Re: [abc] nice paper! :)

On Tue, 28 Sep 2004, Prof. Laurie HENDREN wrote:

> I have read through the paper and made only very minor changes.
> I am very impressed with the paper. It is very professionaly done,
> with lots of related work (thanks Oege for all your research there!...
> I think it helps it be an AOSD paper rather than just a compiler paper)

Erm. There are errors in that, like the fact that I missed the
public distribution of josh. Thanks to Ondrej for noting that,
and a little hint to everyone else to pick a random reference
and check that what I said is actually true!

>
> I don't see any major changes left to be made. Someone needs to
> design our "abc technical report latex format" so we can make
> the technical report version as soon as we submit the paper and
> then put it on the aspectbench.org web site and send copies to
> our friends. I would love to do it, but I seem to have a queue
> of "other papers" I am supposed to be working on.

right, that would be nice.

> I guess we also need an official "last reader" who goes over
> it for a last time looking for typos. I am afraid I'm not so good
> at that any more .... I seem to have gained speed at the expense
> of accuracy ....

Damien is doing a pass tonight.

Personally I think it is worthwhile to spend tomorrow on making
a few more improvements:
 a) check references (as suggested above, but also just the
    details of publication venue etc)
 b) look for further typos, obscurities etc.
 c) run the whole thing through a spell checker

A good technique (and I'm not joking here) is to read the
paper backwards at this stage. If you read it forwards, it
all goes "yeah, seen that before, obvious..." and you read what
you think, not what's written.

> Anyway, a job well done. I hope the "young guys" found this an
> interesting experience. Now we have to shift our focus to our
> release (I go to CASCON next Wednesday, and would love announce
> the real release then...) and we also have to work on writing the
> CC paper and doing the analysis stuff for the PLDI paper.

I'm sorry, but I'm firmly of the opinion we're not ready for
a release. Quite understandably, after the IBM visit we have
slowed down a lot. Before we release, we should

a) compile abc itself (so it is really possible to use aspects
   for extension if people want to do that - it's not just
   a fun meta-circular thing).

b) compile atrack, or at least fix all the problems that
   Ondrej has found

c) pepper the source with javadoc. It is very sparse at
   the moment, and I do not believe the source is of
   use outside our team in its present form.

While I'm really happy with how far we've got, it is a fact
that, even for a simple program like my ants viewer, there
were several things that needed fixing in abc before it could
be compiled. No pathological corner cases: a real program.
We have to have more confidence that abc won't break on
the first sizable example people try.

Now of course we *could* race to bring out a release by CASCON,
and if we're lucky we'll get real users and umpteen bug reports
to fix that same week. Now we have to ask ourselves whether
that is the best use of a team that is going to drop in manpower
a lot as of next week, when term starts and several of the
Oxford gang have to go back (at least part-time :-))
to being undergrads.

I think it is far more productive to get on quietly with
a-c) above, the CC paper, and most importantly with re-weaving.
That was the crucial idea that started all this in January;
it got lost in the development fever but fortunately Ondrej
insisted on getting back to it. We have to get it ready for
PLDI. The realisation of re-weaving will underlie lots of our
future plans, and it will be a big boost to our chances of
continuing the project if we have demonstrated it works. I
don't think a release would have the same impact.

Sorry for a long email. Obviously it's important we make the
right strategic step now, with the very limited resources we've
got.

-Oege
Received on Tue Sep 28 23:04:52 2004