Comment 1 Helen Beeken 2005-10-10 04:31:42 EDT

Can you provide more information about what you were doing when you saw this?
The aspect you're writing etc.

Comment 2 hesse 2005-10-11 08:05:47 EDT

Here is a minimal sample:

package aj;

public class SomeClass {

    public void doSomething() {
    }

    public static void main(String... args) {
        new SomeClass().doSomething();
    }

}

package aj;

public aspect MyAspect {

    public interface MyWorld {
    }

    declare parents : SomeClass implements MyWorld;

    pointcut doSomethingInMyWorld(MyWorld myWorld) :
            execution(void SomeClass.doSomething()) &&
            this(myWorld);

    void around(MyWorld myWorld) : doSomethingInMyWorld(myWorld) {
        System.out.println("this works");
    }

}

package aj;

public aspect DoesntCompile {

    public interface MyWorld {
    }

    declare parents : SomeClass implements MyWorld;

    pointcut doSomething(SomeClass someClass) :
            execution(void SomeClass.doSomething()) &&
            this(someClass);

    pointcut doSomethingInMyWorld(MyWorld myWorld) : doSomething(myWorld);

    void around(MyWorld myWorld) : doSomethingInMyWorld(myWorld) {
    }

}

Comment 3 Helen Beeken 2005-10-11 08:31:00 EDT

Thanks for the simplified testcase - I have been able to recreate the problem.

This is a compiler bug so reassigning to aspectj.

Comment 4 Helen Beeken 2005-10-11 08:45:09 EDT

Created attachment 28111 [details]
patch containing supplied testcase

This patch is a patch to be applied to the tests project and contains the
supplied testcase written to fit in with the aspectj testcase structure.

Comment 5 Andrew Clement 2005-10-20 10:12:41 EDT

I've applied Helens patch then worked on reducing the test program as much as
possible, this fails:

class SomeClass  {
    public void doSomething() { }
}

aspect DoesntCompile {

    declare parents : SomeClass implements Serializable;

    pointcut doSomething(SomeClass someClass) :
            execution(void SomeClass.doSomething()) &&
            this(someClass);

    void around(Serializable myWorld) : doSomething(myWorld) { }
}

If I directly apply Serializable (i.e. not use declare parents) then it works.
If I change the around advice to take a 'SomeClass' then it works.

Comment 6 Andrew Clement 2005-10-20 10:41:00 EDT

Ok ... the problem here is the shadow munger is being concretized before the
type munger that adds the parent has had a chance to take effect.

The message 'illegal change to pointcut' is hopeless.
Before producing the message it is actually looking at the pointcut and checking
that what you are binding 'Serializable' is the same or a supertype of the value
in the pointcut 'SomeClass'.  What you are doing is allowed because all
SomeClass' are Serializable - however, because the declare parents hasnt
occurred yet, we don't know SomeClass is Serializable.

Thats why it works if you either:
a) hardcode SomeClass implements Serializable and dont use declare parents.
b) change the advice to 'void around(SomeClass myWorld)'

what on earth I can do about it ...  hmmmmmmmmm

Comment 7 Andrew Clement 2005-10-20 12:23:22 EDT

The actual problem of using the wrong type in the pointcut is checked in two places.

1. ReferencePointcut.resolveBindings
   produces something vaguely helpful like the message: 
     incompatible type, expected SomeClass found
BindingTypePattern(java.io.Serializable, 0)

2. ReferencePointcut.concretize1
   produces the BCException:
   "illegal change to pointcut declaration:
doSomething(BindingTypePattern(java.io.Serializable, 0))"

I propose to remove the check for illegal change to pointcut decl as it:
1. is done too early
2. is duplicating something checked later
3. is an exception for something that should be an error
4. is extremely unhelpful

Comment 8 Andrew Clement 2005-10-21 03:09:10 EDT

Fix checked in - waiting on build.

Comment 9 Andrew Clement 2005-10-21 06:46:08 EDT

fix available.