[Chris-Wolfgang]

Why

Stage 1 on the sync PR #667 fails with Wolfgang.Etl.Abstractions.Tests.SourceLink (no coverage data) and Wolfgang.Etl.TestKit.Tests.Concurrency (no coverage data) (run 36947263126). The log shows "Unable to find a datacollector with friendly name 'XPlat Code Coverage'" for both.

Cause and fix, per assembly

Assembly	Cause	Fix	Coverage after (local)
Abstractions.Tests.SourceLink	no coverlet.collector	add collector 8.0.1 (same as the other net10.0-only test projects); rework the network probe (below)	none → 100% (first measured at 75.2%)
TestKit.Tests.Concurrency	no coverlet.collector	add collector 8.0.1; run the scenarios in the normal sweep (below)	none → 100% (first measured at 10.7%)

What the new rows exposed, and the test changes

TestKit.Tests.Concurrency/ConcurrencyTests.cs

• Every test returned immediately unless COYOTE_REWRITTEN=1, so the PR sweep ran no scenario at all. RunSystematic now always runs Coyote's TestingEngine and sets .WithSystematicFuzzingEnabled(!rewritten). Un-rewritten, Coyote fuzzes timing without the false "Deadlock detected" that schedule control reports. Rewritten (the coyote workflow), it controls the schedule as before.
• The race scenarios no longer branch on who wins. A straight-line DrainAsync returns (observed, fault) and the scenario asserts with Specification.Assert (fault is null or OperationCanceledException / ObjectDisposedException).
• Three deterministic facts pin each outcome: never cancelled/disposed, pre-cancelled, pre-disposed.
• Checked locally both ways:
  • un-rewritten: 5/5 pass;
  • after coyote rewrite rewrite.coyote.json with COYOTE_REWRITTEN=1: 5/5 pass, three runs;
  • baseline: the original file also passes when rewritten.

Abstractions.Tests.SourceLink/SourceLinkPdbTests.cs

• Which lines ran depended on what raw.githubusercontent.com answered. The retry and classification logic is now ProbeAsync (injectable status source, TimeSpan pause), Classify and IsAcceptable. New deterministic tests cover 200/403/429/503/400, a persistent 404 (3 attempts), a 404 that clears on retry, HttpRequestException and TaskCanceledException, plus the acceptability table. Behaviour is the same as before:
  • pass on success or infra failures;
  • fail on any other 4xx;
  • fail a persistent 404 only in CI.
• A probe URL can always be built: SourceLink substitutes the commit SHA on every build, local ones included (local PDB: https://raw.githubusercontent.com/Chris-Wolfgang/ETL-Abstractions/bdc6c85.../*). So "no mappings" and "no probe URL" are now assertion failures instead of early returns. The urlPrefix.Contains('*') branch could never be taken and was removed. BuildProbeUrl and ReadSourceLinkPayload are now LINQ queries.

Exclusions removed

None existed in these two assemblies. Tests.Unit still carries class-level [ExcludeFromCodeCoverage] on test doubles. That is out of scope here, because that assembly already reports a row (99.1%).

Verified locally

• dotnet build -c Release: 0 warnings, 0 errors.
• Ran Stage 1's per-project loop (dotnet test per test project per net5.0–net10.0 TFM with --collect:"XPlat Code Coverage" --settings coverlet.runsettings, then dotnet reportgenerator → Summary.txt). Every test assembly that ran has a row, and all 34 project/TFM runs pass. Test rows:
  • DocExamples 100%
  • Fuzz 99.1%
  • SourceLink 100%
  • Abstractions.Tests.Unit 99.1%
  • ErrorPolicies.Tests.Unit 100%
  • TestKit.Tests.Concurrency 100%
  • TestKit.Tests.DocExamples 99%
  • TestKit.Tests.Fuzz 100%
  • TestKit.Tests.Unit 99.7%
  • TestKit.Xunit.Tests.Unit 100%

Unblocks #667; part of #658.

Round 2 (2596884): every test assembly at 100% with no test-code exclusions

Fleet policy: every test assembly at 100%, and [ExcludeFromCodeCoverage] in tests only on a dedicated infrastructure-check method. This repo has no such method.

Exclusions removed / added

• Removed: all 71 [ExcludeFromCodeCoverage] attributes in tests/. They sat on test doubles and helpers in Abstractions.Tests.Unit (BaseClassTests/, ConfigureAwaitContextTests, DisposedGuardTests, FinalizationSuppressionTests, ItemErrorHandlingTests, Performance/AllocationFreeTests, PipelineTests/), ErrorPolicies.Tests.Unit, TestKit.Tests.Unit (mutation tests, CountingProgressTimer, CultureSwapper) and TestKit.Xunit.Tests.Unit. The commit message says 70; that is a miscount, the diff removes 71. grep -rn ExcludeFromCodeCoverage tests now returns nothing.
• Added: none. src/ is untouched, including its existing exclusions.

Test changes (for review)

All of these add tests or make mechanical changes. No existing assertion changes.

• New BaseClassTests/StageRunner.cs. Runs a stage once through its progress overload and returns the items and reports. New *_when_run_produces_no_items_and_a_final_report_of_zero tests use it to run each double whose own tests only construct, configure or dispose it: GuardParity, OptionsConstructor, ConvenienceBaseOptionsConstructor, DisposedGuard, TimerSeam, BaseCoverage DisposeRecording, ItemErrorHandling, FinalizationSuppression. The runs reach the doubles' worker and CreateProgressReport.
• New timer tests. ExtractorBaseTests, LoaderBaseTests and TransformerBaseTests each get one: the internal timer-injection constructor, with a ManualProgressTimer fired mid-run, gives reports [2, 3]. The loader test also reads ListLoader.LoadedItems, which nothing read before.
• New positive-control tests:
  • an undisposed Finalizable* double is finalized (the control for the SuppressFinalize tests);
  • CountingSynchronizationContext counts and runs Post and Send (the control for the "posted nothing" assertions);
  • DualDisposableLoader.Dispose sets its flag when called directly.
• New guard tests. OverloadDoubleCoverageTests calls the never-routed CancelOnly* guards directly and pins the WrongOverloadCalledException message.
• Doubles and helpers pinned by new tests:
  • recording timers' Start / Dispose (TestKit mutation tests, *BaseCoverageTests; new CountingProgressTimerTests);
  • RecordingLogger.BeginScope (ErrorPolicies);
  • the helper iterators: EmptyIgnoring*, CancellingSkipSourceAsync, CancelOnFirst, CancelAt, CancellingSequence through its non-generic enumerator;
  • DisposeTrackingEnumerable non-generic Current / Reset;
  • RetryingExtractor on the progress path;
  • ParsingExtractor / CollectingLoader progress reports;
  • ConfigurableLoader / ConfigurableTransformer Abort rethrow;
  • CounterHarness.ExtractAsync, which throws NotSupportedException.
• Fuzz: a new property, LoaderBase_progress_run_reports_every_loaded_item_in_its_final_report.
• Mechanical changes:
  • GuardParityTests: the never-enumerated EmptyAsync() argument becomes AsyncEnumerable.Empty<int>(), and the helper is removed.
  • FaultyExtractorMutationTests: the local iterator in ExtractAsync_cancelling_during_the_skip_phase_throws_promptly moves to the static CancellingSkipSource(cts). The items and cancel point are the same, and a new test drains it.
  • TestKit.Tests.DocExamples: the per-example rot check moves into CheckExample(...). New tests cover a clean snippet and a snippet that is rot in every context (int value = "text"; → CS0029). That second test reaches failures.Add and best ?? ..., which no test reached before.
  • DelayingExtractorCancellationContractTests: both overrides now share one DrainAsync(items, onItem). It counts each item before the callback and sets canceled on OperationCanceledException, the same semantics as the two loops it replaces. A new test drives a run where cancelAfter is never reached, which covers normal completion.
• Stale "cannot be covered" / "never called" comments on the doubles are updated.

Coverage before -> after (local, Linux-equivalent Stage 1 loop)

BEFORE = this branch at 7ac9d97 with its current coverlet.runsettings, net8.0+net10.0. AFTER = 2596884 with the #667 runsettings (repo-template#702), net5.0–net10.0. A config-only run (7ac9d97 + new runsettings) gave the same numbers as BEFORE, because this repo has no TestModels/*Fixture classes. Removing the exclusions before adding tests took Abstractions.Tests.Unit to 96.8%, ErrorPolicies.Tests.Unit to 96.3%, TestKit.Tests.Unit to 98.7% and TestKit.Xunit.Tests.Unit to 99.3%.

Assembly	Before	After
Abstractions.Tests.DocExamples	100%	100%
Abstractions.Tests.Fuzz	99.1%	100%
Abstractions.Tests.SourceLink	100%	100%
Abstractions.Tests.Unit	99.1%	100%
ErrorPolicies.Tests.Unit	100%	100%
TestKit.Tests.Concurrency	100%	100%
TestKit.Tests.DocExamples	99%	100%
TestKit.Tests.Fuzz	100%	100%
TestKit.Tests.Unit	99.7%	100%
TestKit.Xunit.Tests.Unit	100%	100%
src: Wolfgang.Etl.Abstractions	97.4%	97.4%
src: Wolfgang.Etl.ErrorPolicies	100%	100%
src: Wolfgang.Etl.TestKit	98.4%	98.4%
src: Wolfgang.Etl.TestKit.Xunit	98.3%	98.3%

Wolfgang.Etl.Abstractions.Tests.Concurrency (Coyote CLI, no Test SDK) and AotSmoke are not xunit test projects. Stage 1 does not run them, and they are not measured.

Verified locally

• dotnet build -c Release: 0 warnings, 0 errors (all TFMs).
• All 30 project/TFM runs in the net5.0–net10.0 coverage loop pass.
• The four touched multi-TFM projects also pass on net462, net472, net48, net481 and netcoreapp3.1.
• The finalizer and synchronization-context tests passed 10 of 10 repeated runs (net10.0 and net48).

Merge order

Merge this PR (#668) first, then the config PR #667.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟢 Approval recommended

No blocking issues were identified in the collector additions or test refactors.

Review effort: Balanced
Findings: 1

Open (1)

• Update stale comments about Coyote test execution modes · New

What changed in this PR

Adds coverage collection for SourceLink and TestKit concurrency tests, addressing the missing coverage rows exposed by #667 and supporting #658.

Changes:

• Adds coverlet.collector to both test projects.
• Runs concurrency scenarios without rewriting and adds deterministic outcome tests.
• Refactors SourceLink probing for deterministic retry and status tests.

File	Description
tests/​Wolfgang.Etl.TestKit.Tests.Concurrency/​Wolfgang.Etl.TestKit.Tests.Concurrency.csproj	Enables coverage collection.
tests/​Wolfgang.Etl.TestKit.Tests.Concurrency/​ConcurrencyTests.cs	Enables ordinary-sweep scenarios and tests deterministic outcomes.
tests/​Wolfgang.Etl.Abstractions.Tests.SourceLink/​Wolfgang.Etl.Abstractions.Tests.SourceLink.csproj	Enables coverage collection.
tests/​Wolfgang.Etl.Abstractions.Tests.SourceLink/​SourceLinkPdbTests.cs	Separates probe logic and adds deterministic tests.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.