A question about a method with generic bounded type parameter

Question

Hello everybody,
Sorry, if I ask a question which seems basic, but
I'm new to generic types. My problem is about a method
with a bounded type parameter. Consider the following
situation:
abstract class A{     }
class B extends A{     }
abstract class C
     public abstract <T extends A>  T  someMethod();
public class Test extends C
     public <T extends A>  T  someMethod()
          return new B();
}What I want to do inside the method someMethod in the class Test, is to
return an instance of the class B.
Normally, I'm supposed to be able to do that, because an instance of
B is also an instance of A (because B extends A).
However I cannot compile this program, and here is the error message:
Test.java:16: incompatible types
found   : B
required: T
                return new B();
                       ^
1 errorany idea?
many thanks,

Answer 1

Hello again,
First of all, thank you very much for all the answers. After I posted the comment, I worked on the program
and I understood that in fact, as spoon_ says the only returned value can be null.
I'm agree that I asked you a very strange (and a bit stupid) question. Actually, during recent months,
I have been working with cryptography API Core in Java. I understood that there are classes and
interfaces for defining keys and key factories specification, such as KeySpec (interface) and
KeyFactorySpi (abstract class). I wanted to have some experience with these classes in order to
understand them better. So I created a class implementing the interface KeySpec, following by a
corresponding Key subclass (with some XOR algorithm that I defined myself) and everything was
compiled (JDK 1.6) and worked perfectly. Except that, when I wanted to implement a factory spi
for my classes, I saw for the first time this strange method header:
protected abstract <T extends KeySpec> T engineGetKeySpec
(Key key, Class<T> keySpec) throws InvalidKeySpecExceptionThat's why yesterday, I gave you a similar example with the classes A, B, ...
in order to not to open a complicated security discussion but just to explain the ambiguous
part for me, that is, the use of T generic parameter.
The abstract class KeyFactorySpi was defined by Sun Microsystem, in order to give to security
providers, the possibility to implement cryptography services and algorithms according to a given
RFC (or whatever technical document). The methods in this class are invoked inside the
KeyFactory class (If you have installed the JDK sources provided by Sun, You can
verify this, by looking the source code of the KeyFactory class.) So here the T parameter is a
key specification, that is, a class that implements the interface KeySpec and this class is often
defined by the provider and not Sun.
stefan.schulz wrote:
>
If you define the method to return some bound T that extends A, you cannot
return a B, because T would be declared externally at invocation time.
The definition of T as is does not make sense at all.>
He is absolutely right about that, but the problem is, as I said, here we are
talking about the implementation and not the invocation. The implementation is done
by the provider whereas the invocation is done by Sun in the class KeyFactory.
So there are completely separated.
Therefore I wonder, how a provider can finally impelment this method??
Besides, dannyyates wrote
>
Find whoever wrote the signature and shoot them. Then rewrite their code.
Actually, before you shoot them, ask them what they were trying to achieve that
is different from my first suggestion!
>
As I said, I didn't choose this method header and I'm completely agree
with your suggestion, the following method header will do the job very well
protected abstract KeySpec engineGetKeySpec (Key key, KeySpec key_spec)
throws InvalidKeySpecException and personally I don't see any interest in using a generic bounded parameter T
in this method header definition.
Once agin, thanks a lot for the answers.