Model-based, Mutation-driven Test-case Generation Via Heuristic-guided Branching Search

This work introduces a heuristic-guided branching search algorithm for model-based, mutation-driven test-case generation. The algorithm is designed towards the efficient and computationally tractable exploration of discrete, non-deterministic models with huge state spaces. Asynchronous parallel processing is a key feature of the algorithm. The algorithm is inspired by the successful path planning algorithm Rapidly exploring Random Trees (RRT). We adapt RRT in several aspects towards test-case generation. Most notably, we introduce parametrized heuristics for start and successor state selection, as well as a mechanism to construct test cases from the data produced during the search. We implemented our algorithm in the existing test-case generation framework MoMuT. We present an extensive evaluation of the proposed heuristics and parameters of the algorithm, based on a diverse set of demanding models obtained in an industrial context. In total, we continuously utilized 128 CPU cores on three servers for several weeks to gather the experimental data presented. We show that branching search works well and the use of multiple heuristics is justified. With our new algorithm, we are now able to process models consisting of over 2,300 concurrent objects. To our knowledge, there is no other mutation-driven test-case generation tool that is able to process models of this magnitude.