Generic record pattern not working in EJC

[schosin]

I've run into an issue with the EJC in a Java 21 project. The below example does not compile within Eclipse (4.35.0), but does with javac (JDK 21, JDK 24):

Cannot infer record pattern types for ComponentType.Wildcard(var bound)

The problem is line 33 and happens when Wildcard implements ComponentType with anything but ComponentType<T>.

import java.util.List;

public class Example {

    private static boolean matches(ComponentType<?> type, ComponentType.RegularComponentType<?> eventType) {
        return switch (type) {
            case ComponentType.RegularComponentType<?> regular -> switch (regular) {
                case ComponentType.ClassType(var clazz) -> switch (eventType) {
                    case ComponentType.ClassType(var eventClazz) -> clazz == eventClazz;
                };
            };
            /* Workaround: 
            case ComponentType.Wildcard<?> wildcard -> switch (eventType) {
                case ComponentType.ClassType(var eventClazz) -> wildcard.bound().isAssignableFrom(eventClazz);
            }; 
            /**/
            case ComponentType.Wildcard(var bound) -> switch (eventType) {
                case ComponentType.ClassType(var eventClazz) -> bound.isAssignableFrom(eventClazz);
            };
            /**/
        };
    }

    sealed interface ComponentType<T> {

        sealed interface RegularComponentType<T> extends ComponentType<T> {
        }

        record ClassType<T>(Class<T> clazz) implements RegularComponentType<T> {
        }

        // "implements ComponentType<T>" works, javac (JDK 21 & JDK 24) accepts both
        record Wildcard<T>(Class<T> bound) implements ComponentType<List<? extends T>> {
        }

    }

}

[srikanth-sankaran]

@mpalat - thanks for taking a look. Co-tagged @stephan-herrmann as owner for any consulting and code review requirements. Thanks gentlemen.

[stephan-herrmann]

I get three errors:

----------
1. ERROR in /tmp/Example.java (at line 6)
        return switch (type) {
                       ^^^^
A switch expression should have a default case
----------
2. ERROR in /tmp/Example.java (at line 17)
        case ComponentType.Wildcard(var bound) -> switch (eventType) {
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Cannot infer record pattern types for ComponentType.Wildcard(var bound)
----------
3. ERROR in /tmp/Example.java (at line 18)
        case ComponentType.ClassType(var eventClazz) -> bound.isAssignableFrom(eventClazz);
                                                        ^^^^^
bound cannot be resolved
----------
3 problems (3 errors)

While the failure to resolve the record patterns looks like the root cause, I still wonder what causes the failure to resolve bound?

[stephan-herrmann]

@mpalat do you still plan to work on this?

[mpalat]

@mpalat do you still plan to work on this?

I would love to, but maybe able to do so only after Oct 5 (in another 10 days or so). However, if you feel it needs attention earlier, please feel free to reassign.

[stephan-herrmann]

Inference fails when trying to reduce ⟨capture#1-of ? = List<? extends T#0>⟩

This constraint was reduced from ⟨ComponentType<capture#1-of ?> = ComponentType<List<? extends T#0>>⟩

This happens below IC18.findRPrimeAndResultingBounds(), invoked with

• typeBinding = Wildcard#RAW
• tPrime = ComponentType<capture#1-of ?>

Let's focus on tPrime, because that capture doesn't seem to have sufficient information to be matched against List<? extends T#0>

tPrime is the same instance as candidateT being passed into IC18.inferRecordPatternParameterization(..).

candidateT is the resolved type of the switch selector type. Nothing suspicious here.

Should IC18.deriveTPrime() have computed a "better" type? Yes.

Fix and explanation to follow shortly.