test: clear all 504 xunit analyzer warnings

[jeremydmiller]

Closes #582

The 504 is really 129

Parsing the build log by unique (rule, file, line, col) shows the 504 warnings come from 129 distinct source locations — the rest is per-TFM (net9.0/net10.0) and per-node duplication. That changed the plan: 129 is small enough to fix every one by hand, so nothing is suppressed wholesale and there is no blanket NoWarn.

Rule	Locations	What was done
xUnit1051	89	TestContext.Current.CancellationToken threaded through
xUnit1013	26	Triaged individually — see below
xUnit1012	8	Theory parameters made nullable
xUnit1031	4	Two sync tests converted to async
xUnit1026	2	One was a real gap — see below

xUnit1013 was worth the hand triage

The issue predicted actual bugs could hide here. Two did:

ProjectionRebuildSelectionTests implemented IProjectionHost on itself. Interface members have to be public, so all four read as un-attributed test methods — the exact self-stub shape that broke all 16 SliceGroupTests in #577, where a self-stubbed IAsyncDisposable.DisposeAsync ran as test lifecycle instead of as a stub. Extracted to a RecordingProjectionHost class, which removes the warnings and the footgun together.

RetryBlockTests.retry_within_threshold was a commented-out //[Fact] — a test that had silently ceased to exist. Now [Fact(Skip = "Unreliable in CI - timing sensitive under load")], so it shows up as skipped on every run instead of vanishing. This is the one intentional baseline change: CoreTests reports 476 total, 475 passed, 1 skipped.

The rest were benign but fixed properly rather than suppressed:

• Documentation samples on test classes (ResourceHostExtensionsTests, UsingSourceWriter) moved to sibling non-test classes. Deliberately not made private — sample_programmatically_control_resources includes its own method signature inside the #region, so reducing visibility would have changed published docs. Moving keeps the snippet text byte-for-byte identical.
• Reflection targets (EventExtensionsTests ×2, event_parameter_naming_convention) made private, with BindingFlags.Instance | BindingFlags.NonPublic added where the lookup needed it. The expression-tree lookups bind private methods fine from inside the class.

xUnit1026 found a test asserting nothing

AggregateVersioningTests.set_version_single_stream_on_internal ran ten InlineData cases and had an empty body after TrySetVersion — no assertion at all. expected was unused because the assertion was never written. It only ever proved TrySetVersion() doesn't throw.

 versioning.TrySetVersion(aggregate, e);
-
-
+ versioning.GetVersion(aggregate).ShouldBe(expected);

It passes — the behavior was right, the test just wasn't checking it.

xUnit1031

StoreUriStampingObserverTests had two sync tests calling .Wait() on tasks. Both are now async Task with await.

xUnit1051 — one documented exception

All 89 got the token threaded through, except one where the analyzer's advice is wrong:

// SetCanceled(CancellationToken) records which token DID cancel the task; it does not
// observe one. Passing the test's (uncancelled) token would assert something false.
#pragma warning disable xUnit1051
tcs.SetCanceled();
#pragma warning restore xUnit1051

Verification

Clean rebuild:  504 xunit warnings  ->  0
CS warnings:    1920                ->  1920   (unchanged)

The CS count is the check that matters — no warnings were traded for others, and no nullable-annotation changes leaked new ones.

Suite	Result	#581 baseline
CoreTests	475 passed, 1 skipped	475
CodegenTests	419	419
CommandLineTests	295	295
EventTests	648	648
EventStoreTests	72	72
CodegenTests.FSharp	1	1
JasperFx.SourceGenerator.Tests	19	19
JasperFx.Events.SourceGenerator.Tests	26	26

CoreTests and EventTests also verified green on net10.0.

Not included

I did not add WarningsAsErrors for these rules to stop them creeping back — that changes your build's failure behavior and wasn't part of the ask. Happy to add it if you want the guard.

🤖 Generated with Claude Code