Description Bhaskar Maddala 2007-12-09 17:00:46 EST

I have an abstract aspect, where I define my point cuts, and 2 derived concrete aspects that use the pointcuts to implement advice. On building in eclipse and running findbugs I get the following bad practice bug

    Superclass uses subclass during initialization.

  I looked at the byte code and I can see that the abstract aspect had references to the concrete aspects. The reason was an overly broad definition of pertypewithin. Below is a toy code sample that results in this findbugs error

package bar;
public abstract aspect AbstractAspect pertypewithin(bar..*) {
    pointcut loginitialization() : staticinitialization(bar..*);
    pointcut loggedMethods() : (within(bar..*)) && execution(* *.*(..));
}

package bar;
public aspect ConcreteAspect extends AbstractAspect {
    after() : loginitialization() {}
    before() : loggedMethods() {}
    after() : loggedMethods() {}
}

The AbstractAspect satisfies the pertypewithin scope and is advised by ConcreteAspect.after():loginitialization

Having a concrete aspect advice a abstract aspect that it inherits seems to be an anti-pattern (at least when using pertypewithin), which is really the reason for FindBugs to be complaining.

Requested Enhancement : Aspectj compiler should report an appropriate warning indicating such a scenario.

Attachment contains a small eclipse project to reproduce the issue.