[news.eclipse.modeling.gmt.amw] Re: [ATL] what formalism to use to repre

[Marcos Didonet Del Fabro <marcos.didonet-del-fabro@xxxxxxxxxxxxxx>]

Hi,

As Frédéric said, you can use AMW to represent model transformations in a higher 
abstraction level. We create different metamodel extensions with the desired structures, 
e.g., the mapping between classes and structural features, attributes, etc.

I have some precisions about a couple of use cases:

- the tool interoperability use case 
(http://www.eclipse.org/gmt/amw/usecases/interoperability/) has a relatively complex 
weaving model that allows to represent several different transformation expressions.

- the metamodel comparison use case (http://www.eclipse.org/gmt/amw/usecases/compare/) has 
simple kinds of links (equality between elements of the same type). The advantage is that 
the weaving models are created semi-automatically, and then translated into ATL.

- finally, we have created a simple metamodel extension that has structures very close to 
the ATL metamodel (called SimpleATL), e.g., rules, input and output elements, and 
bindings. You can get more information about it at 
http://wiki.eclipse.org/index.php/AMW#Graphical_development_of_ATL_transformations.

For the time being, we have developed transformations from weaving models into 
transformation models. We intend to do the opposite (from ATL to a more abstract 
representation) in a near future.


Regards,

Marcos D. Del Fabro


ps.: I am copying the message to the AMW newsgroup as well.


Frédéric Jouault wrote:
> Hi,
>
>  > I have a question non directly related to ATL: What is the most
>  > appropriate formalism that can be used to represent a model to model
>  > transformation ?
>  > What is your experience with this ? Do you use UML ? and how ???
>
> The best complete representation I can think of that would work for any 
> ATL transformation is the ATL textual representation (i.e., the .atl file).
>
>
> However, several AMW use cases 
> (http://www.eclipse.org/gmt/amw/usecases/) show that model weaving is 
> quite appropriate to visually represent transformations:
>
> - completely: when a weaving extension can be used to represent 
> everything that needs to be represented,
>
> or
>
> - partially: by mapping metamodel classes and structural features, but 
> without detailing complex OCL expressions (you can generate the skeleton 
> of the transformation), or by encoding them textually in the Properties 
> view (which many UML tools do for OCL constraints).
>
>
> Regards,
>
> Frédéric Jouault