Re: [aspectj-users] Interface Constructors

[Jim Hugunin <lists@xxxxxxxxxxx>]

Matthew Webster wrote:
> If I use the pointcut:
>
>       execution (*.new(..));
>
> I will advise the execution of all constructors including interface types.
> Can I exclude interface types without having to enumerate them? Using the

The only way I can think to do this in AspectJ-1.1 is ugly:

    execution(new(..)) &&
if(!thisJoinPointStaticPart.getSignature().getDeclaringType().isInterface())

You could imagine an AspectJ compiler completely optimizing this kind of 
test away; however, the current ajc-1.1.1 will require a couple of cheap 
methods calls here.

One outstanding language request is to add interface, aspect, and class 
to the type specifier language.  This would let you write something like:
  execution( (!interface).new(..) )

You might want to bring this up when we begin the 1.2 language design 
discusions.

> following scope (interface types do not descend from Object):
>
>        within(java.lang.Object+)
>
> does not work.

Whether or not interface types are a subtype of java.lang.Object is 
subject to some debate.  For example, the following is legal code:

  Runnable r = ...;
  String s = r.toString();

This suggests that the Runnable interface is behaving in a fairly 
obvious way as if it is a subtype of Object.  The unofficial Java Spec 
Report includes a detailed discussion of this issue at:
  http://www.ergnosis.com/java-spec-report/java-language/jls-9.2.html

AspectJ treats interfaces as subtypes of Object.  Given the ability to 
add instance fields and concrete methods to interfaces through intertype 
declarations, interfaces in AspectJ behave even more like actual 
subtypes of Object than they do in pure Java.

-Jim