Using similarity metrics for mining variability from software repositories

Abstract

Much activity within software product line engineering has been concerned with explicitly representing and exploiting commonality and variability at the feature level for the purpose of a particular engineering task e.g. requirements specification, design, coding, verification, product derivation process, but not for comparing how similar products in the product line are with each other. In contrast, a case-based approach to software development is concerned with descriptions and models as a set of software cases stored in a repository for the purpose of searching at a product level, typically as a foundation for new product development. New products are derived by finding the most similar product descriptions in the repository using similarity metrics. The new idea is to use such similarity metrics for mining variability from software repositories. In this sense, software product line engineering could be informed by the case-based approach. This approach requires defining and implementing such similarity metrics based on the representations used for the software cases in such a repository. It provides complementary benefits to the ones given through feature-based representations of variability and may help mining such variability.