On the Reuse of Existing Configurations for Testing Evolving Feature Models

Abstract

Software Product Lines (SPLs) are used for representing a variety of highly configurable systems or families of systems. They are commonly represented by feature models (FMs). Starting from FMs, configurations, used as test cases, can be generated to identify the products of interest for further activities. As the other types of software, SPLs and their FMs may evolve due to changing requirements or bug-fixing. However, no guidance is usually given on what to do with derived configurations when an FM evolves. The common approach is based on generating all configurations from scratch, which is not optimal since a greater effort is required for concretizing the new tests, and some of the old ones may be still applicable. In this paper, we present the use of a technique for generating combinatorial tests for evolving feature models: this technique incrementally builds the new combinatorial configuration set starting from the one generated from the previous model. Furthermore, we present a novel definition of dissimilarity among configuration sets that can be used to evaluate how much an evolved test suite differs from the previous one and thus allows evaluating the effort required for adapting old test cases to the new ones. Our experiments confirm that using the proposed technique, in general, leads to lower dissimilarity and test suite size w.r.t. the generation of tests from scratch.