Scala and Java - Implicit Parameters and Inheritance

Question

The following code gives an error:

package test

trait Base {
  def method:String
}

trait Trait extends Base {
  def method()(implicit i:String):String = { "2" }
}

object Object extends Trait {

}

The error is "object creation impossible, since method method in class Base of type => String is not defined"

The above error is fixed by following code

package test

trait Base {
  def method:String
}

trait Trait extends Base {
  def method:String = method()("String") // Over loading
  def method()(implicit i:String):String = { "2" }
}

object Object extends Trait {

}

Now instead of Scala class, when I define a Java interface as follows:

// Java Code
package test;

public interface JBase {
  String method();
}

// Scala Code
package test

trait Trait extends JBase {
  def method:String = method()("10")
  def method()(implicit i:String):String = { "2" }
}

object Object extends Trait {

}

I get an error "ambiguous reference to overloaded definition, both method method in trait Trait of type ()(implicit i: String)String and method method in trait Trait of type ()String match argument types ()"

What is the difference in both these scenarios that makes the compiler behave differently? How do I solve this issue?

Also, I would make it a general rule of thumb that if you resort to overloading in Scala then you are doing something wrong. — Apocalisp
What are the alternatives to overloading? The java interface here is not controlled by me i.e. it is part of a library. I thought use of an implicit is a cool way of maintaining library signatures for api calls while adding additional functionality. — 0n4li

Owen Owen · Accepted Answer · 2013-01-13T19:17:21

Here's an example that I think shows clearly what is going on:

object Test extends App {
    class A { def f(): String = "x" }
    class B extends A { override def f: String = "y" }
    class C { def f: String = "z" }

    println { (new A).f() } // OK
    println { (new B).f() } // OK
    println { (new C).f() } // FAILS
}

A: Has parentheses, called with parentheses, everything good.
B: Has no parentheses, but supertype does have parentheses, so still good. This corresponds to your case.
C: Has no parentheses, called with parentheses, no good.

Basically, Java methods are always considered "with parentheses", so, Trait's supertype has the parentheses, so using parentheses in method()("string") is insufficient to clarify which method you mean.

edit: Honestly I think you are better off renaming the method. Even in the case where there is no ambiguity, the behavior could be quite surprising:

trait Trait {
    def method: String = method()("x")
    def method()(implicit i: String): String = i
}

val t = new Trait { }
implicit val s = "y"

println { t.method }
> "x"

println { t.method() }
> "y"

Furthermore, the fact that the name is the same isn't buying you anything in terms of polymorphism: only the non-implicit method overrides Base.method -- it's just an aesthetic decision to make the names be the same.

Scala and Java - Implicit Parameters and Inheritance

1 Answers