Testing configurable systems is challenging, as faults often arise from interactions between multiple features. Therefore, quality assurance requires to test configurations including such interactions. 𝑇-Wise sampling techniques generate configuration sets to cover each valid interaction among 𝑡 features in at least one configuration. However, state-of-the-art techniques fail to scale to large systems. Furthermore, the variety of sampling techniques results in a variety of samples, raising the question of which sample to use for testing. This thesis proposes leveraging binary decision diagrams (BDDs) to improve sampling methods as well as a fast to compute metric for comparing the quality of samples. Moreover, this thesis presents the first exact 2-wise interaction counting algorithm that successfully scales to the infamous system automotive02ᵥ4.
Aaron Molt (Thu,) studied this question.