November 15, 2007


On November 8, 2007 the student presentation about the metamodeling frameworks EMF and MOF took place. Note, that on this blog there will be many more posts about EMF and MOF (in fact, theses metamodeling frameworks are an essential part of my research), so that this initial post is just to give you a start.

Metalayers up to MOF

So, why do we need metamodels? In computer science we usually describe objects of the real world in terms of models, e.g. by providing a UML class diagram. Advantages of this approach are, amongst others, that we gain more understanding of the particular application domain. A modeling language itself can be very complex as well. For instance, the UML specification consists of hundreds of pages. For such a modeling language it is even more important to be defined as formal and precise as possible, so it is quite self-evident to apply the modeling approach to the modeling language itself. Dealing with a modeling language itself is called metamodeling to avoid confusion. The classical picture on the left from the UML Infrastructure specification (chapter 7) provides a good understanding of this issue. You see that the components of a class diagram can be modeled by (you probably already guess it) a class diagram again. For instance, we could define, that a class may contain several attributes and so on (a composition association between the metaclasses Class and Attribute).

So the language UML is defined on level M2. However, we are not ready yet. We still have not specified how a language like UML can be defined. We raise the level of abstraction once more and get MOF, a language to describe languages (calm down, you do not have to learn yet another language from scratch, because MOF makes heavy use of the well-known class diagrams). Of course we could continue this procedure ad infinitum, but here are the good news: MOF can be described just using itself. So we are done (at least at the moment ;-)).


Why do we need yet another modeling framework? Is MOF not enough to describe arbitrary languages? Yes, in principle it is. But it has two disadvantages:

  • The MOF language is very powerful. Yes, indeed this can be a disadvantage, because you have to understand the concepts to use it. And the standard is very big.

  • The MOF language lacks tool support. Currently there is no widely used and mature tool for MOF. This may change in the future. The research tool MOFLON is quite powerful already and may fill this gap.

In contrast, EMF is widely used. It is integrated in the popular Eclipse platform, provides a powerful code generator and is easy to use. Its Metametamodel is called Ecore. You can think of it as the practically relevant subset of MOF (some quite useful features of MOF, however, are lost).

To conclude, the main advantages of Metamodeling are:

  • Precise definition of languages

  • Helpful for developing CASE tools

  • Use of Reflection possible (exploiting information about the model at runtime)

Further reading:

  • OMG's MOF

  • MOF Specification

  • MOFLON, a MOF diagram editor and meta-CASE tool

  • EMF Website

  • Eclipse Modeling Framework, Budinsky, Steinberg, Merks, Ellersick, Grose, 2003: standard book, however, do not buy it now but wait for the new version that will appear very soon

No comments: