Hi Rachid:

thank you for sending me your paper. It is a hot cake. I was spreading
it at a DARPA PI meeting and people were very interested!

While it is good to have a discussion about the usefulness
of AOP, your paper gives the impression to some people that AOP does not work.
Maybe that is the message you want to send? 

You say in the conclusion: "these features should 
be reserved for smart programmers who
have an advanced sense of the risk". This gives the impression
that AspectJ and DJ etc. are only for advanced programmers.
I have used both tools in my classes and the students were
far from being advanced programmers, yet they could use the tools
effectively.

It is true that AOP has its dangers. Whenever you change the base,
you have to make sure that ALL aspects still work as intended
especially if the aspects involve wild card facilities,
like *, + and traversal specifications.
See:
http://www.ccs.neu.edu/research/demeter/papers/demystified/
for a critique by Mitch Wand written about 4 years ago.
Testing after a base change is very important.

You say in your last sentence:
"... what can possibly be safely aspectized is probably what is not part of
the object semantics". This sentence throws the
Demeter work in the wast basket because we are strong
at aspectizing what is part of the object semantics.

Here is the reaction of Doug Orleans and I agree with him:

Somewhat.  I agree that aspects are often coupled to the base program
(or more generally, each other) in that the modification of one can
require the modification of the other.  But I don't think that AOP
claims (or aims) to completely decouple the concerns of a program.  I
think it's somewhat of a straw man argument to use "aspectize" to mean
a complete semantic decoupling, and then claim that concurrency can't
be aspectized.  I think the goal of AOP (and AP, of course) is to
loosen the coupling as much as is reasonable, and the authors do
demonstrate some aspect definitions that help to loosen the coupling
of concurrency and the base program.  It seems strange that there's no
mention of COOL, though.

Also, I'd be interested in reading reference 27: "AOP = SMP (Structured
Macro Programming)", but I can't find it on the net anywhere.  It
looks like the author, Guerrouai, was on a panel with you, Karl, at
ECOOP 2000-- is this paper in the proceedings?  Do you recall what the
gist of his claim is?

--Doug

Greetings,
-- Karl