CompletableFuture smart dispatch support

[franz1981]

• resolves CompletableFuture smart dispatch support #30133

[franz1981]

@cescoffier Not sure I made it right, I see lot of duplicate codes :/
I will create a separate issue in case it seems a good use case

The reason why I've done it is on this Quarkus profiling workshop snippet where I've been forced to use CompletionStage in order to use the I/O threads, while failing (ie using the blocking thread pool) with the actual concrete type CompletableFuture.

[cescoffier]

What's the rationale? We strongly recommend Uni in this case (for plenty of reason including context handling).

[cescoffier]

So basically, do you have the same issue when using Uni instead of CS?

[franz1981]

So basically, do you have the same issue when using Uni instead of CS?

The issue is related of how we recognize non-blocking endpoint: CS isn't CF and it's going to use a thread pool for that, that's unexpected from a user pov

[cescoffier]

Ah ok, it makes sense (even if I would prefer users to move away from CS/CF as fast as they can)

[franz1981]

Ah ok, it makes sense (even if I would prefer users to move away from CS/CF as fast as they can)

Yep and still my changes are not correctly handling polymorphism there (if someone impl their own CompletionStage children my changes won't help to detect it), given that's not an high prio thing I'm happy to close it unless can still be useful to someone other then me and Mario for the workshop: let me know and I can happily close this ;)

[geoand]

Also, do we really want to do this when RESTEasy Reactive already has support for it?

[franz1981]

@geoand Renaming should look better and I've opened an enhancement request: cannot tell if it's a bug or a missing feature really

[geoand]

Thanks!

[geoand]

CI is complaining :(

[franz1981]

CI is complaining :(

yep, let me check what going on

[franz1981]

@geoand it seems that ServletSimpleRestTestCase::testAsync is failing to run on the I/O thread even if I switch to CompletionStage type on the endpoint signature...is it supposed to happen? @cescoffier too

[geoand]

Servlet is always meant to run on a worker thread.

I am looking for the piece of code that checks this

[franz1981]

Oki so I can just disable that test for the servlet use case

[geoand]

Yeah, that makes sense

[geoand]

I fixed the formatting issue and rebased the PR onto main.