Re: [aspectj-users] Static conditional poincuts

[Alex Villazon <alex.villazon@xxxxxxxxxxx>]

On Apr 21, 2009, at 5:27 PM, Andrew Eisenberg wrote:

> This is currently not a feature on our list.  However, we would
> appreciate if you could raise an enhancement request.  I do notice
> that scope is built on top of the Aspect Bench Compiler (ABC), not
> AJC.  The focus of ABC is extensibility and experimentation, while AJC
> focuses on efficiency and stability in a production environment.  So,
> it may not be a simple port of the mechanism.
>
> Do you have a specific aspect that is suffering from a large runtime
> overhead?  If so, we might be able to look at it and suggest
> alternatives that are more efficient.

Hi,
   We want to advise method calls where the invoked method on the  
target, is defined on an interface and not on possible concrete  
subclasses.
   So if we have an interface "I" and possible classes implementing  
the interface "I", we want to be sure that the invoked method was  
declared in "I" or sub interfaces, but not in a concrete class....   
This information should be available statically (if i'm correct..),  
but currently we need to use dynamic join points to gather this  
info..  as follows:

aspect A {
  before(I o) : call(*.*(..)) && target(o) && !within(A) {
       
if(thisJoinPoint.getSignature().getDeclaringType().isInterface()) {
       // do something.. with o
     }
  }
}

  We could like to put the condition, so that the weaver does not  
weave the aspect for our particular case.. (i.e. somehow describing it  
as a pointcut) but probably this is not even possible with SCoPE..

   The problem is that our application needs to be very fast, and  
computing dyn joinpoints is already introducing too much perturbation..

   Probably there is a nicer way to do that...

Alex



> --a
>
> On Tue, Apr 21, 2009 at 6:56 AM, Alex Villazon
> <alex.villazon@xxxxxxxxxxx> wrote:
> > Hi,
> >    Is there a mechanism similar to SCoPE
> > http://www.graco.c.u-tokyo.ac.jp/ppp/index.php?Projects%2Fscope
> > implemented (or planned) for AspectJ?
> >
> >   This feature seems to be very useful to avoid heavy runtime  
> > tests...
> >
> >  Any comments?
> >
> > Alex
> >
> > _______________________________________________
> > aspectj-users mailing list
> > aspectj-users@xxxxxxxxxxx
> > https://dev.eclipse.org/mailman/listinfo/aspectj-users
> _______________________________________________
> aspectj-users mailing list
> aspectj-users@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/aspectj-users