No generic signature emitted for def in object

[ohze]

minimized code

• scala

import java.util.function.{Function => JFunction}

class Behavior[T]
class BehaviorBuilder[T] {
  def onMessage(handler: JFunction[T, Behavior[T]]): BehaviorBuilder[T] = ???
}
object Behaviors {
  def same[T]: Behavior[T] = ???
}

• BehaviorTestKitTest.java

class BehaviorTestKitTest {
    static class X {
        void foo(){}
    }
    void test() {
        BehaviorBuilder<X> b = null;
        b.onMessage(
            x -> {
                x.foo();
                return Behaviors.same();
            })
        .onMessage(
            x -> {
                x.foo(); // error here
                return Behaviors.same();
            });
    }
}

Compilation output

[error] /dotty-example/src/main/java/BehaviorTestKitTest.java:14:1: cannot find symbol
[error]   symbol:   method foo()
[error]   location: variable x of type java.lang.Object
[error] x.foo
[info] /dotty-example/src/main/java/BehaviorTestKitTest.java: /dotty-example/src/main/java/BehaviorTestKitTest.java uses unchecked or unsafe operations.

expectation

compile successfully as in scala 2

inspect

Decompile Behaviors.class using Procyon v0.5.36:

• dotty's output:

public final class Behaviors {
    public static Behavior same() {
        return Behaviors$.MODULE$.same();
    }
}

• scala 2.13.1 output:

public final class Behaviors {
    public static <T> Behavior<T> same() {
        return Behaviors$.MODULE$.same();
    }
}

Notice the lack of generic signature on same() method.

maybe relate: #6350

[odersky]

The issue here is that the Dotty backend does not demand a generic signature for forwarders. Both scalac and dotc generate a static signature for same in the object class Bahavior$. But only scalac generates a static signature for the static forwarder of same in class Bahavior.

I verified that the BCode backend simply never makes the call to demand a signature for the forwarder. Hopefully this will be fixed once we switch to scalac's backend. Assigning to @lrytz for tracking this.

[smarter]

Closing as a duplicate of #10347 where I provided a workaround: #10347 (comment)