Description John Maraist 2004-10-13 19:55:56 EDT

System: 
 - Eclipse 3.1M2  
 - Mandrake 10.0 
 
This code exhibits inconsistant behavior on the Sun and Eclipse 
compilers. 
 
  interface AA<W, Z extends AA<W, Z>> { 
    public boolean m(AA<W, ?> that); 
    public Z z(); 
    public boolean b(); 
  } 
 
  abstract class BB<U, V extends AA<U, V>> implements AA<U,V> { 
    public boolean m(AA<U, ?> wht) { return wht.z().b(); } 
  } 
 
Note that 
 
 - The formal type parameter Z of AA is bounded 
   recursively by AA, 
 
 - But the corresponding type argument of the 
   declared type of formal parameter wht of 
   method m's implementation is a wildcard 
   with no bound. 
 
My understanding of unbounded wildcards is as an implicit 
bound of Object, which in this case means that in the 
body of m the expression "wht.z()" should be assumed to 
be Object, which does not have a method b().  The Eclipse 
compiler seems to reject this code for this reason:"'Bound 
mismatch: the type ? is not a valid substitute for the 
bounded parameter <Z extends AA<W,Z>> of type AA<W,Z>". 
 
On the other hand, any validly constructed actual 
parameter wht of type AA<U, ?> would certainly have 
been constructed respecting the bounds on Z.  Perhaps 
this is why the Sun JDK javac accepts this code without 
error? 
 
I do not know which is the correct behavior, and am 
posting this bug to both the JDK and Eclipse bug 
pages. 
 
(Interface AA, by itself, compiles without error 
on both compilers.)