[otj-users] Binding inheritance and covariance

[Stephan Herrmann]

Hi,


> Will the compiler make sure that 
> the callin-method will /really/ return the result of the base-call
Yes! Either the base-call result ;-)
OR null :(
But no other value is possible.

> and not  something else that would be fine according to "normal" generics rules?
> E.g.  from looking at
> 
> [callin] <E extends List> E getList() {
> 
> }
> 
> it would appear perfectly fine to return any subclass/implementation of List.
appearances can be deceiving ;-)

Think of who is defining the actual replacement for E: it's the caller.
Any invocation may choose a different E like in
	LinkedList l = getList(); // legal: E == LinkedList
	ArrayList l = getList(); // legal: E == ArrayList
just not:
	String s = getList(); // illegal: can't find an E that extends List AND is conform to String
In the first, legal variant getList() must return a LinkedList! 
The contract of this method is: 
  "I'll return a value,
  the type of which you're free to CHOOSE from all subtypes of List
  AFTER I've implemented the method."
Not very useful in normal settings, ;-)
but for replace callins it looks neat, doesn't it?

(Type theory would say 
"List is an upper bound for E but the exact E is unknown within getList()") 

cheers,
Stephan

PS: I just tried to cheat the java compiler:
	<E extends List> E getList() {
		return (E)(new ArrayList());
		// telling the compiler: please insert a cast, 
		// I'll tell you at runtime which type to cast to!
	}
The compiler reported no(!) error but answered:
  "Note: Test.java uses unchecked or unsafe operations.
   Note: Recompile with -Xlint:unchecked for details."
(The byte code did of course NOT contain a cast)
Hhhmmm ;-)