Honor application class predicate in QuarkusInvokerFactory

[gsmet]

Fixes #47744
Fixes #47473
Related to #47563

@Postremus it was really handy to have the test case all ready, thanks! :).

[geoand]

CI failures seem related

[gsmet]

@geoand yeah, I just pushed a second commit. My first one apparently discovered a bug that was here all along.

[gsmet]

This change is to make things more consistent with GeneratedClassBuildItem and have the same methods available.

[geoand]

Thanks!

[gsmet]

@geoand yeah this issue already happened a few times. Using a fixed port for the TLS reload test is probably not a good idea but not sure how we could do better.

[quarkus-bot]

Status for workflow Quarkus CI

This is the status report for running Quarkus CI on commit 396ab2f.

✅ The latest workflow run for the pull request has completed successfully.

It should be safe to merge provided you have a look at the other checks in the summary.

You can consult the Develocity build scans.

[gsmet]

@holly-cummins while the previous behavior was indeed buggy, do you know why these problems started to pop, everywhere in the code base?

[holly-cummins]

@holly-cummins while the previous behavior was indeed buggy, do you know why these problems started to pop, everywhere in the code base?

I don't. :(

Here are some possible perturbations triggered by my changes that could be surfacing these:

• Tests run in a different order [not a bug, but different, so can expose bugs]
• Classes loaded in a different order. I'd need to check to confirm, but it could be that we used to load the Resource, and then the Test, and now we load the Test, and then the Resource.
• Different TCCL during some stage of the process [usually a bug but there's no one 'correct' TCCL to aim for]
• For this particular issue, it used to be that the test would have been partially executed in the system classloader before being re-executed in the quarkus classloader. I can imagine that maybe would change something?
• Config badness, such as config arriving too late, or being attached to the wrong classloader, so some behaviour-changing config isn't right [usually a bug]
• Extension processors all run at the beginning, instead of just before the test (where there's profiles, so I don't think that change would apply here) [as designed, but needs other code changes to not break stuff]