FSE '17 paper: Continuous Variable-Specific Resolutions of Feature Interactions

Abstract: Systems that are assembled from independently developed features suffer from feature interactions, in which features affect one another’s behaviour in surprising ways. The Feature Interaction Problem results from trying to implement an appropriate resolution for each interaction within each possible context, because the number of possible contexts to consider increases exponentially with the number of features in the system. Resolution strategies aim to combat the Feature Interaction Problem by offering default strategies that resolve entire classes of interactions, thereby reducing the work needed to resolve lots of interactions. However most such approaches employ coarse-grained resolution strategies (e.g., feature priority) or a centralized arbitrator. Our work focuses on employing variable-specific defaultresolution strategies that aim to resolve at runtime features’ conflicting actions on a system’s outputs. In this paper, we extend prior work to enable co-resolution of interactions on coupled output variables and to promote smooth continuous resolutions over execution sequences. We implemented our approach within the PreScan simulator and performed a case study involving 15 automotive features; this entailed our devising and implementing three resolution strategies for three output variables. The results of the case study show that the approach produces smooth and continuous resolutions of interactions throughout interesting scenarios.

For paper reprint click here.