Refactoring of Legacy Software using Model Learning and Equivalence Checking: an Industrial Experience Report
Mathijs Schuts, Jozef Hooman, and Frits Vaandrager
Appeared in: Proceedings 12th International Conference on
integrated Formal Methods (iFM 2016),
LNCS 9681, pages 311-325, Springer,
Many companies struggle with large amounts of legacy software that is difficult to maintain and to extend. Refactoring legacy code typically requires large efforts and introduces serious risks because often crucial business assets are hidden in legacy components. We investigate the support of formal techniques for the rejuvenation of legacy embedded software, concentrating on control components. Model learning and equivalence checking are used to improve a new implementation of a legacy control component. Model learning is applied to both the old and the new implementation. The resulting models are compared using an equivalence check of a model checker. We report about our experiences with this approach at Philips. By gradually increasing the set of input stimuli, we obtained implementations of a power control service for which the learned behaviour is equivalent.