DependentContextTest#testDependentScopedInterceptorsAreDependentObjectsOfBean() does not test the spec assertion

[mkouba]

The org.jboss.cdi.tck.tests.context.dependent.DependentContextTest#testDependentScopedInterceptorsAreDependentObjectsOfBean() test method does not test the linked spec assertion:

Instances of interceptors or decorators with scope @Dependent are dependent objects of the bean instance they decorate.

It merely asserts that if CreationalContext#release() is called then the TransactionalInterceptor#destroy() callback is invoked because the AccountTransaction bean is destroyed. And even this assert is questionable because the test should rather call Bean#destroy() instead; CreationalContext#release() should destroy the dependent objects and not the bean itself.

In fact, this assertion should be testable=false because there's no reasonable way to test it (an interceptor may not declare its own @PreDestroy callback).

I'd recommend to mark the assertion as testable=false and remove the test method completely.

[Azquelt]

Could you test this indirectly by injecting a @Dependent bean into the interceptor and ensuring that it gets destroyed (which should only happen if the interceptor itself is destroyed)?

[Ladicek]

That is a good idea! I'll see if I can rewrite the test like that.

[Ladicek]

Seems to work well -- submitted #462. Thanks @Azquelt!

[manovotn]

Very good idea indeed, that should do the trick!

[manovotn]

Added the accepted label as we seem to agree that the current state isn't OK and we now have a way to test it properly

[manovotn]

As discussed in the meeting today, the solution to this challenge needs to be a test exclusion.
The challenge is correct and so is the fix. However, previous version of the TCK only passed with Weld due to a Weld bug (see https://issues.redhat.com/browse/WELD-2743).
Therefore, in order for Weld releases to keep passing latest TCKs, the correct course of action is excluding this test for 4.0.x and re-enabling them for 4.1. Otherwise, we'll repeat the situation where older Weld version won't pass latest TCKs.

I'll send a PR for this.

[manovotn]

As discussed in the meeting today, the solution to this challenge needs to be a test exclusion. The challenge is correct and so is the fix. However, previous version of the TCK only passed with Weld due to a Weld bug (see https://issues.redhat.com/browse/WELD-2743). Therefore, in order for Weld releases to keep passing latest TCKs, the correct course of action is excluding this test for 4.0.x and re-enabling them for 4.1. Otherwise, we'll repeat the situation where older Weld version won't pass latest TCKs.

I'll send a PR for this.

@Ladicek @Emily-Jiang it seems that I was mistaken and that we got a little confused in the mtg discussion yesterday :)
Weld actually passes this test (after changes made in #462) even on older versions - I tried with 5.0.0.Final.
What wouldn't work is Weld 5.1.1 (not yet released) with older TCK version, such as 4.0.9. However, that doesn't really matter as the requirement is for all impl versions to pass the most recent TCKs.
Therefore, no action is needed here - sorry for the hassle.