A normal scoped bean implemented as Java record cannot be serialised to json

[JackyAnn]

Describe the bug

  /**
   * 
   * Can not be serialised
   * @return
   */
  @Produces
  @ApplicationScoped
  public ApplicationRecord create1(){
    return new ApplicationRecord("ApplicationRecord");
  }

  @Produces
  @Singleton
  public SingletonRecord create2(){
    return new SingletonRecord("SingletonRecord");
  }

  public record ApplicationRecord(String name){

  }

  public record SingletonRecord(String name){

  }

@QuarkusTest
public class GreetingResource1Test {
    @Inject
    private JsonConfigCustomizer.ApplicationRecord applicationRecord;
    @Inject
    private JsonConfigCustomizer.SingletonRecord singletonRecord;
    @Inject
    private Jsonb jsonb;

    @Test
    public void testHelloEndpoint() {
        /**
         * print
         * {}
         */
        System.out.println("---"+jsonb.toJson(applicationRecord));
        
        /**
         * print
         * {
         *     "name": "SingletonRecord"
         * }
         */
        System.out.println("---"+jsonb.toJson(singletonRecord));
    }




}

Expected behavior

No response

Actual behavior

No response

How to Reproduce?

No response

Output of uname -a or ver

Darwin jackydeMacBook-Pro.local 22.5.0 Darwin Kernel Version 22.5.0: Mon Apr 24 20:51:50 PDT 2023; root:xnu-8796.121.2~5/RELEASE_X86_64 x86_64

Output of java -version

java version "20" 2023-03-21 Java(TM) SE Runtime Environment (build 20+36-2344) Java HotSpot(TM) 64-Bit Server VM (build 20+36-2344, mixed mode, sharing)

GraalVM version (if different from Java)

No response

Quarkus version or git rev

3.1.0.Final

Build tool (ie. output of mvnw --version or gradlew --version)

No response

Additional information

No response

[JackyAnn]

demo
https://github.com/JackyAnn/quarkus-devui-issues.git

[mkouba]

@JackyAnn thanks for the rerport. Could you be more specific? What exactly does not work? It seems that you're trying to serialize an @ApplicationScoped CDI bean represented by a Java record to JSON using jsonb, right?

[quarkus-bot]

/cc @Ladicek (arc), @manovotn (arc)

[Ladicek]

At first glance, I don't think it really matters that the bean is a record -- the cause here is that it is normal scoped. That is, you're trying to serialize the client proxy. Could you please try jsonb.toJson(ClientProxy.unwrap(applicationRecord)) and report back?

[JackyAnn]

@Ladicek
I've tried jsonb.toJson(ClientProxy.unwrap(applicationRecord)) and it still doesn't serialise properly.

[Ladicek]

OK, that sounds weird. I'll take a proper look.

[Ladicek]

I have reproduced this locally with my own code and there's one really frustrating thing. If I declare

public class Producers {
    @Produces
    @ApplicationScoped
    public App app() {
        return new App("Application");
    }

    @Produces
    @Singleton
    public Sing sing() {
        return new Sing("Singleton");
    }

    public record App(String name) {
    }

    public record Sing(String name) {
    }
}

then Producers.App.class.isRecord() returns false for me (while Producers.Sing.class.isRecord() is true as expected). I have no idea why.

I'm using OpenJDK 17.0.7.

[Ladicek]

Ahh I think I know! We do a bytecode transformation on App which removes the final flag from the class and adds a zero-parameter constructor. In my JDK source code, Class.isRecord() looks like this:

public boolean isRecord() {
    // this superclass and final modifier check is not strictly necessary
    // they are intrinsified and serve as a fast-path check
    return getSuperclass() == java.lang.Record.class &&
            (this.getModifiers() & Modifier.FINAL) != 0 &&
            isRecord0();
}

The class doesn't have the final flag, therefore Class.isRecord() returns false, and that causes Yasson to not use the special case for treating record component accessor methods as getters. Phew!

[Ladicek]

I think we probably shouldn't remove the final flag from records. That would prevent using records as normal scoped beans, which is probably a good idea anyway.

[manovotn]

That would prevent using records as normal scoped beans, which is probably a good idea anyway.

Yea, that sounds fishy anyway.
I was thinking about records and CDI anyway - now that next CDI version is on JDK 17+ (actually, it's likely 21), we should perhaps add a note about this into the spec? I can create an issue

[mkouba]

I think we probably shouldn't remove the final flag from records. That would prevent using records as normal scoped beans, which is probably a good idea anyway.

👍