[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: Perobject association (was: [aspectj-dev] pertypewithin() vs. pertype() association...)

I don't have time to reply to the whole thing right now, but:

<<<
P.S. - re:
> The set of join points at which the aspect can become 
> disassociated with a previously associated instance

There's no such dissociation.  It's Java; there's only 
garbage collection!  Once Foo.aspectOf(x) returns f
for x, it will always return f for x.  Whether any
advice will ever run again is a whole 'nuther question.
>>>

I know. I'm saying there *should be*. There's no (fundamental) reason why
Foo.aspectOf(x) can't start returning null again after some point.

Cheers,
Nick
--- Wes Isberg <wes@xxxxxxxxxxxxxx> wrote:

> Hi folks -
> 
> So instead of saying
> 
>   aspect A perthis(pc()) { 
>     before(This t) : somePointcut(t) {
>       ...
>     }
>   }
> 
> perhaps I should say, for each possible type,
> 
>   class PerThis {
> 
>      Object thisObject;
> 
>      private PerThis(Object o) { thisObject = o; }
> 
>      static aspect InitPerThis {
>          pointcut pc() : ...;
> 
>          Perthis getInitPerThis(Object o) { return o.perThis; }
> 
>          PerThis Object.perThis; // assuming you could...
> 
>          before(Object) : pc() && this(o) {
>              o.perThis = new PerThis(o);
>          }
>      }
> 
>      static aspect usePerThis {
>         // must use "perthis(..) && " with all advice
>         pointcut perthis(Object o) : this(o) && if(null != o.perThis);
> 
>         before(Object o) : perthis(o) && somePointcut() {
>           This t = (This) o;
>           ...
>         }
>      }
>   }
> 
> Or instead of 
> 
>   after() : pc() {...}
> 
> perhaps I should say
> 
>   Object around() : pc() { 
>      try {
>        return proceed();
>      } finally {
>        ...
>      }
>   }
> 
> or
> 
>   Proxy proxy = ...
> 
> I'm hearing that you can use perobject(..) to implement perthis(..) 
> using lower-level facilities, and get more control.  What would
> be convincing to hear about perobject(..):
> 
> - It's easier to use
> - It can't be done in AspectJ today
> - There are compelling use-cases for it
> 
> I agree that what's confusing about perthis(..) aspects is 
> realizing that the pointcut for each advice in the aspect 
> has an implicit "&& this(Type)".  
> 
> So where the programming guide says
> 
> ------------------------------
> The advice defined in A may then run at any join point where 
> the currently executing object has been associated with an 
> instance of A.
> ------------------------------
> 
> perhaps it should say 
> 
> ------------------------------
> The advice defined in A will only run at a join point ...
> ------------------------------
> 
> wrt IDE's, has anyone checked whether, when AJDT lists the code 
> advised by a particular piece of advice in a perthis aspect,
> it omits those outside the scope of this(Type)? Also, it might
> be nice to list, for a selected type, the per[this|target] 
> aspects that might be associated with it.
> 
> Wes
> 
> P.S. - re:
> > The set of join points at which the aspect can become 
> > disassociated with a previously associated instance
> 
> There's no such dissociation.  It's Java; there's only 
> garbage collection!  Once Foo.aspectOf(x) returns f
> for x, it will always return f for x.  Whether any
> advice will ever run again is a whole 'nuther question.
> 
> > ------------Original Message------------
> > From: Nicholas Lesiecki <ndlesiecki@xxxxxxxxx>
> > To: aspectj-dev@xxxxxxxxxxx
> > Date: Fri, Jan-28-2005 12:05 PM
> > Subject: Re: Perobject association (was: [aspectj-dev] pertypewithin()
> vs. pertype() association...)
> >
> > I like your proposal! Per(target/this) aspects currently behave rather 
> > oddly, because they mix the 4 concerns together. Separating them in the
> 
> > 
> > source would allow programmers to appreciate the four facets of 
> > behavior independently. I can see some other strengths that are 
> > currently missing:
> > 
> > strength 1) Association becomes advice which you can see in various 
> > cross-cutting views, step into more easily, etc.
> > (ok, I only thought of one)
> > 
> > However, there are some strengths to the current approach that your 
> > proposal does not address (and I say this only because I want to 
> > improve the proposal enough to get it accepted).
> > 
> > weakness 1) You could mismatch your association pointcut so that it 
> > never selected objects of the appropriate type (this is probably 
> > livable)
> > weakness 2) New types might match the association pointcut but not 
> > match the per clause type(pattern?)
> > weakness 3) the association advice would appear in an aspect whose 
> > other advice does not apply unless the aspect has already been 
> > associated, meaning that the association advice would need to be 
> > demarcated as "special" somehow
> > 
> > Cheers,
> > 
> > Nicholas Lesiecki
> > Software Craftsman, specializing in J2EE,
> > Agile Methods, and aspect-oriented programming
> > m: 520 591-1849
> > 
> > Books:
> > * Mastering AspectJ: http://tinyurl.com/66vf
> > * Java Tools for Extreme Programming: http://tinyurl.com/66vt
> > 
> > Articles on AspectJ:
> > * http://tinyurl.com/66vu and http://tinyurl.com/66vv
> > On Jan 28, 2005, at 10:48 AM, Eric Bodden wrote:
> > 
> > >
> > > -----BEGIN PGP SIGNED MESSAGE-----
> > > Hash: SHA1
> > >
> > > Nicholas Lesiecki wrote:
> > >> I've been thinking about perobject aspects in the shower and I've
> > >> decided that the current perobject aspects handle several concerns.
> > >> I think it's worth dissecting them for greater clarity of thinking:
> > >>
> > >>
> > > [...]
> > >> Ideally, the perobject aspects would give the programmer control
> > >> over all four concerns. Ok, that's all I have for now. Does anyone
> > >> want to pick up this ball and run it the next few yards?
> > >
> > > I completely agree. Also I think that perobject asoects would
> > > eventually yield a much clearer instantiation model. As Mira Mezini
> > > et al. showed in their CEASAR implementation, perobject aspects and
> > > percflow aspects actually completely suffice to implement all the
> > > instantiation models we have in AspectJ today:
> > >
> > > aspect A perthis(<Type>) {
> > >
> > > }
> > >
> > > is nothing more than:
> > >
> > > aspect perobject(<Type>) {
> > >
> > > 	after(<Type> t) : this(t) {
> > > 		associate(t);
> > > 	}
> > >
> > > }
> > >
> > > Same counts for pertarget *and* the newly proposed pertype as well,
> > > given that "pertype" actually means not more than "perobject", the
> > > "object" being the metaclass <Type>.class for any newly initialized
> > > type.
> > >
> > > Thus in the one could even completely loose perthis, pertarget,
> > > pertype, ... (not percflow, that'S really different), since they
> > > actually all are just special cases of object-associated aspects! In
> > > my personal opinion that would be much clearer than today. You can
> > > associate aspects with a set of objects or with a control flow. Not
> > > more.
> > >
> > > If one still wants for the sake of brevity a "perthis" aspect, one
> > > could probably find a means of inheriting instantiation advice (as
> > > above) down to subaspect types in a well-defined way. So you could
> > > have a PerThis super-aspect as above and all subaspects *may* have
> 
=== message truncated ===