How to force a new instantiation of a lamda-definition

Question

The Java-Spec guarantees that a given lamda-definition, e.g. () -> "Hello World", is compiled/converted to exactly one implementation class (every definition, not every occurence that "looks" the same).

Is there any way I can force the java-compiler/jvm to generate a new lamda-definition instead of sharing a common one? I am currently implementing a library that weaves multiple function parts into a BiFunction which suffers from mega-morphic call-sites because of the guarantees given by the java-spec (EDIT: I stand corrected: the Java-Spec does not guarantee a single shared class - the current reference implementation does this though):

        public <In, Out, A> BiFunction<In, Out, Out> weave(
             Function<? super In, A> getter,
             BiConsumer<? super Out, ? super A> consumer
        ) {
            return (in, out) -> {
                consumer.accept(out, getter.apply(in));
                return out;
            };
        }

Every lamda generated through this code shares the same lamda-definition and is thus mostly uninlineable / unoptimizeable.

Answer 1

In the current implementation, the caching of generated classes (or even instances for the non capturing lambda expressions), is a property of the invokedynamic instruction which will reuse the result of the bootstrapping done on the first execution.

The bootstrap method itself, hosted in the LambdaMetafactory class will generate a new class each time it is invoked. So when you use this factory directly, you’ll get a new class on each invocation, under the current implementation.

public <In, Out, A> BiFunction<In, Out, Out> weave(
     Function<? super In, A> getter,
     BiConsumer<? super Out, ? super A> consumer) {

    MethodHandles.Lookup l = MethodHandles.lookup();
    try {
        MethodHandle target = l.findStatic(l.lookupClass(), "weaveLambdaBody",
            MethodType.methodType(Object.class, Function.class, BiConsumer.class,
                Object.class, Object.class));
        MethodType t = target.type().dropParameterTypes(0, 2);
        return (BiFunction<In, Out, Out>)LambdaMetafactory.metafactory(l, "apply",
            target.type().dropParameterTypes(2, 4).changeReturnType(BiFunction.class),
            t, target, t) .getTarget().invokeExact(getter, consumer);
    }
    catch(RuntimeException | Error e) {
        throw e;
    }
    catch(Throwable t) {
        throw new IllegalStateException(t);
    }
}
private static <In, Out, A> Out weaveLambdaBody(
    Function<? super In, A> getter,
    BiConsumer<? super Out, ? super A> consumer,
    In in, Out out) {

    consumer.accept(out, getter.apply(in));
    return out;
}

First, you have to desugar the lambda body into a method. The captured values come first in its parameter list, followed by the parameters of the functional interface type. The LambdaMetafactory has an exhaustive documentation about its usage.

While I kept the type parameters for documentation purposes, it should be clear that you lose the compile-time safety here, with such an operation.