Java Best Practice for type resolution at runtime.

Question

I'm trying to define a class (or set of classes which implement the same interface) that will behave as a loosely typed object (like JavaScript). They can hold any sort of data and operations on them depend on the underlying type.

I have it working in three different ways but none seem ideal. These test versions only allow strings and integers and the only operation is add. Adding integers results in the sum of the integer values, adding strings concatenates the strings and adding an integer to a string converts the integer to a string and concatenates it with the string. The final version will have more types (Doubles, Arrays, JavaScript-like objects where new properties can be added dynamically) and more operations.

Way 1:

public interface DynObject1 {
  @Override public String toString();
  public DynObject1 add(DynObject1 d);
  public DynObject1 addTo(DynInteger1 d);
  public DynObject1 addTo(DynString1 d);
}


public class DynInteger1 implements DynObject1 {
  private int value;

  public DynInteger1(int v) {
    value = v;
  }

  @Override
  public String toString() {
    return Integer.toString(value);
  }

  public DynObject1 add(DynObject1 d) {
    return d.addTo(this);
  }

  public DynObject1 addTo(DynInteger1 d) {
    return new DynInteger1(d.value + value);
  }

  public DynObject1 addTo(DynString1 d)
  {
    return new DynString1(d.toString()+Integer.toString(value));
  }
}

...and similar for DynString1

Way 2: public interface DynObject2 { @Override public String toString(); public DynObject2 add(DynObject2 d); }

public class DynInteger2 implements DynObject2 {
  private int value;

  public DynInteger2(int v) {
    value = v;
  }

  @Override
  public String toString() {
    return Integer.toString(value);
  }

  public DynObject2 add(DynObject2 d) {
    Class c = d.getClass();

    if(c==DynInteger2.class)
    {
      return new DynInteger2(value + ((DynInteger2)d).value);
    }
    else
    {
      return new DynString2(toString() + d.toString());
    }
  }
}

...and similar for DynString2

Way 3:

public class DynObject3 {

  private enum ObjectType {
    Integer,
    String
  };

  Object value;
  ObjectType type;

  public DynObject3(Integer v) {
    value = v;
    type = ObjectType.Integer;
  }

  public DynObject3(String v) {
    value = v;
    type = ObjectType.String;
  }

  @Override
  public String toString() {
    return value.toString();
  }

  public DynObject3 add(DynObject3 d)
  {
    if(type==ObjectType.Integer && d.type==ObjectType.Integer)
    {
      return new DynObject3(Integer.valueOf(((Integer)value).intValue()+((Integer)value).intValue()));
    }
    else
    {
      return new DynObject3(value.toString()+d.value.toString());
    }
  }
}

With the if-else logic I could use value.getClass()==Integer.class instead of storing the type but with more types I'd change this to use a switch statement and Java doesn't allow switch to use Classes.

Anyway... My question is what is the best way to go about something thike this?

Answer 1

What you are trying to do is called double dispatch. You want the method called to depend both on the runtime type of the object it's called on, and on the runtime type of its argument.

Java and other C derivatives support single dispatch only, which is why you need a kludge like the visitor pattern you used in option 1. This is the common way of implementing it. I would prefer this method because it uses no reflection. Furthermore, it allows you to keep each case in its own method, without needing a big "switchboard" method to do the dispatching.

Answer 2

I'd choose the second option, with the third, I'd better be using generics so you don't rely on that Enum. And with the first option you could be implementing methods for the rest of your life. Anyways you could use "instanceof" operator for Class matching.