Yup!
Sequence Diagrams are also sort of 'orphans' in EA, though that might not apply in this specific context.
I believe that all diagrams showing runtime behavior in sequences are falling in this same category- like object diagrams.
Oliver
It seems that upholding the thesis that diagrams do not contain any information is indeed most problematic for the behavioural diagrams.
Lets take a look at the Activity Diagram. In an activity diagram we model a sequence of activities that are executed in a specific order, with specific constraints.
These same activities however may be used on another diagram having a different order, different constraints, and different transitions.
The context of an activity diagram is usually the a single behavioural entity such as a usecase. Everything on the activity diagram is related to this context, and more importantly, whats not on the activity diagram is NOT behaviour from this usecase.
So how can we make the difference between those activities/constraints/transitions that are part of the usecase and those that are not?
Maybe the key here is to use the classifier/instance paradigm.
Let me explain: I think that for most if not all behavioural diagram we can find some behaviour parent. This parent should contain all the information necesarry to model the actual behaviour. This parent could be a usecase, a collaboration, (others?).
If we not only use instances to model the behaviour of such a parent then we could store those instances nested within the parent.
In the case of sequence diagrams this is pretty obvious since most people are already used to using objects of classes in a sequence diagram.
For Activities that is something else. Most activity diagrams I've seen use the classifiers (the Activity) and model transitions between those.
Of course the classifiers can't be nested in the behavioural parent since they could be part of multiple parents. Therefore we should find a way to model activity diagrams using instances of these Activities.
When asking EA to create an instance of an Activity it creates an Action.
Whether or not the Action can be seen as an instance of an Activity (according to the UML specification) I'm uncertain. I will try to figure it out and post back.
Anyway, if this is really how OMG intented modelling behaviour then EA should make that possible. The tools should not store any important model information in diagrams.
Diagrams should only contain information with regards to layout etc...
Any thoughts?
Geert
