependencies, no temp files | PASS | Strict Moq mock of `IAppItemEngines`; no filesystem, network, process or clock. | +| Test file location | PASS by repository convention | C# tests live in `.Test//` mirroring `//` throughout this repository; the new partial sits beside its two siblings. The rule file's `tests/` wording is the cross-language default and is not the convention applied to C# projects here (pre-existing, informational). | +| Coverage, no regression on changed lines | PASS | See the comparison block below and section 5. | + +### Coverage Evidence Checklist + +- C# baseline coverage artifact: `coverage/baseline-942.cobertura.xml` (git-ignored, read directly at root and class level) projected to `evidence/baseline/coverage-baseline.md` +- C# post-change coverage artifact: `coverage/final-942.cobertura.xml` (git-ignored, read directly at root and class level) projected to `evidence/qa-gates/coverage-post-change.md` +- TypeScript baseline coverage artifact: `N/A - out of scope` +- TypeScript post-change coverage artifact: `N/A - out of scope` +- PowerShell baseline coverage artifact: `N/A - out of scope` +- PowerShell post-change coverage artifact: `N/A - out of scope` +- Python baseline coverage artifact: `N/A - out of scope` +- Python post-change coverage artifact: `N/A - out of scope` +- Per-language comparison summary: section 1.2.1 of this document + +### 1.2.1 Per-Language Coverage Comparison + +- C#: Baseline: 85.32% lines (56083/65736), 79.73% branches (13597/17054). Post-change: 85.31% lines (56080/65736), 79.72% branches (13596/17054). Change: -0.01% lines (-3 covered lines, denominator unchanged at 65736) and -0.01% branches (-1 covered branch), located in packages other than the changed file and within the plan's D-7 run-to-run tolerance; the changed file is identical at both stages at 100% lines (143/143) and 97.37% branches (37/38). New/changed-code coverage: 100%. Disposition: PASS. Evidence: evidence/baseline/coverage-baseline.md, evidence/qa-gates/coverage-post-change.md (COMPARISON), coverage/final-942.cobertura.xml class element at document line 230308 read directly. +- TypeScript: Baseline: N/A. Post-change: N/A. Change: N/A. Disposition: N/A. Evidence: N/A - zero TypeScript files changed on this branch. +- Python: Baseline: N/A. Post-change: N/A. Change: N/A. Disposition: N/A. Evidence: N/A - zero Python files changed on this branch. +- PowerShell: Baseline: N/A. Post-change: N/A. Change: N/A. Disposition: N/A. Evidence: N/A - zero PowerShell files changed on this branch. + +### 1.2.2 Coverage Artifact State + +| Language | Artifact read by this review | State | +|---|---|---| +| C# | `coverage/final-942.cobertura.xml` and `coverage/baseline-942.cobertura.xml` in the item worktree (post-processed by `ConvertTo-KoverageCoberturaXml`, workspace-relative filenames) | Present; root element and the coordinator class element read directly; figures match the committed projections exactly | +| C# (hook canonical path) | `artifacts/csharp/coverage.xml` in the item worktree | Not populated; the DIRECT route writes to `coverage/` and CLAUDE.md "Committed Test Evidence Format" prohibits committing the raw document. Recorded as observation PA-1, non-blocking, because the same document was verified at its runner location | + +C# coverage verdict (thresholds and changed-line regression, first-party processed Cobertura): PASS. +C# coverage note (canonical hook path `artifacts/csharp/coverage.xml`): not populated in the item worktree; substitute document verified directly at `coverage/final-942.cobertura.xml`; this is an observation, not a threshold verdict. + +**Coverage Metrics by Language:** + +| Language | Files Changed | Tests | Test Result | Baseline Coverage | Post-Change Coverage | New Code Coverage | +|---|---|---|---|---|---|---| +| C# | 4 (1 production, 2 test, 1 project file) | 25 fixture tests (1 new); 7324 in the coverage-enabled suite | 7324/7324 passed, 0 failed | 85.32% lines / 79.73% branches | 85.31% lines / 79.72% branches | 100% | +| TypeScript | 0 | N/A | N/A | N/A | N/A | N/A | +| Python | 0 | N/A | N/A | N/A | N/A | N/A | +| PowerShell | 0 | N/A | N/A | N/A | N/A | N/A | + +## 2. General Code Change Policy Compliance + +| Requirement | Verdict | Evidence | +|---|---|---| +| Bugfix workflow: failing regression test first | PASS | `evidence/regression-testing/prime-fault-ordering-fail-before.md`: EXIT_CODE 1 = ExpectedExitCode 1; 24 passed / 1 failed; the failing message is the FluentAssertions `BeSameAs` text ("Expected handleSeenBySink to refer to ... ContinuationTaskFromTask ... but found System.Threading.Tasks.Task"), i.e. the sink observed `Task.CompletedTask`; production file hash equals the baseline hash. | +| Minimal, targeted fix | PASS | Numstat `10 5` on the production file: five removed lines are three documentation lines, the `TryRemove` statement and one blank line; no line of the early return, the failure computation or the synthesized exception was rewritten (`evidence/qa-gates/production-reorder-scope.md`, confirmed by reading lines 344–360). | +| Verify locally before review, full toolchain in order | PASS | `evidence/qa-gates/toolchain-final-pass.md`: pass 1 clean; format (0 rewrites), check ("no differences"), analyzer rebuild (0 errors, 0 warnings, `SKIP_CORECOMPILE_LINES: 0`), nullable rebuild (same), fixture run 25/25, coverage-enabled suite 7324/7324. | +| Design principles (simplicity, separation of concerns) | PASS | Two statements reordered; no new abstraction, seam, lock or constructor parameter; I/O stays behind the injected delegates. | +| Error handling, fail fast | PASS | No new `try`/`catch`/`finally`; the type keeps exactly one `catch (` (line 181) and one `lock (` (line 271). | +| Module and file size (500 lines) | PASS | 420 / 470 / 77 lines (`evidence/qa-gates/file-line-counts.md`; confirmed by direct read: 420, 470, 77). | +| Naming, docs, comments explain why | PASS | Summary of `CompletePrime`, returns of `GetPrimeTask` and a three-line "Report-then-clear is load-bearing" comment state the invariant and its reason. | +| Dependencies | PASS | None added; the new partial omits `using Moq;` because it does not name Moq directly. | +| Public API compatibility | PASS | `CompletePrime` and `Harness` are private; `GetPrimeTask` is internal with an unchanged signature and a strengthened post-condition. `GetPrimeTask(` has no production caller (Grep over `TaskMaster/`: only the declaration at line 249). | +| Seven-stage loop stages 4, 6, 7 (architecture-boundary, contract/schema, integration) | Not evaluable, pre-existing | No `*.ArchitectureTests` project, contract-check tooling or integration-test stage exists for C# in this repository; CLAUDE.md's four-step C# toolchain is the operative loop and was run. Observation PA-3. | + +## 3. Language-Specific Code Change Policy Compliance + +C# (`.claude/rules/csharp.md`, CLAUDE.md C#1–C#7): + +| Requirement | Verdict | Evidence | +|---|---|---| +| CSharpier via `dotnet tool run`, check clean | PASS | Step 1/2 of the final pass; "Checked 1626 files", no differences. The csproj is excluded by `.csharpierignore` line 12 (confirmed). | +| Analyzer rebuild command verbatim, `/t:Rebuild` | PASS | Command matches CLAUDE.md character-for-character; `SKIP_CORECOMPILE_LINES: 0`, `CSC_OUT_TASKMASTER 2`, `CSC_OUT_TASKMASTER_TEST 2`. | +| Nullable rebuild command verbatim, no `/p:Nullable=enable` | PASS | Command matches; no `/p:Nullable=enable`; 0 errors, 0 warnings. | +| Test step (`Invoke-MSTestWithCoverage.ps1` / VS Code task) | PASS with disclosed substitution | The runner hard-codes its assembly filter; plan D-6 selects the DIRECT route when the shell-icon stall probe does not read CLEAR (`evidence/baseline/stall-probe.md`: one `ShellUtilities_Tests` test failed with a Win32 icon-handle `ArgumentException`, a pre-existing workstation defect). The DIRECT route issues the runner's own inner `dotnet-coverage collect ... vstest.console.exe` invocation with the four classes excluded and post-processes with the runner's own helpers; CI executes the four classes. Observation PA-2. | +| Naming, XML docs, `internal` surface | PASS | PascalCase members, camelCase locals; XML docs updated on both touched members; nothing made public. | +| Null-safety | PASS | No nullable directive in any touched file (pre-existing state); no new nullable warning under `/p:TreatWarningsAsErrors=true`. | +| Banned symbols (`Thread.Sleep`, `Task.Delay`, `DateTime.Now`) | PASS | Zero occurrences added (determinism-tokens; direct read). | +| Prohibited behaviors (sleeps, retries, weakened assertions, `[DoNotParallelize]`) | PASS | None; the original test is byte-for-byte unchanged (method SHA equal at base and head, `evidence/qa-gates/original-test-unchanged.md`). | + +## 4. Language-Specific Unit Test Policy Compliance + +| Requirement | Verdict | Evidence | +|---|---|---| +| MSTest attributes | PASS | `[TestMethod]` on the new test; `[TestClass]` remains on the primary partial only (the Race partial follows the same shape). | +| Moq for mocks, strict | PASS | `Harness.Engines` is `new Mock(MockBehavior.Strict)` (line 424); the new test sets up exactly the one call the prime makes. | +| FluentAssertions with reasons | PASS | Six assertions, each with a reason string. | +| AAA structure | PASS | Explicit markers at lines 29, 40, 44 of the new partial. | +| Deterministic test rules; parallel regime unchanged | PASS | `TaskMaster.runsettings` and `scripts/vscode/TaskMaster.cli.runsettings` diff-clean against the base (`RUNSETTINGS_DIFF_EXIT=0`); the test passes under Workers=0 / ClassLevel in the pass-after and final runs. | +| Coverage: new module/class/method >= 90% | PASS (vacuous) | No new production module, class or method; the changed method remains 100% covered. | +| Coverage: changed lines not reduced | PASS | Line 359 (moved `TryRemove`) hits=1; lines 345–348, 351–353, 358, 360 hits=1; branch conditions 2/2 (346) and 4/4 (351) — read directly from the final Cobertura. | +| Test files excluded from the denominator | PASS | `coverage.config` excludes third-party and test modules; the JaCoCo projection lists production packages only. | + +## 5. Test Coverage Detail + +Per-file figures (first-party processed Cobertura, both stages read directly at the class element): + +| File | Baseline | Post-change | Changed executable lines | Notes | +|---|---|---|---|---| +| `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | 143/143 lines (100%), 37/38 branches (97.37%) | 143/143 lines (100%), 37/38 branches (97.37%) | 1 (line 359, hits=1) | The single uncovered branch is the `RenderEngineName` null-name arm (line 367, 1/2), pre-existing and outside the change. Lines-valid unchanged at 143: the change adds no executable statement. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` | excluded (test) | excluded (test) | — | Harness hook only. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` | excluded (test) | excluded (test) | — | New partial; its one test executed and passed (RESULT line in the pass-after and FINAL-FIXTURE-RUN sections). | +| `TaskMaster.Test/TaskMaster.Test.csproj` | not coverage-bearing | not coverage-bearing | — | One `Compile Include` at line 360, immediately after the Race entry. | + +Repo-wide (package-level JaCoCo projection, identical package set at both stages): the -3 covered lines are UtilitiesCS -4 and QuickFiler +1; the -1 covered branch is UtilitiesCS. Denominators are identical at every package. The TaskMaster package is unchanged at 2443/3245 lines and 517/728 branches. This variation is run-to-run noise outside the changed file, consistent with the nondeterminism recorded for this suite on earlier reviews, and is recorded rather than attributed. + +Thresholds evaluated: `.claude/rules/quality-tiers.md` 85% lines / 75% branches — met (85.31 / 79.72). CLAUDE.md 80% lines / 75% branches, 90% for new code — met (no new production code; changed method 100%). Observation PA-4 records the floor discrepancy between the two documents; the verdict is identical under either. + +## 6. Test Execution Metrics + +| Run | Command (abbreviated) | Total | Passed | Failed | Exit | Evidence | +|---|---|---|---|---|---|---| +| Baseline fixture (P0-T12) | vstest, coordinator class filter | 24 | 24 | 0 | 0 | `evidence/baseline/coordinator-tests-baseline.md` | +| Baseline suite with coverage (P0-T14) | dotnet-coverage collect + vstest, 9 assemblies, DIRECT filter | 7323 | 7323 | 0 | 0 | `evidence/baseline/coverage-baseline.md` | +| Fail-before (P1-T5) | vstest, coordinator class filter, base production file | 25 | 24 | 1 (new test, `BeSameAs`) | 1 (expected 1) | `evidence/regression-testing/prime-fault-ordering-fail-before.md` | +| Pass-after (P2-T4) | same command after the reorder | 25 | 25 | 0 | 0 | `evidence/regression-testing/prime-fault-ordering-pass-after.md` | +| Final fixture (P3-T7) | same command on the analyzer/nullable-rebuilt assembly | 25 | 25 | 0 | 0 | same file, FINAL-FIXTURE-RUN | +| Final suite with coverage (P3-T8) | dotnet-coverage collect + vstest, 9 assemblies, DIRECT filter | 7324 | 7324 | 0 | 0 | `evidence/qa-gates/coverage-post-change.md` | + +Population arithmetic corroborated independently: 15 `[TestMethod]` + 1 `[DataTestMethod]` with 3 rows in the primary partial, 6 in the Race partial, 1 in the new partial = 25; suite 7323 + 1 = 7324. + +## 7. Code Quality Checks + +| Check | Result | Evidence | +|---|---|---| +| Formatting (CSharpier check) | PASS, no differences | toolchain-final-pass step 2 | +| Linting (.NET analyzers, `/t:Rebuild`) | PASS, 0 errors, 0 warnings, CoreCompile ran | toolchain-final-pass step 3 | +| Type checking (`/p:TreatWarningsAsErrors=true`, `/t:Rebuild`) | PASS, 0 errors, 0 warnings, CoreCompile ran | toolchain-final-pass step 4 | +| Testing (MSTest, coverage-enabled) | PASS, 7324/7324 | toolchain-final-pass step 5b | +| File-size ceiling | PASS, 420 / 470 / 77 | direct read | +| Host-path hygiene of committed evidence | PASS | `evidence/qa-gates/evidence-hygiene.md` (36 files, 0 account/machine/drive-profile hits); independent Grep of the feature folder for any drive-letter path (`[A-Za-z]:[\\/][A-Za-z]`) returned no match, so the tool-banner escape seen on #930 is absent here | +| Raw tool documents committed | PASS, none | footprint enumeration (`RAW-DOCS-COMMITTED: 0`, `RAW-DOCS-UNTRACKED-IN-FEATURE: 0` with `--ignored`); Glob of the feature folder lists Markdown only | + +## 8. Gaps and Exceptions + +| ID | Severity | Blocking | Description | Disposition | +|---|---|---|---|---| +| PA-1 | Low | No | The hook-canonical C# coverage path `artifacts/csharp/coverage.xml` is not populated in the item worktree; the DIRECT route writes `coverage/final-942.cobertura.xml`, which this review read directly, and the committed projection carries every figure. | Accept. Populate the canonical path only if a later gate requires the hook to parse it; do not commit the raw document (CLAUDE.md Committed Test Evidence Format). | +| PA-2 | Low | No | Toolchain step 4 ran the DIRECT route (inner collector invocation with four shell-icon test classes excluded) instead of `Invoke-MSTestWithCoverage.ps1` verbatim, per plan D-6 after the stall probe recorded a workstation-local `ArgumentException` in `ShellUtilities_Tests`. Baseline and final used the identical exclusion, so the comparison is like-for-like; CI executes the excluded classes. | Accept as a disclosed substitution. The PR CI `mstest-coverage` run remains the repo-wide gate for those four classes. | +| PA-3 | Informational | No | `quality-tiers.yml` is absent at the repository root (Glob), so tier-dependent gates (property-test density, mutation score) cannot be evaluated; stages 4, 6 and 7 of the seven-stage loop have no C# tooling in this repository. Pre-existing; unchanged by this branch. | Record only. | +| PA-4 | Informational | No | CLAUDE.md states 80% line / 90% new-code floors (maintainer decision 2026-09-11, #563) while `.claude/rules/quality-tiers.md` and `general-unit-test.md` state 85% / 75% uniform floors. Both are met here; no verdict depends on which governs. | Record only; maintainer-owned reconciliation. | +| PA-5 | Informational | No | PR context artifacts (`artifacts/pr_context.summary.txt` / `.appendix.txt`) do not exist in the item worktree; the session checkout holds a stale pair for a different branch (#936). They could not be regenerated (no shell or MCP). Scope was taken from the resolved base, the orchestrator's verified footprint and the executor's anchored `git diff --name-status` enumeration, and cross-checked against direct reads. | Record only. | + +No exception to any unit-test rule was claimed by the change and none is needed. + +## 9. Summary of Changes + +| File | Status | Lines (+/-) | Nature | +|---|---|---|---| +| `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | M | +10 / -5 | `CompletePrime`: `_primeTasks.TryRemove` moved after `_logError`; three-line why-comment; summary and `GetPrimeTask` returns documentation updated. Early return, failure unwrap, synthesized `TaskCanceledException` unchanged. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` | M | +12 / -1 | `Harness.OnLogError` (`Action`, documented) invoked null-conditionally after `Errors.Add` in the error-log lambda; both hunks inside the `Harness` type; `[TestMethod]` count unchanged at 15. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` | A | +77 | Third partial; one test `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged`. | +| `TaskMaster.Test/TaskMaster.Test.csproj` | M | +1 / -0 | `Compile Include` for the new partial at line 360. | +| `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/**` | A | — | issue, spec (v0.3), research, plan (v1.4 with C1/C2), 37 evidence projections. | +| `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` | A (inherited) | +60 | Promotion record written by the promotion commit; named as AC13's sole exemption. | + +## 10. Compliance Verdict + +**PASS.** 0 FAIL, 0 blocking PARTIAL. Non-blocking observations: PA-1 to PA-5. No remediation inputs are produced. The bugfix workflow (RED first, minimal fix, full toolchain) is evidenced end to end, the invariant the fix restores is stated in code and documentation, the regression test discriminates on program order rather than scheduling, and coverage of the changed file is unchanged at 100% lines with the moved statement covered. + +## Appendix A: Test Inventory + +Coordinator fixture (`TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests`, three partials), 25 executed tests in the pass-after and final runs: + +| Partial | Test | Status (final) | +|---|---|---| +| Primary | `Constructor_WithNullEnginesAccessor_ThrowsArgumentNullException` | Passed | +| Primary | `Constructor_WithNullInvalidateDelegate_ThrowsArgumentNullException` | Passed | +| Primary | `Constructor_WithNullNotifyDelegate_ThrowsArgumentNullException` | Passed | +| Primary | `Constructor_WithNullLogErrorDelegate_ThrowsArgumentNullException` | Passed | +| Primary | `GetPressed_WithNullOrWhitespaceKey_ReturnsFalseWithoutPrimeOrInvalidate` (3 data rows) | Passed x3 | +| Primary | `GetPressed_WithUnmappedKey_ReturnsFalseWithoutPrime` | Passed | +| Primary | `GetPressed_WhenEnginesAccessorReturnsNull_ReturnsFalseAndStartsNothing` | Passed | +| Primary | `GetPressed_OnCacheMissWithEnginesAvailable_StartsExactlyOnePrime` | Passed | +| Primary | `GetPressed_AfterPrimeCompletes_ReturnsPrimedValueAndInvalidatesMappedControl` | Passed | +| Primary | `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` (original reproduction, unchanged) | Passed | +| Primary | `ExecuteToggleAsync_PerformsToggleThenRefreshThenCacheThenInvalidate_InOrder` | Passed | +| Primary | `ExecuteToggleAsync_WhenToggleFaults_PropagatesUnchanged` | Passed | +| Primary | `ExecuteToggleAsync_WithUnmappedKey_ThrowsArgumentException` | Passed | +| Primary | `HandleToggleClickAsync_WhenToggleFaults_LogsErrorDoesNotThrowDoesNotInvalidate` | Passed | +| Primary | `HandleToggleClickAsync_WithNullEngines_NotifiesOnceAndInvokesNothing` | Passed | +| Primary | `HandleToggleClickAsync_WhenEnginesAvailable_TogglesAndInvalidates` | Passed | +| Race | `ApplyPrimeAsync_WhenPrimeResolvesAfterToggle_DoesNotOverwriteToggleResult` | Passed | +| Race | `ExecuteToggleAsync_WhenOlderObservationCompletesLast_DoesNotOverwriteNewerResult` | Passed | +| Race | `ExecuteToggleAsync_WithNoCompetingWriter_CachesValueAndInvalidatesExactlyOnce` | Passed | +| Race | `ExecuteToggleAsync_WithNullEngines_ThrowsInvalidOperationExceptionWithoutTogglingEngine` | Passed | +| Race | `GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker` | Passed | +| Race | `GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked` | Passed | +| PrimeFaultOrdering (new) | `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged` | Failed at fail-before (expected); Passed at pass-after and final | + +Suite: 7324 tests across 9 assemblies, 7324 passed, 0 failed, 0 skipped (derived), `TestCategory!=LiveOutlook` plus the four shell-icon classes excluded under the DIRECT route. + +## Appendix B: Toolchain Commands Reference + +Executed by the executor and recorded in `evidence/qa-gates/toolchain-final-pass.md` (none re-run by this review): + +1. `dotnet tool run csharpier format .` then `dotnet tool run csharpier check .` +2. `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true` +3. `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true` +4. `dotnet-coverage collect --output coverage\final-942.cobertura.xml --output-format cobertura --settings coverage\effective-coverage-942.config -- vstest.console.exe <9 test assemblies> /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:TestCategory!=LiveOutlook&FullyQualifiedName!~HelperClasses.ShellUtilities_Tests&FullyQualifiedName!~HelperClasses.ShellUtilitiesStatic_Tests&FullyQualifiedName!~HelperClasses.SysImageListHelperTests&FullyQualifiedName!~EmailIntelligence.OSBrowser_Tests" "/ResultsDirectory:coverage\test-results\942\final" "/Logger:trx;LogFileName=final-942.trx" "/Blame:CollectHangDump;TestTimeout=4min;HangDumpType=None"` followed by the runner's own post-processing (`ConvertTo-KoverageCoberturaXml`, threshold functions, first-party line, JaCoCo projection, trx summary) — the DIRECT substitution for `Invoke-MSTestWithCoverage.ps1` (PA-2) + +Regression-test commands: `vstest.console.exe TaskMaster.Test\bin\Debug\TaskMaster.Test.dll /Settings:scripts\vscode\TaskMaster.cli.runsettings /InIsolation "/TestCaseFilter:FullyQualifiedName~TaskMaster.Test.Ribbon.EngineToggleStateCoordinatorTests" ...` (fail-before P1-T5, pass-after P2-T4, final fixture P3-T7; identical apart from the results-directory and trx-name segments). diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md new file mode 100644 index 000000000..456d4c7c9 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/research/2026-09-29T23-20-engine-toggle-prime-fault-race-research.md @@ -0,0 +1,186 @@ +# Research: issue #942 — `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` races the prime-fault log + +- Issue: #942 +- Feature folder: `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/` +- Branch: `bug/engine-toggle-prime-fault-logging-test-races-942` +- Date: 2026-09-29T23-20 +- Evidence tags: `[V]` verified by reading the named file/lines in this worktree; `[D]` derived from documented .NET Framework 4.8 TPL behaviour; `[M]` recalled from agent memory and re-checked against the worktree where possible. Bash was unavailable in this session, so no git history was consulted and no test was executed. + +## 1. Summary of conclusions + +1. **H1 (fault observed in a continuation the test does not await) — REFUTED.** `StartObservedPrime` returns the `ContinueWith` continuation itself and that continuation is what `_primeTasks` stores and `GetPrimeTask` returns (`EngineToggleStateCoordinator.cs:276, 296-302, 254`). Awaiting the handle covers `CompletePrime`, including `_logError`, whenever the handle is obtained. +2. **H2 (marker removed before the log, so a post-trigger `GetPrimeTask` can return `Task.CompletedTask` while the log is still pending) — CONFIRMED.** `CompletePrime` executes `_primeTasks.TryRemove` at line 348 and `_logError` at line 354. The continuation runs on a thread-pool thread (`TaskContinuationOptions.None`, `TaskScheduler.Default`, lines 299-301), while the test calls `GetPrimeTask` on its own thread *after* the trigger (`EngineToggleStateCoordinatorTests.cs:223-224`). If the pool thread reaches line 348 first, the test awaits `Task.CompletedTask` and asserts on `Errors` before, or concurrently with, the `Errors.Add` at line 415. +3. A second, pre-existing ordering hazard (registration at line 276 racing the removal at line 348 when the prime completes synchronously) is reachable from the same test's re-prime at line 237 and in production after a cached configuration-load fault. It does not fail any current test. It was recorded as NB-2 in the #735 code review with a recommendation to promote it to its own issue; no such issue or potential entry exists. It is out of scope for #942 and should be promoted separately. +4. **Recommended fix:** reorder `CompletePrime` so the fault is reported through `_logError` before the marker is cleared (production, two statements), which makes the documented `GetPrimeTask` contract true on every path; add one deterministic regression test that probes the prime handle from inside the injected log sink (no sleeps, gates, timers, blocking, `[DoNotParallelize]`, or worker limits); the negative control is the same test run against the current statement order, where it fails deterministically. +5. No other test in the repository has the same pattern. Two other production sites observe faults in discarded continuations (the genuine H1 shape), but no test asserts on their logs. + +## 2. Current state analysis + +### 2.1 Production: `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` `[V]` + +| Lines | Member | Behaviour relevant to the race | +|---|---|---| +| 77-80 | `_primeTasks` | `ConcurrentDictionary`; doc at 72-76 says its *presence* is the at-most-one-prime guard and its *value* is the test-observable handle. | +| 136-150 | `GetPressed` | Cache miss with engines available calls `StartPrimeIfNeeded` (148) and returns `false`. | +| 247-255 | `GetPrimeTask` | Returns `_primeTasks[engineName]` or `Task.CompletedTask` when no marker is registered. Doc (238-246): "exposed so tests can await the prime deterministically instead of polling or sleeping … The returned task never faults: a prime fault is observed inside the prime itself and reported through `logError`." | +| 261-278 | `StartPrimeIfNeeded` | Under `lock (_primeGate)`: `ContainsKey` check (271), then `_primeTasks[engineName] = StartObservedPrime(...)` (276). `StartObservedPrime` is fully evaluated — including scheduling the continuation — before the assignment. | +| 290-303 | `StartObservedPrime` | `ApplyPrimeAsync(...).ContinueWith(completed => CompletePrime(completed, engineName), CancellationToken.None, TaskContinuationOptions.None, TaskScheduler.Default)`. The **continuation task** is what is returned and stored. Doc (284-288): "The returned continuation task always completes successfully, which is what makes it safe for a test to await." | +| 310-325 | `ApplyPrimeAsync` | `await engines.EngineActiveAsync(engineName).ConfigureAwait(false)` (319); no `catch`. | +| 341-355 | `CompletePrime` | `if RanToCompletion return;` (343) → `_primeTasks.TryRemove(engineName, out _)` (**348**) → unwrap/synthesize exception (350-352) → `_logError(BuildPrimeFailedMessage(engineName), failure)` (**354**). | + +Production wiring `[V]` `TaskMaster/Ribbon/RibbonController.EngineCommands.cs:67-77`: `logError` is `(message, exception) => logger.Error(message, exception)`; `invalidateControl` is `controlId => _viewer?.InvalidateEngineToggle(controlId)`. Neither re-enters the coordinator. `GetPrimeTask` has no production caller (grep over `*.cs`: the only non-test hits are its declaration and its own ``). + +### 2.2 Tests `[V]` + +`TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` (459 lines; the 500-line cap is close): + +- 213-243 `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` (the flaky test): + - 220 `GetPressed(SpamEngine)` starts the prime against a held `TaskCompletionSource` (`probe`). + - 223 `probe.SetException(failure)`; 224 `await harness.Coordinator.GetPrimeTask(SpamEngine)` — **the handle is fetched after the trigger**. + - 227-234 assert `Errors` has exactly one entry, its message and exception, and `Invalidations` empty. + - 236-239 `GetPressed` again (re-primes, because the marker was cleared); 242 cleanup `await GetPrimeTask` (also post-trigger). +- 403-441 `Harness`: `Errors` is a plain `List` (440) appended from the injected `logError` lambda (415); `OnInvalidate` (434) is an extra observer invoked from inside the invalidation sink (409-413) — the existing "probe from inside the sink" precedent used by the ordering test at 270-276. + +`TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.Race.cs` (277 lines, second partial): + +- 204-246 `GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker`: **captures `firstPrime = GetPrimeTask(...)` at 211 before `probe.SetCanceled()` at 214** and awaits that captured handle (215) before asserting `Errors` (218-229). This is the capture-before-trigger pattern and is immune to H2 for its error assertion. Its remarks (195-202) already document why the error count must be asserted before the re-prime. +- 253-273 `GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked`: post-trigger `GetPrimeTask` at 263, but the only assertion (266-269) is `GetPressed == false`, which holds under every interleaving because a canceled prime never writes the cache. +- 38-74 and the two toggle tests use `SetResult` (success path); `CompletePrime` returns at 343 without touching the marker, so `GetPrimeTask` always returns the continuation and the await is deterministic. + +Run regime `[V]`: `TaskMaster.runsettings:4-7` and `scripts/vscode/TaskMaster.cli.runsettings:4-7` both set `Workers=0`, `Scope=ClassLevel`. MSTest is 4.4.1 (`TaskMaster.Test/packages.config:42-44`). `[M]` MSTest 4.4 runs no-`[Timeout]` test bodies on a `Task.Run` worker, so the test thread is itself a thread-pool thread (memory `taskrun-getresult-inlines-on-pool-thread-900`, re-checked against the packages.config version). + +## 3. Hypothesis verdicts + +### H1 — REFUTED + +`StartObservedPrime` (296-302) returns `ApplyPrimeAsync(...).ContinueWith(...)`, i.e. the continuation, and line 276 stores exactly that in `_primeTasks`. `GetPrimeTask` (254) returns the stored value. Awaiting it therefore resumes only after `CompletePrime` has returned, which is after `_logError` has returned. The fault observer is *inside* the awaited task. The issue's "Suspected Cause" is therefore not the mechanism. + +### H2 — CONFIRMED + +The mechanism is the order of lines 348 and 354 combined with the test fetching the handle after the trigger. Detailed interleaving in section 4. + +## 4. Exact interleaving that produces the failure + +Let T be the MSTest worker executing the test (a pool thread `[M]`), and P another pool thread. + +1. T, line 220: `GetPressed` → `StartPrimeIfNeeded` → `ApplyPrimeAsync` runs synchronously to line 319, where `engines.EngineActiveAsync` returns the incomplete `probe.Task` (Moq `.Returns(probe.Task)`, test line 219). The state machine suspends; because of `ConfigureAwait(false)` and the absence of a `SynchronizationContext` on an MSTest worker, the resumption is stored on `probe.Task` as a plain `Action` continuation `[D]`. `ApplyPrimeAsync` returns incomplete task A. `ContinueWith` registers continuation C on A; C is stored at line 276. `[V]` +2. T, line 223: `probe.SetException(failure)`. `TaskCompletionSource.SetException` completes `probe.Task` and runs its continuations **inline on T** (await continuations are inlined by `FinishContinuations` unless the completing thread is aborting or `RunContinuationsAsynchronously` was requested — neither applies) `[D]`. The `ApplyPrimeAsync` state machine resumes on T, the `await` rethrows `failure`, and A becomes Faulted on T. A's continuation C was registered with `TaskContinuationOptions.None` (no `ExecuteSynchronously`), so it is **queued to `TaskScheduler.Default`**, not inlined `[D]` (lines 299-301 `[V]`). Because T is a pool thread, the item lands on T's local work-stealing queue and is picked up either by T once the test body yields, or by an idle pool thread P that steals it `[D]`. `SetException` returns to the test. +3. From here T and P run concurrently: + - T, line 224: `GetPrimeTask(SpamEngine)` → `_primeTasks.TryGetValue` (254). + - P, running C: `CompletePrime(A, "Spam")` → status is Faulted (343) → **`TryRemove` (348)** → `GetBaseException` (350-352) → `BuildPrimeFailedMessage` (`string.Format` with `CultureInfo.CurrentCulture`, 393-401) → **`_logError` (354)** → harness `Errors.Add` (test line 415). + +Interleavings: + +| Label | Order | `GetPrimeTask` returns | Outcome | +|---|---|---|---| +| I-1 | T's `TryGetValue` before P's `TryRemove` | C | `await C` resumes after `CompletePrime` returns; `Errors` populated. PASS. | +| I-2 | P finishes `Errors.Add` before T's `TryGetValue` | `Task.CompletedTask` | Assertions read a list that already has one entry. PASS in practice; note that no synchronization edge orders `Errors.Add` (after `TryRemove`) before T's read, so even this passing case has no formal happens-before. | +| **I-3** | P's `TryRemove` before T's `TryGetValue`, and T's `Errors` read before P's `Errors.Add` is complete or visible | `Task.CompletedTask` | `await` returns synchronously; `harness.Errors.Should().ContainSingle()` (227-231) fails with an empty collection, or, if T reads while `List.Add` is mid-write, `harness.Errors[0]` can be null and line 232 throws `NullReferenceException`. **FAIL.** | + +Why I-3 is rare but real: the window on P between line 348 and the completed `Errors.Add` is ordinarily microseconds, but on the first execution in the process it includes JIT compilation of `BuildPrimeFailedMessage`, the `_logError` lambda, `LoggedError..ctor` and `List.Add`, plus culture-data initialisation for `CultureInfo.CurrentCulture` `[D]`. Under `Workers=0` with other classes running on the same runner, T can be preempted after `SetException` and P can be preempted inside that window. One failure on PR #939 head `9624376dc` with passes on rerun and locally (issue.md:36) is consistent with this. + +### Cancellation path + +`CompletePrime` is shared by the Faulted and Canceled outcomes (343 tests only `RanToCompletion`), so the same window exists after `probe.SetCanceled()`. The two cancel tests are not exposed: `..._LogsErrorAndClearsPrimeMarker` captures the handle before the trigger (Race.cs:211-215); `..._LeavesToggleReportingUnchecked` asserts only `GetPressed == false` (Race.cs:266-269). + +## 5. Complete list of ordering hazards found + +| # | Hazard | Where | Effect | In scope for #942 | +|---|---|---|---|---| +| A | Marker cleared (348) before the fault is reported (354). | `CompletePrime` | A `GetPrimeTask` call made after the trigger can return `Task.CompletedTask` while the report is pending; violates the documented "await the prime deterministically" contract (238-246) and is the direct cause of the CI failure. | **Yes — root cause.** | +| B | Registration (276) races removal (348) when `ApplyPrimeAsync` completes synchronously in a non-success state. | `StartPrimeIfNeeded` / `CompletePrime` | If `TryRemove` runs before the assignment lands, the assignment re-registers an already-completed continuation and the key never re-primes for the session (the CR-2 defect class). Reachable in this very test at line 237: Moq returns the same already-faulted `probe.Task`, the awaiter reports `IsCompleted`, the state machine throws synchronously, A2 is faulted before `ContinueWith` is called, and C2 is queued before line 276 executes. No assertion follows line 242, so the test cannot fail from it, but the cleanup `await` at 242 can return before the second `_logError`, leaving a stray `Errors.Add` on a pool thread after the test method returns (benign: the harness is per-test). Reachable in production: `AppItemEngines.EngineActiveAsync` (`TaskMaster/AppGlobals/AppItemEngines.cs:101-109`) awaits `Globals.AF.Manager.Configuration`, an `AsyncLazy` backed by `Lazy>` (`UtilitiesCS/ReusableTypeClasses/AsyncLazy/AsyncLazy.cs:24,32`) that caches a faulted task permanently (`UtilitiesCS.Test/ReusableTypeClasses/AsyncLazy_Tests.cs:58-73`), so after one configuration-load fault every later prime faults synchronously and races on each `getPressed` poll. Already documented as NB-2 in `docs/features/active/2026-09-02-ribbon-engine-toggle-defects-735/code-review.2026-09-03T06-19.md:188-221` with "Recommendation for a follow-up issue"; grep of `docs/features/potential/` finds no entry for it. | **No — promote as its own issue.** The recommended #942 reorder neither worsens nor fixes it. | +| C | Test-side: `Harness.Errors` (`List`) is appended on a pool thread and read on the test thread with no synchronization in I-3. | test fixture | Secondary to A; produces the `NullReferenceException` variant of the failure. Eliminated by either fix in section 6 because both restore a happens-before edge between `Errors.Add` and the assertions. | Resolved by the fix to A. | +| D | The identity assertion at Race.cs:240-245 is weaker than its comment states: `Task.CompletedTask` is also "not the same as" `firstPrime`, so it passes even when the second marker was removed (or never re-registered under hazard B) before line 235 ran. | Race.cs test | Not flaky; the conclusion "a second prime actually started" is not fully established by that assertion. Strengthening it to `NotBeSameAs(Task.CompletedTask)` would itself be subject to hazard A, so leave it. | Observation only. | + +## 6. Candidate approaches + +### Approach 1 — Reorder `CompletePrime`: report the fault, then clear the marker (production) + +Change lines 348-354 so the order is: compute `failure` → `_logError(...)` → `_primeTasks.TryRemove(engineName, out _)`. Update the `` at 327-331 (which lists "marker cleared … and the failure reported" in the current prose order) and add a one-sentence "why" comment: the report precedes the removal so that any holder of the prime handle, and anyone who observes the marker gone, is guaranteed the fault has already been reported. + +Why it fixes the test with no test change: in I-1 the await still covers the report; in I-2/I-3 the only way `GetPrimeTask` returns `Task.CompletedTask` is by observing the removal, and `Errors.Add` now precedes the removal in P's program order. `ConcurrentDictionary.TryRemove` publishes the bucket change under a lock with a volatile write and `TryGetValue` reads the bucket with a volatile read `[D]`, so observing the removal acquires everything P wrote before it, including the list append. Hazard C disappears with it. + +Invariant check (all `[V]`): +- Lines 327-331 describe the order in prose only; no sentence names it as load-bearing. +- The #735 acceptance wording is order-agnostic: plan P3-T9 requires only that "the marker removal and the log call are on the non-completed path" (`plan.2026-09-02T12-04.md:265-266`); `spec.md:172` likewise. +- No production consumer depends on the order: the sink is `logger.Error` (does not re-enter the coordinator); `GetPrimeTask` has no production callers. +- "Does a fault re-prime rely on the marker being cleared before observers see the fault?" No. A re-prime is triggered only by a later `GetPressed` poll from Office, which is independent of the sink and of any awaiter. With the reorder, a poll arriving between the report and the removal sees the marker still present and does not re-prime on that poll; the next poll re-primes. Before the reorder the opposite window existed (a re-prime could start, and even log, before the first fault's report), so log order could invert. Neither window is asserted by any test. +- The type's "exactly one `catch`" invariant (153-155, 284-285) is untouched. +- A throwing sink: today a throwing `_logError` faults the continuation (breaking the "never faults" contract at 288) with the marker already cleared; after the reorder it would additionally leave the marker registered. The production sink is log4net's `logger.Error`, which does not throw on appender failure. If unconditional marker clearing is wanted, `try { _logError(...); } finally { _primeTasks.TryRemove(...); }` preserves it without adding a `catch`; this is optional hardening, not required for #942. + +Alignment: minimal (two statements plus comments), keeps the fixture unchanged, and makes the `GetPrimeTask` doc contract true. Production changes: yes, one method. + +### Approach 2 — Capture the handle before the trigger (test-only) + +In the flaky test, replace lines 220-224 with: `GetPressed(SpamEngine); var prime = GetPrimeTask(SpamEngine); probe.SetException(failure); await prime;`. Precedent: Race.cs:211-215. Deterministic under the current production order because the captured handle is C regardless of when P runs. + +Limitations: leaves the documented `GetPrimeTask` contract false on the failure path, so any future post-trigger call (the cleanup awaits at :242 and Race.cs:272 already are such calls) can still return early; cannot serve as a "fails before, passes after" regression test because it changes the observation, not the behaviour; the issue's proposed validation ("show that the test fails when the ordering is inverted", issue.md:57) presupposes an ordering in the code under test. Production changes: none. + +### Recommendation + +**Approach 1**, with a new deterministic regression test (section 8) whose red state before the reorder is the negative control. Approach 2 may additionally be applied to the original test as hardening; it is not required, and leaving the original test untouched preserves it as the "original repro" the bugfix workflow asks to re-run. + +### Rejected alternatives (brief) + +- `TaskContinuationOptions.ExecuteSynchronously` on the continuation: with an already-completed antecedent it runs inline *inside* `ContinueWith`, i.e. before line 276, which makes hazard B deterministic instead of racy; it also does not restore the contract when the antecedent completes later on another thread. +- Taking `_primeGate` inside `CompletePrime`: addresses hazard B, not A; out of scope. +- Injecting a `TaskScheduler` seam so tests can run the continuation on demand: adds a fifth constructor argument and a production seam that nothing else needs; the in-sink probe (section 8) obtains the same determinism without it. +- `[DoNotParallelize]`, `Workers=1`, retries, sleeps, `Task.Delay`: prohibited by the issue and by `.claude/rules/csharp.md` "Prohibited Behaviors". + +## 7. Behaviour semantics after the fix + +- On any non-success prime outcome the coordinator: leaves the cache unset; reports the failure through `logError` exactly once; then clears the in-flight marker so a later read may re-prime. +- `GetPrimeTask(key)` returns a task that is incomplete for as long as the marker is registered, and the marker is registered for as long as the fault report has not yet been delivered. Corollary: a caller that obtains `Task.CompletedTask` for a key whose prime failed can rely on the report having already been delivered. +- Success path, toggle path, cancellation synthesis (350-352), message text, and the "never faults" nature of the handle are unchanged. + +## 8. Testing implications + +### 8.1 Regression test (deterministic, no timing) + +Discriminator: the state of the prime marker **at the moment the sink is invoked**. That moment is strictly after line 348 in the current code and strictly before the removal in the fixed code, on the same thread, so the outcome is a function of program order, not of scheduling. + +Design (MSTest, Moq strict mock, FluentAssertions; Arrange-Act-Assert): + +- Harness change (main fixture file): add `internal Action OnLogError { get; set; }` and invoke it from the `logError` lambda after `Errors.Add`, mirroring `OnInvalidate` (409-413, 430-434). About eight lines; the file goes from 459 to roughly 467 lines. +- New test, suggested name `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged`: + - Arrange: harness; held `probe`; strict `EngineActiveAsync(SpamEngine)` returns `probe.Task`; `GetPressed(SpamEngine)`; `var prime = GetPrimeTask(SpamEngine)` (captured before the trigger so the await itself is deterministic — Race.cs:211 precedent); `Task handleSeenBySink = null; harness.OnLogError = (_, _) => handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine);` (probe from inside the sink — the OnInvalidate precedent at 252-276). + - Act: `probe.SetException(failure); await prime;` + - Assert: `handleSeenBySink.Should().BeSameAs(prime, "while the fault is being reported the prime handle must still be registered, so a caller that fetches it after the trigger awaits the report")`; `harness.Errors.Should().ContainSingle()`; `harness.Errors[0].Exception.Should().BeSameAs(failure)`; `harness.Coordinator.GetPrimeTask(SpamEngine).Should().BeSameAs(Task.CompletedTask, "once the handle has completed the marker has been cleared so a later read may re-prime")`. + - Thread safety: `handleSeenBySink` is written on P before C completes and read on T after `await prime`; task completion provides the happens-before edge. No blocking, no gate, no timer, no `Thread.Sleep`/`Task.Delay`, no temp file, no `[DoNotParallelize]`. + - Before the reorder the sink runs after `TryRemove`, so `GetPrimeTask` returns `Task.CompletedTask` and the first assertion fails deterministically. After the reorder it returns C (`== prime`) and passes. +- A cancellation twin (`probe.SetCanceled()`, `BeAssignableTo`) is optional; `CompletePrime` is a single path for both outcomes, so one test covers the changed lines. + +File placement: `EngineToggleStateCoordinatorTests.cs` is at 459/500 lines; put the new test in a third partial (for example `EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`) or append a new region to `Race.cs` (277 lines). The test project uses explicit `` items (`TaskMaster.Test/TaskMaster.Test.csproj:352, 359`), so a new file needs a csproj entry. + +### 8.2 Negative control + +Run the new test against the current statement order (before applying the reorder, per the bugfix workflow's failing-test-first step): it fails on the `BeSameAs(prime)` assertion every time. Then apply the reorder and rerun: it passes. Record both as TRX-derived summaries under `/evidence/regression-testing/` (raw TRX is not committable per CLAUDE.md "Committed Test Evidence Format"). This is stronger than the original test, which cannot serve as a negative control because its failure depends on scheduling. An equivalent second demonstration is to temporarily invert the two statements after the fix and observe the same deterministic failure. + +### 8.3 Existing tests + +- `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` becomes deterministic without modification under Approach 1 (section 6 reasoning). Optionally harden it with the capture-before-trigger form. +- No existing assertion depends on the current order (section 6 invariant check). +- Coverage: the reorder changes no branch structure; the changed lines are exercised by the existing fault test, the two cancel tests, and the new test. + +### 8.4 Alternative regression design considered and not recommended + +Gating the injected `logError` on a `ManualResetEventSlim`/`TaskCompletionSource` wait so the test thread can call `GetPrimeTask` while the sink is held open also discriminates the two orders, but it blocks a pool thread on a test-controlled gate. Under `Workers=0` saturation the test thread's own resumption may need a pool thread, and a blocking wait is adjacent to the banned "real wall-clock waits". The in-sink probe gives the same discrimination without blocking. + +## 9. Files that would change (repository-relative) + +| File | Change | +|---|---| +| `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` | `CompletePrime`: move `_primeTasks.TryRemove` after `_logError`; update `` at 327-331 and add the why-comment; optionally tighten the `GetPrimeTask` `` (243-246) to state the guarantee. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` | `Harness`: add `OnLogError` hook invoked from the `logError` lambda. Optional: capture-before-trigger hardening at 220-224. | +| `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (new partial) — or a new region in `EngineToggleStateCoordinatorTests.Race.cs` | The regression test in 8.1. | +| `TaskMaster.Test/TaskMaster.Test.csproj` | `` for the new partial (only if a new file is created). | +| `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` | Fill Root Cause (replace the H1 wording with H2), Proposed Fix, Test Strategy, ACs. | +| `docs/features/potential/-engine-toggle-prime-marker-registration-race.md` | New potential entry for hazard B (NB-2), promoted through the MCP lifecycle; not part of the #942 code change. | + +## 10. Survey: does any other test have the same pattern? + +Exhaustive grep of `GetPrimeTask` across `*.cs` finds only the two fixture files listed in section 2.2; every post-trigger use is either on the success path (marker never removed), assertion-free, or capture-before-trigger. Production `ContinueWith` sites outside tests (`Grep "\.ContinueWith\("`, excluding `*.Test/**`): `TaskMaster/AppGlobals/AppEvents.ReadinessHookup.cs:46` and `UtilitiesCS/OutlookObjects/Folder/OutlookFolderTreeService.cs:405` discard a fault-logging continuation (`_ = …`, the genuine H1 shape), but no test asserts on those logs (`TaskMaster.Test` grep for "Startup inbox processing failed"/`ProcessStartupInboxItemsAfterReadinessHookup`: none; `UtilitiesCS.Test` grep for `ObserveFault`/`cleanupFailureObserver`: none). `TaskMaster/AppGlobals/AppOlObjects.FolderTreeService.cs:148-162` uses `ExecuteSynchronously` plus an explicit already-completed check, and `QuickFiler/Controllers/QfcFormController.SetupDisposal.cs:240` stores its continuation as the awaited handle; neither removes a marker before logging. Conclusion: no other test in the repository currently carries the #942 pattern. + +## 11. Memory hygiene + +No absolute host paths are embedded in this artifact. No evidence artifacts were produced by this research session. diff --git a/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md new file mode 100644 index 000000000..4c6e0a7e0 --- /dev/null +++ b/docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md @@ -0,0 +1,256 @@ +# 2026-09-29-engine-toggle-prime-fault-logging-test-races (Spec) + +- **Issue:** #942 +- **Parent (optional):** none +- **Owner:** drmoisan +- **Last Updated:** 2026-09-30T01-30 +- **Status:** Ready for planning +- **Version:** 0.3 + +> Formatting note for later editors: inline code spans around repository paths in this document are the change footprint. Only files listed under "Write Set" in the Scope section, and the evidence projections named in the Test Strategy section (which sit under that same feature folder), are backticked. Out-of-scope files are cited in plain prose on purpose; do not add backticks to them. Acceptance-criterion lines deliberately contain no digits, no angle brackets, and no percent signs; counts are written as words. + +## Context + +`EngineToggleStateCoordinatorTests.GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` failed once in CI and passed on rerun. The fault logging it asserts appeared to race the task the test awaits. The research record under this feature folder's research directory established the actual mechanism: it is not an unawaited continuation but a statement-ordering defect inside the coordinator, which lets a post-trigger call to `GetPrimeTask` return an already-completed task while the fault report is still pending on a thread-pool thread. + +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 + +Impact / Severity: +- [ ] Blocker +- [x] High +- [ ] Medium +- [ ] Low + +The severity is High because the failure lands on a required CI check and blocks unrelated pull requests (the first observed failure was on PR #939, which does not touch this code). + +## Repro & Evidence + +Steps to Reproduce: +1. Run the TaskMaster.Test project under the parallel regime configured by the repository run settings (Workers set to zero, class-level scope). +2. Observe `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` (the test begins at line 213). It fails intermittently. + +Expected: +The test is deterministic: the error log it asserts is written before the awaited task completes, so any caller that obtains the prime handle after the trigger still awaits the report. + +Actual: +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. The failure is timing-dependent and cannot be reproduced on demand with the existing test; the research record (section 4) gives the exact interleaving and explains why the window is ordinarily microseconds wide but is widened on first execution by JIT compilation and culture-data initialisation inside the reporting path. + +Logs / Screenshots: +- [ ] Attached minimal logs or screenshot +- Snippet: CI run for PR #939 at head `9624376dc` (first attempt). +- Research record: the file under this feature folder's research directory dated 2026-09-29T23-20 (read in full; its verified line citations were re-checked against this worktree while authoring this spec). + +## Scope & Non-Goals + +- In scope: + - Reorder two statements in `CompletePrime` in `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` so the fault is reported through the injected error-log delegate before the in-flight marker is removed, and update the method's documentation to describe and justify that order. + - Add an `OnLogError` observer hook to the private `Harness` fixture in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, mirroring the existing `OnInvalidate` hook. + - Add one deterministic regression test in a new third partial, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`, registered with an explicit compile item in `TaskMaster.Test/TaskMaster.Test.csproj`. + - Capture fail-before and pass-after evidence projections, a coverage baseline and post-change comparison, and the final toolchain pass under this feature folder's evidence tree. +- Out of scope / non-goals: + - Hazard B from the research record (registration of the prime marker racing its removal when the prime completes synchronously in a non-success state; recorded as NB-2 in the issue #735 code review). It is promoted separately as its own issue. This fix neither worsens nor addresses it, and no lock is added to CompletePrime. + - Any change to the existing test `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse`. It is left unmodified so it remains the original reproduction the bugfix workflow asks to re-run. + - Any change to the second existing partial, EngineToggleStateCoordinatorTests.Race.cs, or to the production wiring in RibbonController.EngineCommands.cs. + - Any change to the parallel run settings (TaskMaster.runsettings, scripts/vscode/TaskMaster.cli.runsettings). + - Retries, sleeps, `Task.Delay`, `Thread.Sleep`, `[DoNotParallelize]`, a single-worker setting, timeouts, blocking gates inside the sink, temporary files, or an injected scheduler seam. Each is either prohibited by repository policy or rejected in the research record with a stated reason. + - The optional try-finally hardening around the log call discussed in the research record (section 6). The production sink is log4net's error method, which does not throw on appender failure, so the hardening is not required for this issue. + - The two other production sites that discard fault-observing continuations (AppEvents.ReadinessHookup.cs and OutlookFolderTreeService.cs). No test asserts on their logs, so they are not part of this defect. +- Explicitly excluded systems, integrations, or datasets: Outlook host process, Office ribbon callbacks, log4net configuration. + +### Write Set + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` (new) +- `TaskMaster.Test/TaskMaster.Test.csproj` +- `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/spec.md` +- Evidence projections under `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/` as named in the Test Strategy section + +## Root Cause Analysis + +### Confirmed mechanism + +`CompletePrime` in `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` (lines 341 to 355 in this worktree) runs as a `ContinueWith` continuation on `TaskScheduler.Default` with `TaskContinuationOptions.None`. On any outcome other than ran-to-completion it executes, in this order: + +1. `_primeTasks.TryRemove(engineName, out _)` (line 348) — clears the in-flight marker. +2. Unwraps or synthesizes the failure exception (lines 350 to 352). +3. `_logError(BuildPrimeFailedMessage(engineName), failure)` (line 354) — reports the fault. + +`GetPrimeTask` (lines 247 to 255) returns the stored continuation when the marker is present and `Task.CompletedTask` when it is absent. The flaky test fetches the handle after the trigger: it calls `probe.SetException(failure)` at line 223 and only then calls `GetPrimeTask` at line 224. + +Because `SetException` runs the awaiting state machine inline on the test thread and the state machine then queues the continuation to the thread pool, the test thread and a pool thread run concurrently from that point. If the pool thread reaches step 1 before the test thread's `TryGetValue`, the test awaits `Task.CompletedTask`, returns synchronously, and asserts on `harness.Errors` before (or while) the pool thread performs step 3. The assertion `Errors.Should().ContainSingle()` then fails on an empty list, or, if the read overlaps the list append, the indexer returns null and the message assertion throws. That is the CI failure. + +### Contract violated + +The documentation on `GetPrimeTask` (lines 238 to 246) states that the handle is "exposed so tests can await the prime deterministically instead of polling or sleeping" and that a prime fault "is observed inside the prime itself and reported through logError". The second sentence is true only for a caller that already holds the continuation. A caller that observes the marker absent has no guarantee the report has happened, because removal precedes the report in program order. The invariant the fix restores is stated in one sentence: **for a key whose prime did not run to completion, the in-flight marker is present until the fault report has returned, so any caller that observes the marker absent observes a report that has already completed.** + +### Original hypothesis refuted + +The issue's suspected cause ("the prime fault is probably observed and logged in a continuation that is not part of the awaited task") is refuted. `StartObservedPrime` (lines 290 to 303) returns the `ContinueWith` continuation itself, `StartPrimeIfNeeded` stores exactly that at line 276, and `GetPrimeTask` returns the stored value. The fault observer is inside the awaited task. The defect is the order of two statements inside that observer, not a missing await. + +### Why the window is real + +The window between removal and the completed list append is ordinarily microseconds, but on the first execution in a process it includes JIT compilation of the message builder, the injected log lambda, the `LoggedError` constructor and the list append, plus culture-data initialisation for the current-culture string format. Under the parallel run regime with other classes on the same runner, the test thread can be preempted after `SetException` and the pool thread can be preempted inside that window. One failure with passes on rerun and locally is consistent with this. + +### Why this cannot be made deterministic from the test side alone + +Capturing the handle before the trigger (the pattern already used by `GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker` in the Race partial) makes that one test immune, but it leaves the documented contract false on the failure path: every other post-trigger call, including the cleanup awaits already present at line 242 and in the Race partial, can still return early. It also cannot serve as a fails-before, passes-after regression test, because it changes the observation rather than the behavior. + +## Proposed Fix + +### Design summary (what changes where): + +1. **Production, `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`, `CompletePrime`.** Move `_primeTasks.TryRemove(engineName, out _)` to after the `_logError(...)` call, so the order becomes: early return on ran-to-completion; compute `failure`; report through `_logError`; remove the marker. Update the `summary` element so its prose lists the report before the marker removal, and add a one-sentence comment stating why the order is load-bearing: the report precedes the removal so that any holder of the prime handle, and any caller that observes the marker gone, is guaranteed the fault has already been reported. Tighten the `returns` element of `GetPrimeTask` with one sentence stating that guarantee. No other member changes. +2. **Test fixture, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, `Harness`.** Add an internal settable `OnLogError` property typed as an action taking the message string and the exception, documented as an optional extra observer invoked from inside the error-log sink. Invoke it from the injected `logError` lambda immediately after the `Errors.Add(...)` call, using the null-conditional invoke form already used for `OnInvalidate`. No existing test method in the file changes. +3. **New regression test, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`.** A third partial of the same `[TestClass]` (the class attribute stays on the first partial only, as the Race partial already does), containing `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged`. Design in the Test Strategy section. +4. **Registration, `TaskMaster.Test/TaskMaster.Test.csproj`.** Add a `Compile Include` item for the new file next to the two existing coordinator test entries (lines 352 and 359). The project uses explicit compile items; without the entry the file is not compiled and the test silently does not exist. + +### Boundaries and invariants to preserve: + +- The type keeps exactly one `catch` (the click boundary). `CompletePrime` remains a continuation, not a `catch`, and gains no `try` block. +- The success path is untouched: on ran-to-completion `CompletePrime` still returns before touching the marker or the sink, so the handle for a successful prime stays registered and `GetPrimeTask` keeps returning it. +- The cancellation path still synthesizes a `TaskCanceledException` when the completed task carries no exception, and the faulted path still reports the unwrapped base exception. +- The returned continuation still always completes successfully; the message text and the invalidation behavior are unchanged. +- `StartPrimeIfNeeded`, `_primeGate`, and the at-most-one-prime guard are not modified. No lock is added to `CompletePrime` (that would address hazard B, which is out of scope). +- No new constructor parameter, interface, scheduler seam, or public surface is introduced. + +### Dependencies or blocked work: + +- None. `GetPrimeTask` has no production caller (verified by a repository-wide search over C# sources: the only non-test hits are its declaration and a `see cref` in the `_primeTasks` documentation). The production error sink is a log4net error call that does not re-enter the coordinator. +- Hazard B is promoted as a separate issue and is not a dependency of this fix. + +### Implementation strategy (what changes, not sequencing): + +#### Files/modules to change: + +- `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs` — `CompletePrime` body and its documentation; `GetPrimeTask` `returns` documentation. +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` — `Harness.OnLogError` and its invocation from the log lambda. +- `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` — new partial with the regression test. +- `TaskMaster.Test/TaskMaster.Test.csproj` — one new compile item. + +#### Functions/classes/CLI commands impacted: + +- `EngineToggleStateCoordinator.CompletePrime` (private): statement order and documentation. +- `EngineToggleStateCoordinator.GetPrimeTask` (internal): documentation only; body unchanged. +- `EngineToggleStateCoordinatorTests.Harness` (private nested): one new property; constructor lambda gains one invoke. +- `EngineToggleStateCoordinatorTests` (partial test class): one new test method in a new partial. + +#### Data flow and validation changes: + +- None in data flow. The only semantic change is the order in which the sink observes the fault relative to the marker removal. +- After the change, a `getPressed` poll that arrives between the report and the removal sees the marker still present and does not re-prime on that poll; the next poll re-primes. Before the change the opposite window existed (a re-prime could start, and even log, before the first fault's report), so log order could invert. Neither window is asserted by any test, and the new window is bounded by the duration of the log call. + +#### Error handling and logging updates: + +- No new log messages. The existing prime-failed message is emitted exactly once per failed prime, as before, but now before the marker is cleared. +- A throwing error sink would now leave the marker registered as well as faulting the continuation (before the change it only faulted the continuation). The production sink does not throw; the optional `finally` hardening is recorded as a non-goal. + +#### Rollback/feature-flag considerations (if applicable): + +- Not applicable. The change is two reordered statements and can be reverted by a single commit revert. No flag. + +### Technical specifications (interfaces/contracts): + +#### Inputs/outputs and formats: + +- `GetPrimeTask(engineName)` — unchanged signature. Post-condition strengthened: when it returns `Task.CompletedTask` for a key whose prime did not run to completion, the fault report for that prime has already returned. +- `Harness.OnLogError` (test-only) — an optional action receiving the message and the exception, invoked from inside the log sink after the error has been appended to `Errors`, on whatever thread the sink runs. + +#### Required configuration keys and defaults: + +- None. + +#### Backward-compatibility expectations: + +- No public API change. `CompletePrime` and `Harness` are private; `GetPrimeTask` is internal and its signature is unchanged. The #735 acceptance wording ("the marker removal and the log call are on the non-completed path") remains satisfied because both statements stay on that path. + +#### Performance constraints (latency/throughput/memory): + +- No measurable change. The marker is held for the additional duration of one log call on the failure path only. + +## Assumptions, Constraints, Dependencies + +- Assumptions (environment, data, access): .NET Framework TPL semantics as documented — `TaskCompletionSource.SetException` runs await continuations inline on the completing thread; a `ContinueWith` registered with `TaskContinuationOptions.None` is queued to the default scheduler; `ConcurrentDictionary.TryRemove` publishes under a lock with volatile writes and `TryGetValue` performs a volatile read, so observing the removal acquires all writes the removing thread made before it (including the list append performed by the sink). The MSTest version pinned by the test project's packages.config runs test bodies without a synchronization context. +- Constraints (budget, performance, compatibility): The main fixture file is at 459 of the 500-line ceiling in this worktree; the harness hook adds roughly eight lines. The new test goes in a separate partial for that reason. All C# work must pass the toolchain in CLAUDE.md order. Tests must use MSTest, Moq, and FluentAssertions, and must not add sleeps, delays, retries, timeouts, blocking waits, `[DoNotParallelize]`, or temporary files. +- External dependencies (services, libraries, releases): none beyond packages already referenced by the test project. + +## Data / API / Config Impact + +- User-facing or API changes: none. +- Data or migration considerations: none. +- Logging/telemetry updates (if any): none in content; ordering of the existing prime-failed log relative to internal marker removal changes as described. +- Compatibility notes (CLI flags, config schemas, versioning): none. + +## Test Strategy + +Seeded from issue: + +- [x] 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. — Resolved by the in-sink probe design below; a scheduler seam and a blocking gate were both considered and rejected (research record, sections 6 and 8.4). +- [x] Negative control: show that the test fails when the ordering is inverted. — Resolved by the fail-before run described below; the ordering under test is the statement order inside `CompletePrime`. + +- Regression tests to add or update: + - Add `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged` in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs`. Discriminator: the state of the prime marker at the moment the sink is invoked. That moment is after the removal in the current code and before the removal in the fixed code, on the same thread, so the outcome is a function of program order, not scheduling. + - Arrange: `new Harness()`; a held `TaskCompletionSource` named `probe`; `InvalidOperationException` named `failure`; strict mock setup `EngineActiveAsync(SpamEngine)` returning `probe.Task`; `harness.Coordinator.GetPressed(SpamEngine)` to start the prime; `var prime = harness.Coordinator.GetPrimeTask(SpamEngine)` captured before the trigger so the await itself is deterministic; a `Task` local `handleSeenBySink` initialised to null; `harness.OnLogError = (_, _) => handleSeenBySink = harness.Coordinator.GetPrimeTask(SpamEngine)`. + - Act: `probe.SetException(failure)`; `await prime`. + - Assert: `handleSeenBySink.Should().BeSameAs(prime, ...)` with a reason stating that while the fault is being reported the prime handle must still be registered so that a caller fetching it after the trigger awaits the report; `harness.Errors.Should().ContainSingle()`; `harness.Errors[0].Message.Should().Contain(SpamEngine)`; `harness.Errors[0].Exception.Should().BeSameAs(failure)`; `harness.Invalidations.Should().BeEmpty()`; `harness.Coordinator.GetPrimeTask(SpamEngine).Should().BeSameAs(Task.CompletedTask, ...)` with a reason stating that once the handle has completed the marker has been cleared so a later read may re-prime. + - Thread safety of the test itself: `handleSeenBySink` is written on the pool thread before the continuation completes and read on the test thread after `await prime`; task completion supplies the happens-before edge. No blocking, gate, timer, sleep, delay, temporary file, or parallelism attribute. + - Documentation: an XML summary on the test naming issue #942 and stating the invariant, plus an in-file comment stating that if this test passes without the production reorder, isolation of the negative control has been lost and the run must be investigated rather than accepted. + - Modify the `Harness` in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` to expose `OnLogError`; no existing test method is edited. + - Leave `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` unchanged. Under the fixed order it is deterministic: if it obtains the continuation it awaits the report; if it obtains `Task.CompletedTask` it has observed a removal that the report preceded in program order, and the volatile read acquires the list append. +- Unit tests (pytest) for the fixed behavior and boundaries: not applicable (C# / MSTest). The new test, the existing fault test, and the two existing cancellation tests in the Race partial together exercise every statement of the changed method on both the faulted and canceled outcomes. +- Edge cases and negative scenarios (invalid inputs, missing data, boundary values): + - Cancellation outcome: covered by the existing `GetPressed_WhenPrimeIsCanceled_LogsErrorAndClearsPrimeMarker` and `GetPressed_WhenPrimeIsCanceled_LeavesToggleReportingUnchecked`, which pass unchanged because `CompletePrime` is a single path for both outcomes. A cancellation twin of the new test is optional and not required. + - Success outcome: covered by the existing prime-success tests; the early return is untouched. + - Null or empty engine key on `GetPrimeTask`: body unchanged; existing constructor and key tests are unaffected. +- Error handling and logging verification: the new test asserts exactly one logged error carrying the injected exception and a message containing the engine name, with no invalidation. The existing fault test asserts the same on the original observation path. +- Coverage impact and targets for changed lines/modules: the reorder changes no branch structure. The changed lines in `CompletePrime` are exercised by the new test, the existing fault test, and the two cancellation tests, so line coverage of the changed lines must not decrease relative to the baseline and the method is expected to remain fully covered. Test files are excluded from the coverage denominator. Record a baseline projection before the change and a post-change projection after it, with an explicit comparison of the coordinator file's line and branch figures: + - `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/baseline/coverage-baseline.md` + - `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/coverage-post-change.md` (includes the comparison against the baseline for the coordinator file and the first-party total) +- Toolchain commands to run (format → lint → type-check → test), exactly as in CLAUDE.md, restarting from the first step if any step changes files or fails: + 1. `dotnet tool run csharpier format .` then verify with `dotnet tool run csharpier check .` + 2. `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:EnableNETAnalyzers=true /p:EnforceCodeStyleInBuild=true` + 3. `msbuild TaskMaster.sln /t:Rebuild /m /p:Configuration=Debug "/p:Platform=Any CPU" /p:TreatWarningsAsErrors=true` + 4. Run the `test: MSTest with Coverage (Koverage)` VS Code task, or invoke Invoke-MSTestWithCoverage.ps1 under scripts/vscode directly. + The final passing pass is recorded as a projection with Timestamp, Command, and EXIT_CODE fields at `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/qa-gates/toolchain-final-pass.md`. For the two msbuild steps the projection must also state that no project reported a skipped CoreCompile target, so the analyzer and nullable gates are known to have compiled rather than short-circuited. +- Fail-before / pass-after evidence (bugfix workflow, regression test first): + - Fail-before: with the harness hook, the new partial, and its csproj entry in place and the production reorder NOT applied, run the coordinator test class (a vstest invocation filtered to the fully qualified test class name is sufficient). The new test fails on the `BeSameAs(prime)` assertion every time, because the sink runs after the removal and observes `Task.CompletedTask`. Record the projection with a non-zero EXIT_CODE and `ExpectedExitCode` set to that non-zero value at `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-fail-before.md`. The projection must state the environment of the control explicitly: production file unchanged from the merge base (cite the commit), hook present, new test present, run settings unchanged. + - Pass-after: apply the reorder and rerun the same command. The new test and `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` both pass, and every other test in the class passes. Record the projection with EXIT_CODE zero at `docs/features/active/2026-09-29-engine-toggle-prime-fault-logging-test-races-942/evidence/regression-testing/prime-fault-ordering-pass-after.md`, including a statement that the only production difference between the two runs is the statement reorder in `CompletePrime` (the harness hook and the new test are test-side and were present in both runs). + - An equivalent additional demonstration, if wanted, is to temporarily invert the two statements after the fix and observe the same deterministic failure; it does not replace the fail-before run. + - Per CLAUDE.md "Committed Test Evidence Format", only Markdown projections are committed. No raw TRX, Cobertura XML, or coverage file is added to the repository under any path. +- Manual validation steps (if required): none. The defect is not observable from the Outlook UI. + +## Acceptance Criteria + +- [x] AC1 — In `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`, `CompletePrime` invokes the injected error-log delegate before it removes the engine key from the in-flight prime dictionary on every outcome other than ran-to-completion; the early return on ran-to-completion, the base-exception unwrap, and the synthesized `TaskCanceledException` for a canceled prime are unchanged; and no `catch`, `try`, or lock is added to the method. +- [x] AC2 — The `summary` documentation on `CompletePrime` describes the report-then-clear order, a comment adjacent to the two statements states why the order is load-bearing (a caller that observes the marker absent is guaranteed the fault has already been reported), and the `returns` documentation on `GetPrimeTask` states that guarantee in one sentence. +- [x] AC3 — The private `Harness` in `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs` exposes an internal settable `OnLogError` hook typed as an action receiving the message and the exception, documented as an optional observer invoked from inside the error-log sink, and the injected log lambda invokes it with the null-conditional form immediately after appending to `Errors`; no existing test method in that file is modified (the diff for the file contains hunks only inside the `Harness` type). +- [x] AC4 — A new partial `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` contains the test `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged`, which captures the prime handle from `GetPrimeTask` before triggering the fault, assigns `OnLogError` to record `GetPrimeTask` from inside the sink, faults the held completion source, awaits the captured handle, and asserts that the handle seen by the sink is the same instance as the captured handle, that `Errors` contains exactly one entry whose exception is the injected failure and whose message contains the engine name, that `Invalidations` is empty, and that `GetPrimeTask` after the await returns `Task.CompletedTask`. +- [x] AC5 — The new test uses MSTest attributes, a strict Moq mock of the engines interface, and FluentAssertions with reason strings; it is organised as Arrange, Act, Assert; it carries an XML summary naming the issue and the invariant; and it carries an in-file comment stating that a pass without the production reorder means the negative control has lost isolation. +- [x] AC6 — `TaskMaster.Test/TaskMaster.Test.csproj` contains an explicit `Compile Include` item for the new partial, and the pass-after projection lists `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged` as an executed, passed test (the test is not silently absent from the run). +- [x] AC7 — Fail-before evidence exists as the projection prime-fault-ordering-fail-before.md under the feature folder's regression-testing evidence directory (full path in Test Strategy), recording a run in which the harness hook, the new partial, and its csproj entry are present and the production reorder is absent; the projection carries Timestamp, Command, a non-zero EXIT_CODE, a matching `ExpectedExitCode`, the merge-base commit of the unchanged production file, and shows the new test failing on the same-instance assertion of the sink-observed handle. +- [x] AC8 — Pass-after evidence exists as the projection prime-fault-ordering-pass-after.md under the same directory, recording the same command after the reorder with EXIT_CODE zero, listing both `GetPressed_WhenPrimeFaults_PrimeHandleStaysRegisteredUntilFaultIsLogged` and `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` as passed, and stating that the only production difference between the two runs is the statement reorder in `CompletePrime`. +- [x] AC9 — Every test method in the coordinator fixture across all its partials passes in the pass-after run, and the text of `GetPressed_WhenPrimeFaults_LogsErrorAndStillReturnsFalse` is byte-for-byte unchanged from the merge base. +- [x] AC10 — Neither the new partial nor the `Harness` change introduces `Thread.Sleep`, `Task.Delay`, a retry loop, a wall-clock read, a timeout attribute, `[DoNotParallelize]`, a blocking wait inside the sink, a temporary file, or a scheduler seam; and the repository run-settings files that configure the parallel test regime are not modified. +- [x] AC11 — The final toolchain pass, recorded in the projection toolchain-final-pass.md under the feature folder's qa-gates evidence directory, shows `dotnet tool run csharpier check .` reporting no differences, both CLAUDE.md msbuild rebuild commands (analyzers enabled with code style enforced; warnings treated as errors) exiting zero with no project reporting a skipped CoreCompile target, and the MSTest-with-coverage run exiting zero, all in one uninterrupted pass in the CLAUDE.md order. +- [x] AC12 — Coverage evidence exists as the baseline projection coverage-baseline.md under the feature folder's baseline evidence directory and the post-change projection coverage-post-change.md under its qa-gates evidence directory; the post-change projection compares the coordinator file's line and branch figures against the baseline and shows that coverage of the changed lines in `CompletePrime` did not decrease and that the method remains fully covered; and the diff adds no raw test-result or coverage document (no file with a trx, xml, or coverage extension) anywhere in the repository. +- [x] AC13 — The diff against the merge base modifies no repository file outside the four code files listed in the Write Set, this feature folder, and the promotion record `docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md` written by the promotion commit this branch inherits; in particular the second existing test partial, the ribbon controller engine-command wiring, and the run-settings files are untouched, and `StartPrimeIfNeeded` and the prime gate lock are unchanged (hazard B remains out of scope). +- [x] AC14 — Each of `TaskMaster/Ribbon/EngineToggleStateCoordinator.cs`, `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.cs`, and `TaskMaster.Test/Ribbon/EngineToggleStateCoordinatorTests.PrimeFaultOrdering.cs` is at or below the five-hundred-line ceiling after the change. + +## Risks & Mitigations + +- Technical or operational risks: + - A throwing error sink would leave the marker registered (before the change it only faulted the continuation). The production sink is log4net's error method, which does not throw on appender failure. Mitigation: documented as a non-goal; the optional `finally` hardening can be added in a follow-up if a throwing sink is ever introduced. + - A `getPressed` poll arriving during the log call now observes the marker present and does not re-prime on that poll. Mitigation: the next poll re-primes; the window is bounded by one log call and no test or production behavior depends on re-priming within it. + - Hazard B (registration racing removal on synchronous non-success completion) remains. Mitigation: promoted separately; this change neither widens nor narrows it, and the planner must not attempt to fold it in. + - The main fixture file is close to the line ceiling. Mitigation: the new test lives in a separate partial; AC14 gates the ceiling. +- Mitigations and rollbacks: single-commit revert restores the previous order; no configuration or data migration is involved. + +## Rollout & Follow-up + +- Release/rollout steps: merge through the normal PR gate; no deployment step beyond the next add-in build. +- Post-fix monitoring or clean-up tasks: watch the required MSTest-with-coverage check on subsequent PRs for any recurrence of a failure in the coordinator test class; none is expected. +- Links: issue #942; PR #939 (first observed failure); issue #735 code review note NB-2 (hazard B); the separately promoted issue for hazard B (number recorded by the coordinator at promotion time). diff --git a/docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md b/docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md new file mode 100644 index 000000000..6e4dee197 --- /dev/null +++ b/docs/features/potential/promoted/2026-09-29-engine-toggle-prime-fault-logging-test-races.md @@ -0,0 +1,60 @@ +# engine-toggle-prime-fault-logging-test-races (Issue #942) + +- Date captured: 2026-09-29 +- Author: Dan Moisan +- Status: Promoted -> docs/features/active/engine-toggle-prime-fault-logging-test-races/ (Issue #942) + +> Automation note: Keep the section headings below unchanged; the promotion tooling maps each of them into the GitHub bug issue template. + +- Issue: #942 +- Issue URL: https://github.com/drmoisan/TaskMaster/issues/942 +- Last Updated: 2026-09-30 +## 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 #939 at head `9624376dc` (first attempt). + +## Impact / Severity + +- [ ] Blocker +- [x] High +- [ ] Medium +- [ ] Low + +## 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 that hits a required check. 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. + +## Next Step + +- [x] Promote to GitHub issue (bug-report template) +- [ ] Move to active fix folder / branch