[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
RE: [aspectj-users] Logging enabled check idea

Oh... and forgot to mention: Of course you also get into all sorts of
trouble for example when using around advice, which get the *evaluated*
arguments passed as arguments to the advice body. So even if you had
more powerful compilers that could find out all objects possibly
reaching the joinpoint for you, it would not much sense semantically to
intercept them with around.

Eric 

> -----Original Message-----
> From: aspectj-users-bounces@xxxxxxxxxxx [mailto:aspectj-users-
> bounces@xxxxxxxxxxx] On Behalf Of Eric Bodden
> Sent: Tuesday, June 06, 2006 9:29 PM
> To: aspectj-users@xxxxxxxxxxx
> Subject: RE: [aspectj-users] Logging enabled check idea
> 
> Well, I am not from the ajc team, but I know Java bytecode and I know
> that from the bytecode you cannot tell where the evaluation of an
> argument starts. The "two styles" that were presented indeed share
> virtually the same byte code representation. What you need are
dataflow
> pointcuts or something like that. PQL for example can handle similar
> problems already: http://pql.sourceforge.net/. But generally you will
> still have to wait a few months or so until practical solutions for
> such
> problems are available. People are working on it. That's what I know
> for
> sure. I am not sure however, if any of such things are ever gonna make
> it into the ajc compiler. Problably not (because that can have quite
an
> impact on compilation time).
> 
> Eric
> 
> > No, unfortunately not; please disassemble your woven class and see
> for
> > yourself. I'm actually surprised that the around advice is not
> wrapped
> > around the argument construction as well. Could someone from the
> > AspectJ
> > team please comment on why this is so, and whether this might be
> added
> > in
> > the future? Considering that this.foo(this.bar(this.baz(z))) can be
> > pretty
> > expensive depending on bar and baz, wrapping around them as well
> would
> > be
> > really really nice and more in line what one (meaning me :) would
> > expect.
> 
> _______________________________________________
> aspectj-users mailing list
> aspectj-users@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/aspectj-users