[REGRESSION] - quarkus.rest.jackson.optimization.enable-reflection-free-serializers=true broken in 3.18

[edeandrea]

Describe the bug

I've found that setting quarkus.rest.jackson.optimization.enable-reflection-free-serializers=true on 3.18 causes some things to break

How to Reproduce?

rest-fights.zip

1. Unzip
2. cd rest-fights
3. ./mvnw clean verify

There are two unit tests - 1 should pass and one should fail. If you change the quarkus version back to 3.17.8 all the tests pass.

[quarkus-bot]

/cc @geoand (jackson), @gsmet (jackson), @mariofusco (jackson)

[geoand]

I'll look into this as it's really weird

[geoand]

Actually, the status code is a result of how the test is setup and a bug in the generated deserializers. Fixing the deserializers will automatically fix the issue.

Here is what the deserializer for FightLocation looks like:

public class FightLocation$quarkusjacksondeserializer extends StdDeserializer {
   public FightLocation$quarkusjacksondeserializer() {
      super(FightLocation.class);
   }

   public Object deserialize(JsonParser var1, DeserializationContext var2) throws IOException, JacksonException {
      JsonNode var3 = (JsonNode)var1.getCodec().readTree(var1);
      FightLocation var9 = new FightLocation();
      Iterator var4 = var3.fields();

      while (var4.hasNext()) {
         Entry var5 = (Entry)var4.next();
         JsonNode var8 = (JsonNode)var5.getValue();
         if (!var8.isNull()) {
            Object var6 = var5.getKey();
            int var7 = var6.hashCode();
            switch (var7) {
               case -1724546052:
                  if ("description".equals(var6)) {
                     var8.asText();
                  }
                  break;
               case -577741570:
                  if ("picture".equals(var6)) {
                     var8.asText();
                  }
                  break;
               case 3373707:
                  if ("name".equals(var6)) {
                     var8.asText();
                  }
            }
         }
      }

      return var9;
   }
}

The switch looks very weird, so at this point I'll turn it over to @mariofusco.

[mariofusco]

@geoand That's weird, I don't think that I have changed that code recently, but I will give a look immediately.

[geoand]

Maybe it's a bug in gizmo? cc @Ladicek

[Ladicek]

The last Gizmo release has happened a year ago, so if there's a difference between Quarkus 3.17 and 3.18, I'm fairly sure it's not Gizmo's fault.

[Ladicek]

That switch statement in one of the previous comments is basically a switch on Strings, there's nothing wrong with it.

EDIT: actually, there is. The obtained values are not assigned anywhere.

[geoand]

The last Gizmo release has happened a year ago, so if there's a difference between Quarkus 3.17 and 3.18, I'm fairly sure it's not Gizmo's fault.

Gotcha, thanks

[geoand]

With Quarkus 3.17.8, the deserializer looks like this:

public class FightRequest$quarkusjacksondeserializer extends StdDeserializer {
   public FightRequest$quarkusjacksondeserializer() {
      super(FightRequest.class);
   }
}

[mariofusco]

With Quarkus 3.17.8, the deserializer looks like this:

public class FightRequest$quarkusjacksondeserializer extends StdDeserializer {
public FightRequest$quarkusjacksondeserializer() {
super(FightRequest.class);
}
}

In Quarkus 3.17.8 that deserializer is generated and then discarded (not used) because it doesn't work with records. I made a change ee668b2 to overcome that limitation, but I didn't think to a situation of a record with an empty constructor like

public record FightLocation(String name, String description, String picture) {
  public FightLocation() {
    this(null, null, null);
  }
}

(@edeandrea why are you doing this? just joking 😄 ) so that the deserializer try to use that empty contructor but then it is not able to set any value on the immutable record, thus generating that useless switch. I will work on a fix.

[Ladicek]

For the record, in Jandex, you can find the canonical record constructor using ClassInfo.canonicalRecordConstructor().

[mariofusco]

For the record, in Jandex, you can find the canonical record constructor using ClassInfo.canonicalRecordConstructor().

Nice, that's super helpful, thank you.

[mariofusco]

This should fix the problem #46002

Thank you @Ladicek for your suggestion.

[edeandrea]

(@edeandrea why are you doing this? just joking 😄 )

Sorry @mariofusco I must be using it wrong. Always user error :)

But seriously, thank you all for getting this fixed so fast.