Bug 73971

Summary: [1.5] Ambiguous method error with overloaded parameterized methods
Product: [Eclipse Project] JDT Reporter: Markus Keller <markus.kell.r>
Component: CoreAssignee: Philipe Mulet <philippe_mulet>
Status: VERIFIED FIXED QA Contact:
Severity: normal    
Priority: P3 CC: kent_johnson
Version: 3.0   
Target Milestone: 3.1 M4   
Hardware: PC   
OS: Windows XP   
Whiteboard:

Description Markus Keller CLA 2004-09-15 05:58:21 EDT 
I200409140800

The second and third call to m(..) yield wrong errors. After erasure, there's no
ambiguity, since the three methods m(..) are distinct. Maybe connected to bug 73970.

public class X {
    static <E extends A> void m(E e) { System.out.println("A:"+e.getClass()); }
    static <E extends B> void m(E e) { System.out.println("B:"+e.getClass()); }
    static <E extends C> void m(E e) { System.out.println("C:"+e.getClass()); }

    public static void main(String[] args) {
        m(new A());
        m(new B()); //error: "The method m(B) is ambiguous for the type X"
        m(new C()); //error: "The method m(C) is ambiguous for the type X"
    }
}

class A {}
class B extends A {}
class C extends A {}

javac 1.5.0-beta3-b62 compiles it and prints:
A:class q.A
B:class q.B
C:class q.C
Comment 1 Kent Johnson CLA 2004-11-19 15:48:29 EST 
Philippe - This is a bug in ParameterizedGenericMethodBinding 
computeCompatibleMethod()

When the message send is m(new B()) and we substitute for the method:
   <E extends A> void m(E e)
we create a ParameterizedGenericMethodBinding:
   m(B)
instead of:
   m(A)

So we get 2 ParameterizedGenericMethodBindings with B as their substituted 
parameter instead of one with A & one with B.

Added MethodVerifyTest.test016()
Comment 2 Philipe Mulet CLA 2004-11-21 09:32:27 EST 
Type inference is not meant to substitute parameters with their bounds, but
rather  with type inferred from argument types.

In this case, both (A)m(E) and (B)m(E) are substitutable into: m(B), which
causes the ambiguity in the end.

I suspect the tiebreak should use original methods rather than substituted ones,
or at least have original bounds play.
Comment 3 Philipe Mulet CLA 2004-11-22 07:02:25 EST 
Tuned tiebreak. Added regression tests: GenericTypeTest#test425--426.
Fixed
Comment 4 Olivier Thomann CLA 2004-12-14 15:40:11 EST 
Verified in 200412140800