Improve stereotypes support

[Ladicek]

This commit adds a better API for registering custom stereotypes, similar
to existing APIs: StereotypeRegistrar and StereotypeRegistrarBuildItem.
The existing AdditionalStereotypeBuildItem is very confusing to use, so
it is deprecated and all its usages are replaced with the new API.

Also, this commit allows defining stereotypes for synthetic beans. Scope,
name, alternative status and priority are all applied to the synthetic bean,
but interceptor bindings are not. We don't apply interceptors to synthetic
beans in general, this case is not an exception.

[Ladicek]

Creating as draft to let CI run in my fork, and to discuss the TODO items with @mkouba.

[Ladicek]

Rebased. @mkouba could you please take a look? Thank you!

[Ladicek]

Removed all TODOs, so marking as ready.

[Ladicek]

Rebased to get rid of the RestClient test failures.

[mkouba]

The CI failure may be related?

[Ladicek]

Yea I need to check why the SmallRye Health test failed in CI, it all passed locally.

[mkouba]

Yea I need to check why the SmallRye Health test failed in CI, it all passed locally.

So what's the status of this PR?

[Ladicek]

I'm just getting back to this. (I wanted to work on #27574 and figured it would be best to have this done first).

[Ladicek]

I can also just get rid of the SmallRye Health changes, as they are not really relevant to the subject of this PR. I just stumbled upon that piece of code and found it's suboptimal, so I figured I could just fix it and squeeze it in. I'll investigate for a bit and if I don't manage to find the issue, I'll get rid of the changes here.

[Ladicek]

So the test failure from CI is:

Expected: iterable with items ["alpha1", "bravo1"] in any order
  Actual: <[alpha2, bravo1]>

The health checks are:

static class HealthCheckProducers {
    static final AtomicInteger ALPHA_COUNTER = new AtomicInteger();
    static final AtomicInteger BRAVO_COUNTER = new AtomicInteger();

    // No scope - @Singleton is used by default
    @Readiness
    HealthCheck alpha() {
        int idx = ALPHA_COUNTER.incrementAndGet();
        return () -> HealthCheckResponse.builder().up().name("alpha" + idx).build();
    }

    @RequestScoped
    @Readiness
    HealthCheck bravo() {
        int idx = BRAVO_COUNTER.incrementAndGet();
        return () -> HealthCheckResponse.builder().up().name("bravo" + idx).build();
    }
}

So alpha has no scope annotation, thereby defaulting to singleton. There should only ever be alpha1. The existence of alpha2 means the alpha health check has a different scope. I have no idea why.

Gonna revert those changes, then.

[Ladicek]

Done.

[mkouba]

So alpha has no scope annotation, thereby defaulting to singleton. There should only ever be alpha1. The existence of alpha2 means the alpha health check has a different scope. I have no idea why.

I believe that this would deserve a little investigation. Do you have a diff somewhere or a branch with a state before you reversed the change?

[Ladicek]

I do: https://github.com/Ladicek/quarkus/commits/smallrye-health-default-scope The interesting thing is, the test didn't fail for me locally, and only failed in CI on Windows. I don't know what to take from it.

[mkouba]

The interesting thing is, the test didn't fail for me locally, and only failed in CI on Windows. I don't know what to take from it.

That's superweird.