I’m trying to build an SDK using an adapter pattern. Here’s what I’ve got so far:
interface Adapter<T> {
void doWork(WorkUnit<T> unit);
Class<T> getT();
}
class WorkUnit<T> {
public int getId() { ... }
public T getExtras() { ... }
}
class OldWorkUnit {
public <T> void setExtra(T data) { /* Store data in Map<Class, Object> */ }
public <T> WorkUnit<T> toNewWorkUnit(Adapter<T> adapter) { /* Map.get(adapter.getT()) */ }
}
There’s a good amount of generics in there, but I can’t know T at compile time, and there may be multiple Adapters, all with different types of T. This is meant to be exposed to third parties, so I also need as little implementation as possible in the interface implementation, and it has to be an interface (no abstract class).
Now I want to take this and call doWork with a WorkUnit. My first pass at the code looks like this:
class FooAdapter implements Adapter<FooWorkUnit> {
...
}
OldWorkUnit w = new OldWorkUnit();
w.setExtra(new FooWorkUnit());
Adapter<?> a = new FooAdapter();
WorkUnit<?> unit = w.toNewWorkUnit(a);
a.doWork(unit);
Uh oh, that doesn’t compile:
The method doWork(WorkUnit<capture#2-of ?>) in the type Adapter<capture#2-of ?>
is not applicable for the arguments (WorkUnit<capture#4-of ?>)
I know that the WorkUnit generic argument is the same type as the Adapter.doWork() generic argument, but without knowing the type I can’t cast it appropriately.
So is there a way to work my way through this?
Why do you wildcard the template class here?:
If you do this:
Doesn’t that preserve what you need?
Now the compiler knows that
FooWorkUnitis the common denominator, so to speak.Edit: OK point taken about the runtime variability. How about strongly-typing the method that does the work, so that the wildcards are eliminated, but still consistent: