Developing a DSL Abstract Syntax with the Eclipse Modeling Framework
- Apr 7, 2009
- In this chapter, we walk through the development of a domain-specific language (DSL) using the Eclipse Modeling Framework (EMF) and supporting components. Specifically, we develop the DSL’s abstract syntax using the Ecore metamodel. But first we cover some basics on what to consider when creating a DSL and the different implementation strategies you might want to employ when using EMF. Next, we provide an overview of EMF, leaving detailed information to the book  dedicated to this purpose. We cover some additional components of EMF and Model Development Tools (MDT) that enable you to further refine DSLs, and we develop a series of domain models for use in the sample projects.
3.1 DSL Considerations
Many considerations are involved in creating a DSL. Does a model already exist that is close enough? If so, can an existing model be extended, or is it fixed? Does the model need to be based on a standard? Does the DSL lend itself to graphical display and editing? Does the DSL require a textual syntax and editor? Will a product line be built on the DSL, and perhaps others? Is the Ecore metamodel expressive enough to suit your needs for a DSL? How can you model dynamic behavior?
A key consideration is the amount of flexibility you need or will tolerate in the DSL. As you can see in the examples, sometimes a change in the domain model makes your transformation definitions much easier to write. Also, frameworks such as GMF have certain limitations—or, rather, were designed with particular use cases in mind. Your particular style of modeling might not lend itself well to graphical representation, but a few changes might allow mapping to diagram elements much easier. For example, certain mappings in Query/View/Transformation (QVT) and template expressions can be facilitated by adding derived features or methods to the domain model. Complex queries using Object Constraint Language (OCL) (and, therefore, useful ones in QVT and Xtend) can be added to the domain model with code generated for their implementation at runtime. Having a feature available in the model will greatly simplify transformations and templates that access them.
This is not to say that you should let the tooling influence your DSL to an extent you are not comfortable with. The question is, how do you maintain a satisfactory level of “purity” in your DSL when considering the additional complexity associated with developing and maintaining the other Model-Driven Software Development (MDSD) artifacts? In general, the more complex the metamodel (DSL) is, the more complex the transformation definitions, templates, and diagram definitions are.
A set of conventions and best practices for the definition of DSLs, transformations, and templates likely will arise, as it has for Java and other popular programming languages. With conventions and best practices comes tooling to support refactorings, static analysis, and cleanup. At this stage of the Modeling project’s evolution, operations are still quite manual and even error prone. As an open source project that forms the basis for commercial products, Eclipse eventually will see more advanced features pushed down into it, thereby improving the Toolsmith experience.