Why doesn't the first of the following examples work?

  • run(R::new); method R.run is not called.
  • run(new R()); method R.run is called.

Both examples are compiled-able.

public class ConstructorRefVsNew {

  public static void main(String[] args) {
      new ConstructorRefVsNew().run(R::new);
      System.out.println("-----------------------");
      new ConstructorRefVsNew().run(new R());
  }

  void run(Runnable r) {
      r.run();
  }

  static class R implements Runnable {

      R() {
          System.out.println("R constructor runs");
      }

      @Override
      public void run() {
          System.out.println("R.run runs");
      }
  }
}

The output is:

  R constructor runs
  -----------------------
  R constructor runs
  R.run runs

In the first example, the R constructor is called, it returns lambda (which is not object):

But then how is it possible, that the example is compiled successfully?


Your run method takes a Runnable instance, and that explains why run(new R()) works with the R implementation.

R::new is not equivalent to new R(). It can fit the signature of a Supplier<Runnable> (or similar functional interfaces), but R::new cannot be used as a Runnable implemented with your R class.

A version of your run method that can takeR::new could look like this (but this would be unnecessarily complex):

void run(Supplier<Runnable> r) {
    r.get().run();
}

Why does it compile?

Because the compiler can make a Runnable out of the constructor call, and that would be equivalent to this lambda expression version:

new ConstructorRefVsNew().run(() -> {
    new R(); //discarded result, but this is the run() body
});

The same applies to these statements:

Runnable runnable = () -> new R();
new ConstructorRefVsNew().run(runnable);
Runnable runnable2 = R::new;
new ConstructorRefVsNew().run(runnable2);

But, as you can notice, the Runnable created with R::new does just call new R() in its run method body.


A valid use of a method reference to execute R#run could use an instance, like this (but you'd surely rather use the r instance directly, in this case):

R r = new R();
new ConstructorRefVsNew().run(r::run);

The first example:

new ConstructorRefVsNew().run(R::new);

is more or less equivalent to:

new ConstructorRefVsNew().run( () -> {new R();} );

The effect is you just create an instance of R but do not call its run method.


Compare two calls:

((Runnable)() -> new R()).run();
new R().run();

By ((Runnable)() -> new R()) or ((Runnable) R::new) , you create a new Runnable which does nothing1.

By new R(), you create an instance of the R class where the run method is well-defined.


1 Actually, it creates an object of R which has no impact on execution.


I was thinking of treating 2 invocations identically without modifying the main method. We would need to overload run(Runnable) with run(Supplier<Runnable>).

class ConstructorRefVsNew {

    public static void main(String[] args) {
        new ConstructorRefVsNew().run(R::new);
        System.out.println("-----------------------");
        new ConstructorRefVsNew().run(new R());
    }

    void run(Runnable r) {
        r.run();
    }

    void run(Supplier<Runnable> s) {
        run(s.get());
    }

    static class R implements Runnable { ... }
}

The run method expects a Runnable.

The easy case is new R(). In this case you know the result is an object of type R. R itself is a runnable, it has a run method, and that's how Java sees it.

But when you pass R::new something else is happening. What you tell it is to create an anonymous object compatible with a Runnable whose run method runs the operation you passed it.

The operation you passed it is not R's run method. The operation is the costructor of R. Thus, it's like you have passed it an anonymous class like:

new Runnable() {

     public void run() {
         new R();
     }
}

(Not all the details are the same, but this is the closest "classical" Java construct ).

R::new, when called, calls new R(). Nothing more, nothing less.