Backward propagation of code refinements on transformational code generation environments

Transformational code generation is at the core of generative software development. It advocates the modeling of common and variable features in software-system families with domain-specific languages, and the specification of transformation compositions for successively refining the abstract domain models towards eventually enriching them with execution semantics. Thus, using code-generation environments, families of software systems can be generated, based on models specified in high-level domain languages. The major advantage of this software-construction methodology stems from the fact that it enables the reuse of verified execution semantics, derived from domain models. However, like all software, once an implementation is generated, it is bound to evolve and manually refined to introduce features that were not captured by its original generation environment. This paper describes a conceptual framework for identifying features that have to be propagated backwards to generation engines, from refined generated references. Our conceptual framework is based on static and symbolic execution analysis, and aims to contribute to the maintenance and evolution challenges of model-driven development.