Oops.  I think we just didn't think of this.  It would be a shame for both to be illegal, as that would force folks to use raw types.  I think javac's behavior is what we intended to specify.  Section 4.5.2 failed to take into consideration that sometimes generic types are used in a context where performing a capture conversion is inappropriate.  The definition of a "Reifiable Type" failed to consider inner classes too.  I suspect there are other places in the spec that nested types are not properly handled. 

-Neal

On 12/7/06, ... wrote:

javac accepts this case:
        
        
        class X<T> {
                class Y<T> {}
        }
        
        
        class B {
                void f() {
                        X<B>.Y<B> xy = null;
                        boolean b2 = (xy instanceof X<?>.Y<?>);
                }
        }   
        
        
        But rejects this one:
        
        
        class X<T> {
                class Y<T> {}
        }
        
        
        class B {
                void f() {
                        X<B>.Y<B> xy = null;
                        boolean b2 = (xy instanceof X<B>.Y<?>);
                }
        } 
        
        
        with the error:
        
        
        t6.java:14: illegal generic type for instanceof
                        boolean b2 = (xy instanceof X<B>.Y<?>);
                                                          ^ 
        
        
        This seems to be treating X<?>.Y<?> as a type with only unbounded wildcards, but that is not how I read the JLS.
        
        
        - 4.5.2 says that the members of wildcarded types are undefined, so we know that the parent class of Y<?> is not a wildcard.
        
        
        - As far as I can tell, the X<?> in that context is replaced with its capture conversion.
        
        
        So I would expect an error on both of these cases.  What am I missing?
This comment was invalid and has been deleted