Bottom-up model-driven development

Prominent researchers and leading practitioners are questioning the long-term viability of model-driven development (MDD). Finkelstein recently ranked MDD as a bottom-ten research area, arguing that an approach based entirely on development and refinement of abstract representations is untenable. His view is that working with concrete artifacts is necessary for learning what to build and how to build it. What if this view is correct? Could MDD be rescued from such a critique? We suggest the answer is yes, but that it requires an inversion of traditional views of transformational MDD. Rather than develop complete, abstract system models, in ad-hoc modeling languages, followed by top-down synthesis of hidden concrete artifacts, we envision that engineers will continue to develop concrete artifacts, but over time will recognize patterns and concerns that can profitably be lifted, from the bottom-up, to the level of partial models, in general-purpose specification languages, from which visible concrete artifacts are generated, becoming part of the base of both concrete and abstract artifacts for subsequent rounds of development. This paper reports on recent work that suggests this approach is viable, and explores ramifications of such a rethinking of MDD. Early validation flows from experience applying these ideas to a healthcare-related experimental system in our lab.