[Modeling] Behavior in Interaction Diagrams

Radovan Cervenka rce@whitestein.com
Fri, 21 Mar 2003 10:29:10 +0100

Dear Marc-Philippe,

> I just received a good question that must be considered even if I have
> my thoughts about it. The person represents agents as a set of objects:
> one for communication, one for means-end reasoning, one for scheduling,
> etc. She wants to write an interaction diagram that presents the
> exchange of messages between agents but as well, the method invocation
> between the agent as whole and the objects constituting it. The aim is
> to link messages to actions triggered by these messages. My first answer
> to the person was: actually, interaction diagrams are for messages and
> that's all, after a second thought, I think about UML 2 where we can
> have method invocations (grey rectangles on lifelines) and a recent
> discussion where AUML was criticized, yes, some people dare! ;-),
> because it is not possible to relate messages to actions.
> I would like to know your opinion and particularly if such approach must
> be added to the specifications or do I have to answer that statecharts
> and activity diagrams are done for that, I know continuations inside a
> diagram but between different diagrams of different kinds, that's
> possible?

I think that this case could be modeled by a sequence diagram with
interaction among "whole" agents. For each agent that needs to be modeled in
details, use PartDecomposition that refer another "detailed" Interaction
owned by this agent Classifier. This detailed (referenced) Interaction
(represented e.g. by a sequence diagram) depicts interaction of agent's
parts. Of course detailed "agent-internal" Iteraction can contain also a
Lifeline for the agent used for showing the way the "whole" communicates
with "its parts".

You can also use several detail "agent-internal" Interactions (e.g. each
Interaction handling of particular incoming message or a group of messages
from the overall agent-level Interaction), but in this case it is necessary
to put several PartDecompositions for one classifier in the overall
agent-level sequence diagram - what is little bit inconvenient in the case
of larger amount of messages.

But of course you can omit PartDecompositions and use "pointing to" other
diagrams as suggested Jim. For this purpose I would suggest to use a special
kind of dependency (not some tool-specific hyperlinking), to be able to
check the "pointing" semantics explicitly.

BTW I'm sorry for relatively late answer, but I was very busy up to now.


Radovan Cervenka | rce@whitestein.com
Whitestein Technologies | www.whitestein.com
Panenska 28 | SK-81103 Bratislava | Slovak Republic
Tel +421(2)5443-5502 | Fax +421(2)5443-5512
If you are not the intended recipient of this email,
you are not authorized to make any use of it;
please delete it and notify us by return email.
Thank you.