Stop IntegrationContext from disposing a class fixture it doesn't own (GH-3423)

[jeremydmiller]

Partial fix for #3423. This does not close the issue — see "What this doesn't fix" below. I'd rather land the correct-ownership change on its own merits than bundle it with a cause I haven't pinned yet.

The defect

DefaultApp is an xUnit class fixture — xUnit owns its lifetime and disposes it when the test class finishes. IntegrationContext.Dispose() disposed it anyway, after every test method:

public virtual void Dispose()
{
    _default.Dispose();   // disposing a fixture this class does not own
}

DefaultApp.Dispose() nulls the host, and the guard that papered over the resulting damage silently substituted a differently-configured host:

public void RecycleIfNecessary()
{
    if (Host == null) Host = WolverineHost.Basic();   // NOT the DefaultApp configuration
}

WolverineHost.Basic() has none of DefaultApp's IncludeType<MessageConsumer>(), IncludeType<InvokedMessageHandler>(), RegisterMessageType(...) calls, or the IAdditionService/IIdentityService/ITrackedTaskRepository registrations. So from the second test method in a class onward, tests could be running against a host they were never written for — and whether any given test passed came down to ordering.

The ownership was backwards in the other direction too: with(...) builds a host for the test, and that one was never disposed at all — a straight leak of a started IHost per call.

The fix

The test disposes what it built; the fixture is left to xUnit.

private IHost? _ownedHost;   // only set by with()

public virtual void Dispose() => _ownedHost?.Dispose();

protected async Task with(Action<WolverineOptions> configuration)
{
    _ownedHost?.Dispose();
    _ownedHost = await WolverineHost.ForAsync(configuration);
    Host = _ownedHost;
}

RecycleIfNecessary() and the Host = null! dance are deleted rather than kept as a band-aid — with the fixture no longer torn down mid-class, there's nothing to recycle. The one caller (find_handlers_with_the_default_handler_discovery) was already a no-op and is removed.

What this fixes

• The NullReferenceExceptions out of DefaultApp.ChainFor<T>() (a null Host)
• The leaked with() hosts
• The xUnit fixture-lifetime violation itself

What this doesn't fix

19 of the 20 filtered-subset failures survive this change, with a separate cause. Running a StorageActionCompliance-derived class before another IntegrationContext class leaves the later host's HandlerGraph unable to handle conventionally-discovered messages — HandlerGraph.CanHandle(...) returns false, so findInvoker falls through to routing and throws IndeterminateRoutesException.

Deterministic repro:

# green
--filter "FullyQualifiedName~CoreTests.Persistence.using_multiple_storage_actions"
# 2 failures
--filter "...using_multiple_storage_actions|...using_storage_return_types_and_entity_attributes"

The obvious explanation — DisableConventionalDiscovery() on the compliance host leaking into hosts built later — is disproven: I built a host with DisableConventionalDiscovery(), disposed it, then built a plain host, and conventional discovery worked fine. Mechanism still unidentified. Written up on #3423 so the next person doesn't re-run the same dead end.

Verification

• Full CoreTests assembly: 1931 passed, 0 failed (unchanged — this is what CI runs)
• CoreTests.Persistence subset: 20 → 19 failures

🤖 Generated with Claude Code