Quarkiverse extensions can build on other OSes too

[gastaldi]

This PR introduces some improvements to Quarkiverse extensions' build script:

• Changes the build script to easily allow building on other OSes (commented out by default)
• Use actions/setup-java's cache

[quarkus-bot]

This workflow status is outdated as a new workflow run has been triggered.

Failing Jobs - Building 4096d94

Status	Name	Step	Failures	Logs	Raw logs
✔️	JVM Tests - JDK 11
✖	JVM Tests - JDK 11 Windows	Build	Failures	Logs	Raw logs
✔️	JVM Tests - JDK 17

Full information is available in the Build summary check run.

Failures

⚙️ JVM Tests - JDK 11 Windows #

- Failing: extensions/vertx/deployment 
! Skipped: core/test-extension/deployment extensions/agroal/deployment extensions/amazon-lambda-http/deployment and 315 more

📦 extensions/vertx/deployment

✖ io.quarkus.vertx.DuplicatedContextTest.testThatBlockingEventConsumersAreCalledOnDuplicatedContext - More details - Source on GitHub

(RECIPIENT_FAILURE,8185) io.smallrye.mutiny.TimeoutException
	at io.vertx.core.eventbus.Message.fail(Message.java:141)
	at io.quarkus.vertx.runtime.VertxRecorder$3$1$1.handle(VertxRecorder.java:113)

[maxandersen]

Assuming tests passes lgtm

[gsmet]

Just want to open a debate, please dismiss if you think it's useless.

Is it really a good thing to multiply our CI usage by 3, when in most cases this will have little value? I would think it could be enabled by the Quarkiverse authors when they think it makes sense for their libraries (we had one so far AFAICS).

Also, we wanted to enable native testing by default for Quarkiverse extensions and it will require some non trivial work with Windows in the matrix.
So if we decide to do that, I think we should add native testing on Linux (and I would limit it to Linux for the default behavior) because it might not be easy for the Quarkiverse contributor to do it afterwards with this new setup.

[gsmet]

Btw this question: Is it really a good thing to multiply our CI usage by 3, when in most cases this will have little value? has 3 dimensions:

• GitHub Actions resources in general
• The planet
• Our GitHub Actions limited resources -> we multiply by 3 our GitHub Actions usage and I know some of you were already worried about the day when we would hit the limit

[gsmet]

I wouldn't be against having this matrix setup with Linux by default and a commented line with Linux/Windows.

I see little value in testing with macOS in the Java world, except in very specific cases.

[maxandersen]

You are right @gsmet it does not make sense to enable This for All by default.

Maybe parameteriZe the matrix and let users enable whats needed.

Osx is relevant for native but yes; not by default.

[gastaldi]

@gsmet The reason why I am doing this is because some Quarkiverse builds were reported to fail in Windows (see quarkiverse/quarkus-openapi-generator#7).

But I agree that it wouldn't be necessary from day one. I like your suggestion of having the matrix setup with Linux by default and some commented out lines for Windows. I also agree that testing in macOS is not very valuable (perhaps if one likes to investigate too many open files errors, but that's another story) :)

[gsmet]

some Quarkiverse builds were reported to fail in Windows (see quarkiverse/quarkus-openapi-generator#7).

Yes, I have seen that, that's the one occurrence I was referencing to. I'm just unsure we should generalize this based on a few occurrences where it makes absolutely sense.

CI has a cost, even if it appears free to us, and I think we should try to be good citizens about it when we can.