Position Statement on the Role and Future of Search Based Software Engineering

SSBSE 2009 1st International Symposium on Search Based Software Engineering Windsor, UK 13-15 May 2009 During the 1970’s and 1980s there were many computer scientists - and software engineers especially - who dreamed of perfect solutions to complex problems based purely on mathematical semantics and logic. With the right formal notation and proof rules they imagined that it would be possible, for example, to create complex software systems automatically from mathematical specifications that were perfect in the sense of requiring no testing [1]. Although for many this dream has never died, and examples of modestly complex, verified systems and tools to support them have been developed, in practice most software engineering challenges can never support perfect solutions [2]. Since even a perfectly verified system can only have been verified against a formal specification, the need for testing still remains for such a system since there can never be a formal automated means of verifying real-world requirements. The quest for perfect solutions is, of course, also provably infeasible for a wide class of algorithmic problems, namely all of those such as the traveling salesman problem with complexity NP-complete or worse. Since such problems are pervasive and will not just go away, we have been forced to consider solutions that were less than perfect. This led to the broad discipline of approximation in algorithm design with different approaches to optimization and also probabilistic solutions [3]. It seems to me that the current discipline of search-based software engineering [4] is the inevitable extension of this work into areas beyond algorithm design, with a special emphasis on testing. The growing popularity of search-based software engineering therefore seems to be a sign of the maturity of the entire software engineering discipline as it finally moves away from the naivety of those earlier years and confronts the reality of accepting uncertainty and approximation. While the need for search-based techniques to support testing of complex systems seems to me to be self-evident, my view is that the subject as a whole may be cast in too narrow a light, with too much focus on specific types of metaheuristic search [5]. I can understand why, to date the emphasis has been so. The need for optimising test data is both an obviously important problem and a natural candidate for the class of metaheuristic search techniques commonly used. However, as is made clear in [4] the challenges of search-based software engineering are far greater and could encompass a much broader range of methods. Search-based software engineering essentially deals with any software engineering problem for which there is no perfect solution and for which no efficient deterministic algorithm will find an optimal solution. Such problems are inevitably characterized by uncertainty, and I feel there is a much broader class of methods, which can be classified as ‘intelligent’, that may be