Pitfalls in applying model learning to industrial legacy software

Abstract

Maintaining legacy software is one of the most common struggles of the software industry, being costly yet essential. We tackle that problem by providing better understanding of software by extracting behavioural models using the model learning technique. The used technique interacts with a running component and extracts abstract models that would help developers make better informed decisions. As promising in theory, as slippery in application it is, however. This report describes our experience in applying model learning to legacy software, and aims to prepare the newcomer for what shady pitfalls lie therein as well as provide the seasoned researcher with concrete cases and open problems. We narrate our experience in analysing certain legacy components at Philips Healthcare describing challenges faced, solutions implemented, and lessons learned.