5.1 Modelling Inheritance

In felix, objects are defined dynamically by constructor functions. And we can combine objects with the extend combinator. Lets see how we can model inheritance with this mechanism!

  // The base.
  interface A[T] { print : 1 -> 0; get: 1 -> list[T]; }
  object fred[T] (x:list[T]) implements A[T] = 
  {
    method proc print () { println$ x; }
    method fun get () => x;
  }
  
  var lst = list$ 1,2,3;
  var a = fred lst;
  
  a.print();
  println$ a.get();
  
  println$ "*" * 20;
  
  // OK, so an inheritance example.
  
  interface B[T] extends A[T] { pprint: 1 -> 0; }
  
  object joe () extends a implements B[int] = {
    method proc pprint () { print "Hello "; a.print(); }
  }
  
  var b = joe();
  b.print();
  b.pprint();
  println$ b.get();

Here we have inherited from an object! The two objects, a and b, encapsulate two distinct state variables, and can be modified independently.

This is basically prototype based OO. Here's a more conventional class-like way to do it, with only a single state variable:

  interface C[T] extends A[T] { pprint: 1 -> 0; }
  
  object joe[T] (x:list[T]) extends fred x as var y implements C[T] = {
    method proc pprint () { print "Hello "; y.print(); }
    method proc print () { print "OVERRIDE "; y.print(); }
  }
  
  var c = joe lst;
  c.print();
  c.pprint();
  println$ c.get();

This is quite different because the fred object we're extending is created by the joe constructor, in fact we have to use the "as var y" notation to give it a name. fred is very similar to a C++ base class here, except we don't make the mistake of C++ and refer to the base by its type: instead we name the subobject (which is what C++ should have done).

In this example you can also observe overriding the print method. This works because during extension if two record fields have the same name, only the later one is kept. Since methods are record fields, the print method of fred is discarded in favour of the print method of joe.

I hope you can see that this system is very powerful. It can do both prototype based OO as well as class based OO, or both in any combination, depending entirely on how you construct the private state. In fact, you get multi-methods for free as well.

What one needs to get used to is that the object is divorced from the private states its methods implement. It's also necessary to throw out the weak nominal typing an inheritance model used by C++ and Java in favour of the much more usable structural typing provided by records.