FlowActionServiceTest: 12 tests could not construct the service; 9 behavioural failures underneath

[rubenvdlinde]

On development, every test in tests/Unit/Service/Flow/FlowActionServiceTest.php errored before reaching an assertion:

ArgumentCountError: Too few arguments to function FlowActionService::__construct(),
7 passed in FlowActionServiceTest.php on line 52 and exactly 9 expected

FlowActionService gained EventCatalogService and ObjectService when the object-CRUD action nodes landed (e6925cc3b), and the test construction was never updated. So the suite has been reporting nothing at all for that service.

Fixed mechanically in #2072 — the constructor call now passes both collaborators, which turns 12 "cannot construct" errors into real signal.

What that signal says — 9 real failures

These are behavioural and I have deliberately NOT touched them, because "fixing" them by rewriting assertions to match whatever the code now does is exactly the test theatre this repo has been bitten by before. They need whoever owns the object-CRUD action work to say which side is wrong.

Representative case — testAgentActionMissingResultFieldIsSkippedAndLogged:

• The test asserts that an agent action with no resultField is skipped and logged as a warning.
• The action is skipped (dispatchTyped is never called, as expected).
• But no warning is emitted.

So an agent action that is silently misconfigured now disappears without a trace. That is either a lost guard (the warning should be restored) or an intentional change (the test should be updated to match). From outside the change that introduced it, I cannot tell which — but a step that skips silently is the failure mode this codebase keeps paying for, so my guess is the guard.

The other 8 are the same shape: expectations about warnings and about IMailer::send() being called that no longer hold.

Why it matters beyond this file

FlowActionService is the pre-engine action runner. As consumers migrate onto FlowEngine + IFlowNode (#2072), its behaviour is the reference for what the equivalent nodes must do. A test suite that has not actually run since e6925cc3b is a poor reference.

[rubenvdlinde]

Fixed in #2074.

The 9 behavioural failures were not regressions in FlowActionService. They were caused by the constructor fix in that same PR: I wired EventCatalogService in as a mock, and run() matches a flow's trigger through EventCatalogService::aliasesFor(). A mock returns [], so in_array($flowTrigger, []) was always false and every flow was skipped before any action ran — which is exactly why all nine failures shared one shape, with createEvent, send, dispatchTyped and warning never invoked at all.

EventCatalogService is a pure lookup over a class constant with no collaborators, so the test now constructs the real one.

The production code was correct throughout, including the agent-action guard I suspected of having been lost. Had the assertions been "fixed" to match observed behaviour instead, nine correct tests would have been deleted and the belief shipped that a misconfigured agent action legitimately passes in silence.

All 59 tests in tests/Unit/Service/Flow pass.