Elimination of constraints for parallel analysis of feature models.

Abstract

Cross-tree constraints give feature models maximal expressive power since any interdependency between features can be captured through arbitrary propositional logic formulas. However, the existence of these constraints increases the complexity of reasoning about feature models, both for using SAT solvers or compiling the model to a binary decision diagram for efficient analyses. Although some works have tried to refactor constraints to eliminate them, they deal only with simple constraints (i.e., requires and excludes) or require the introduction of an additional set of features, increasing the complexity of the resulting feature model. This paper presents an approach that eliminates all the cross-tree constraints present in regular boolean feature models, including arbitrary constraints, in propositional logic formulas. Our approach for removing constraints consists of splitting the semantics of feature models into orthogonal disjoint feature subtrees, which are then analyzed in parallel to alleviate the exponential blow-up in memory of the resulting feature tree.