Motivations

Model-Driven Software Development focuses on managing domain-specific models, which represent concerns at conceptual level. Such models are generally represented as various artifacts, in many languages and representations. Model-Driven approach also generally requires code generators to generate source code from models. The underlying semantics of this generated code is generally encoded in the code generators and can be inlined or implicitly defined by code generation process, or may be sometimes explicit to the code generation. The following figure illustrates the classical vision for for Model-Driven Engineering where some Java classes (on the right) are generated from a UML class diagram (on the left).

A first major drawback of this approach is the gap created between the conceptual level (the model) and the source code, where semantics may be totally hidden or implicit. Another classical issue resides in the development or maintenance process, where model and source code must evolve independently (and sometimes by different actors, e.g. architect and developer). Round-tripping mechanisms are commonly used to overcome those difficulties. But round-tripping tools are difficult to use and maintain, and the process can lead to inextricable issues in a context of concurrent modifications of model and source code. This explains why lots of major software developments projects abandon the idea of maintaining the link between models and source code during the development process. In that context, the model is developed at the early stages of development process, and is used to prototype software application. The model may be manually reverse engineered at the end of development process, for documentation purposes.