Bounds

In the process of discussing generics, we've already had to mention bounds a few times. A bound is a constraint on the type of a type parameter. Bounds use the extends keyword and some new syntax to limit the parameter types that may be applied to a generic type. In the case of a generic class, the bounds simply limit the type that may be supplied to instantiate it. A type variable may extend a class or interface type, meaning that its instantiation must be of that type or a subtype:

 class EmployeeList< T extends Employee > { ... }


Here, we made a generic EmployeeList type that can be instantiated only with Employee types. We could further require that the Employee type implement one or more interfaces using the special & syntax:

 class EmployeeList< T extends Employee & Ranked & Printable > { ... }


The order of the & interface bounds is not significant, but only one class type can be specified and if there is one, it must come first. When a type has no specific bounds, the bound extends Object is implicit. By applying bounds to our type, we not only limit the instantiations of the generic class but we make the type arguments more useful. Now that we know our type must extend some type or implement some set of interfaces, we can use variables and arguments declared with T by those other type names. Here is a somewhat contrived extension of our previous example:

 class EmployeeList< T extends Employee & Ranked & Printable >
 {
 Ranked ranking;
 List<Printable> printList = new ArrayList<Printable>( );
 public void addEmployee( T employee ) {
 this.ranking = employee; // T as Ranked
 printList.add( employee ); // T as Printable
 }
 }


Type variables can also refer to other type variables within the type declaration:

 class Foo <A, B extends A> { ... }


We'll see a particularly vicious example of this later when we talk about the definition of the Enum class. We'll also see a more convenient technique for declaring how individual elements of a generic class relate to the parameter type when we cover wildcards in the next section.

Erasure and Bounds (Working with Legacy Code)

We mentioned earlier in our discussion of erasure that the resulting type used in place of the type parameter in the raw type for the generic class is the bound of the type variable. Specifically, we have seen many generics with no explicit bounds that defaulted to a bound of type Object. We also showed a quick example of a type that imposed a bound of extends Date and said that the type of its methods would be Date instead of Object. We can now be a little more specific. The type after erasure used for the parameter type of a generic class is the leftmost bound. That is, the first bound specified after extends becomes the type used in the erasure. This implies that if the type extends a class type, it is always the erased type because it must always come first. But if the type extends only interface types, the choice is up to us. This fine point is important for backward compatibility with nongeneric code. Often when creating generic versions of nongeneric APIs, we have the opportunity to "tighten up" the specification a bit. Being aware of the leftmost bound gives us a way to explicitly control the type of the erased class. For example, suppose we create a generic List class that we only want instantiated on Listable objects, but we'd prefer not to change the API of our old List class, which accepted Object type elements. Our initial attempt:

 class List< E extends Listable > { ... }


produces a raw type that accepts only Listable. However, we can insert a somewhat gratuitous additional type, Object, as the leftmost bound in order to get back our old API, without changing the new generic bounds:

 class List< E extends Object & Listable > { ... }


Inserting Object doesn't change the actual bounds of the generic class but does change the erased signature.

Java ScreenShot
Comments