Custom emf for "lollipop" notation?

[Molto Mike]

Hi,

we are modeling certain types of interfaces that need graphical representation different from the lollipop/socket.
While you can give alternative EMF's for a stereotype when displayed in "box" notation, it would be great to also allow changing the display of the lollipops, according to stereotype.

What do you think?

Mike

[manfred]

Changing the cup/lollipop connector notation would make the diagram illegal UML. While the specification allows creativity for stereotype icons, connector notations (or any other notation) is a normative representation of a particular semantic and cannot be changed. Please note, UML is not just pictures, it is a formal graphical language with a given syntax. Opening up for more "flexibility" here would be the immediate end of model-driven development.

The solution in your case would be a pair of stereotypes extending the Interface metaclass from StructuredClassifier. Since you are introducing a pair of new language elements, you can assign your own notation.

Manfred

[Molto Mike]

Hi Manfred,

you suggest: "extending the Interface metaclass through stereotypes".

This is exactly what I have in mind!

But: for stereotyped refinements of an Interface (and hence the refinement of semantics of the meta level M2 class) you still have no way to associate a different visual representation (lollipop like).

And hey: thanks for the hint that UML is not just pictures...    

Mike

[manfred]

Oh, Mike,

You didn't know that with the pictures...?  

You are touching a difficult point here.... What you are talking about is to substitute a notation in the context of stereotyping an element. Unfortunately this is not easy for most tools... I'm working on a component modeling tool which allows this. But it is only possible since my tool has a MOF-2 (CMOF) implementation underneath, so I can store enough metainformation associated with the new notation to make it working harmonically with the "old" stuff. Not easy....

Manfred

[Tom_Andries]

This thing with connectors not being able to be stereotyped, I find it quite inconsistent.
Apart from all meta modeling issues, the general question is whether you want a standard notation that is universal or a tuned notation that is specfic to it's goal. Being an engineering matter, it's typically a trade off to make.

Whether or not opening up for more flexibility would kill MDA and the like, only depends on HOW you do that. Indeed sematics need to be respected.

Changing the cup/lollipop connector notation would make the diagram illegal UML.

 Opening up for more "flexibility" here would be the immediate end of model-driven development.


Manfred