Re: [abc-dev] Extension

Hi again,

Sorry for any confusion, I figured it out. I just added
java.util.ArrayListto the basic classes and load it at SIGNATURES
level in a static block
inside my AbcExtension class. Seems to work fine.

Albert

On 4/1/06, Albert Bachand <albertb@gmail.com> wrote:
>
> Hi Pavel,
>
> Thanks for your help. I'm almost done with the extension. I just have one
> last problem. It might be more related to Soot than Abc so sorry in advance
> if that's the wrong mailing list. So, the problem is in my extension of
> AspectCodeGen. I want to add a field of type java.util.ArrayList to
> associable aspects. I've tried several different ways of getting a hold of a
> SootClass for java.util.ArrayList but nothing seems to work, I always get
> an error about the resolving level being HIERARCHY instead of SIGNATURES. I
> read on the sable wiki that a way around that would be to add
> java.util.ArrayList to the basic classes before calling abc.main.Main but
> the problem is that this is the very first method called by abc so I'm not
> sure how to go about it. Is there an easy way in Abc to load some classes at
> a higher resolving level?
>
> Thanks!
>
> Albert
>
>
> On 3/23/06, Pavel Avgustinov <pavel.avgustinov@magdalen.oxford.ac.uk >
> wrote:
> >
> > Hi,
> >
> > > My first question is where should I generate the code that creates the
> >
> > > List field inside the aspect? (Every associable aspect needs a static
> > > List that keeps track of associated instances.) Should it go in a
> > > ContextVisitor that visits all nodes until it finds an aspect that has
> >
> > > been marked as associable and then generate the code in enter()?
> >
> > Essentially, you're trying to introduce new static members onto the
> > aspect class. A very similar case arises in pure AspectJ -- think of the
> >
> > static aspectOf() method that is added onto aspect classes.
> >
> > Since your Aspect.associate() and Aspect.disassociate() methods need to
> > be visible to the base code, you need to introduce them (or at least
> > stub methods that you fill in later) before method calls are resolved.
> > The natural place for this is the AspectDecl_c class. Take a look at how
> > it handles the aspectOf() method -- you'll need to do something very
> > similar. Override AspectDecl_c in your extension, and override the
> > factory that instantiates it so that it creates your updated class.
> >
> > The bodies of those methods can be generated whenever convenient; it
> > might make sense to stick to the aspectOf() pattern. The body of that
> > method is filled in in src/abc/weaving/weaver/AspectCodeGen.java, and
> > happens on Jimple using Soot.
> >
> > > My second question is about the association Residue: I started writing
> > > an IsAssociated residue that can be used for both 'this' and 'target'
> > > association but I'm not sure how to refer back to the aspect from the
> > > codeGen method. I would need to generate code that checks whether the
> > > instance is in the List of that aspect before proceeding with the
> > adivce.
> >
> > How you actually implement the isAssociated stuff is of course your
> > choice, but note that if you introduce a method on the aspect class
> > called "isAssociated(Object)", you can reduce the checks to if()
> > pointcuts. Thus, you can rewrite your custom pointcuts to if() in the
> > frontend.
> >
> > Note that if you want to do this globally, there is a mechanism for
> > adding a new conjunct to every single pointcut. Thus, you could make all
> > pointcuts respect aspect association. Take a look at
> > src/abc/om/AbcExtension.java and its use of residueConjuncts.
> >
> > We gave a tutorial on extending abc at AOSD yesterday. The slides should
> > appear on http://aspectbench.org sometime soon, in the meantime you can
> > access a copy at http://musketeer.comlab.ox.ac.uk/~pavel/slides.pdf<http://musketeer.comlab.ox.ac.uk/%7Epavel/slides.pdf>--
> > there are some further explanations about residueConjuncts there, if
> > nothing else.
> >
> > Apologies for the delay in this reply. Most of us are currently visiting
> > the AOSD conference, so our access to email is somewhat sporadic. Hope
> > this helps, feel free to ask for more information if you need it.
> >
> > Best wishes,
> > Pavel
> >
> >
> >
>
Received on Sat Apr 01 18:37:31 2006