- From: Kohei Honda <kohei@dcs.qmul.ac.uk>
- Date: Tue, 21 Feb 2006 01:54:12 +0000
- To: 'WS-Choreography List' # <public-ws-chor@w3.org>
- CC: Steve Ross-Talbot <steve@pi4tech.com>, Gary Brown <gary@pi4tech.com>, Nobuko Yoshida <yoshida@doc.ic.ac.uk>, Marco Carbone <carbonem@dcs.qmul.ac.uk>, Kohei Honda <kohei@dcs.qmul.ac.uk>
teve Ross-Talbot wrote:
>
> Check out:
>
> http://polarlake.typepad.com/warren/
>
>
Quite an interesting observation. Two points:
(1) Mr. Buckley's view focusses on a run-time view (run-time means
what happens in situ, in live) of CDL, and says fixing description is
useful for controlling the run-time. A charming understanding of
dynamism between description and what is realised.
(2) And this is far from being inconsistent with the use of static
abstraction such as types, logics and others:
--- First, it helps you to write down a reference description which
you are sure means what you want.
--- Second, it enhances clarity of description itself, and gives
us the idea what it is like (as dtd and schemas do).
--- Third, it can give useful info to a run-time: a monitor can
rely on extra information for controlling interaction. A simple
one: you should send two integers rather than one boolean.
A more complex: You never give low quote to more than 10
customers. Analogous to the use of run-time types in JVM.
Well I realise there are more.
--- Fourth, you can write a meta-specification. You start from a fixed
choreo, which, depending on what is observed at run-time, can be
updated and replaced at run-time. A new choreography can even be
generated at run time. Perhaps like some modern dance performance?
(Pina Bausch's)? Or like writing scripts for the next scenes as a
movie is being screened, as some directors seem to be doing?
(close Mr. Buckley's feedback loop).
--- Fifth, your static abstraction is even more important when we
allow such dynamic configuration. In fact it will be near
impossible to control such evolvable behaviour without a basic notion
of typing. If a director can change dancing in situ, and if that
director
can go completely mad (at run-time, say), how can you be sure
everything works all right?
Surely there will be more, but let's finish here. As an academic note,
process algebras such as pi-calculus and CSP would offer a basic tool
for studying such complex run-time behaviour.
kohei
Received on Tuesday, 21 February 2006 01:53:53 UTC