MDD: How dynamic is MDD at runtime?

Question

Over the years, I've investigated a lot of ways to use code generators and MDD. I've always felt that something is lacking: Patching and changes to the model at runtime.

Patching: If you have a code generator, all your classes should look the same. Now you have a single exception. All code generators so far would require that I modify the template or the template engine to make this work.

Wouldn't it be better if I could apply patches to the result of the code generation step to fix the exceptions?

Answer 1

MDD doesn't work because it is based on a view of the domain and not the entire domain. I mean that usually MDD take an XMI in entry coming from an UML diagram. The problem is that this diagram is only a view of the domain and therefore you have many alternative and the real world is a mot more complex specially at deployment stage.

The only company which has provided me real value in my project was Omondo with EclipseUML. EclipseUML doesn't try to do MDD but create UML at diagram level live synchronized with code. Deployment is made using stereotypes which are added in the java annotation in the code. I can therefore model and if I add deployment stereotypes then my application can be deployed immediately/ If I manually change my code, then my model is refactored and all my views updated. If I want to add a documentation then I just add notes in the metamodel. These notes are live available when I click on each element. No more printed documentation needed because live navigation, dynamic views creations etc...

My EclipseUML model is always up to date, and I can deploy it immediately because Java annotations are lived synchronized between the model, the metamodel, the diagrams and the code. Really cool :-) :-)

Answer 2

Well, it depends on how you build your model. If fact, it depends on what code generator you are using, its approach, and what it lets you do.

Creating an exception to a rule (model) is more or less against the nature of MDD, unless the applied modeling approach allows you to add exceptions as modeling entities.

I think ABSE is the only modeling approach that accepts "custom code" as a first-class entity, just like a text or an integer. If you create a template that contains a "CustomCode" parameter, you can later add your exception code only when necessary, without breaking your model rules. This can be used to add or replace code. You just need to specify it in your template.

AtomWeaver is a free implementation of the ABSE modeling methodology.