Leveraging relational concept analysis for automated feature location in software product lines

ABSTRACTFormal Concept Analysis (FCA) has been introduced for almost a decade as a suitable method for Feature Location (FL) on a collection of product variants. Even though FCA-based FL techniques allow to locate the core of a feature implementation, they do not propose a solution to trace feature interactions to their implementation. Thus, the extracted traceability links (traces) are too inaccurate, and, in the context of SPL extraction, cannot be used to generate complete products. In this paper, we propose to complement the FCA-based FL techniques by leveraging the power of Relational Concept Analysis, an extension of FCA to multi-relational data. From two given formal contexts, one for the product’s artefact and one for their features, our technique computes the traces that link the features and the feature interactions to their corresponding artefacts. Additionally, we introduce a stage that removes unnecessary features from the extracted traces, to make them easier to understand by an expert. Our FL technique can be applied at any artefact granularity (from files to statements) and independently from software languages. The results show that our technique produces valid traces, from which we were able to completely rebuild the set of artefacts for each initial product. Moreover, they show that our trace reduction removes, on average, between 31% and 85% of unnecessary features from the traces.