Towards statistical prioritization for software product lines testing

What is it about?

Highly-configurable systems (such as software product lines) allow to adapt their behavior to the user via the triggering of options (or features). Though a blessing for the user, this customization is a curse regarding quality assurance, since the number of possible behaviors grows exponentially with the number of features. Exploration of all possibilities is generally intractable for realistic software, pushing the research community to define sampling criteria based on feature models. Contrasting with existing sampling approaches, this paper offers to mine logs to extract information on which behaviors are actually used and uses it to select interesting behaviors to test.

Featured Image

Photo by Jon Moore on Unsplash

Photo by Jon Moore on Unsplash

Why is it important?

This work is important because it brides the idea of statistical testing with highly-configurable systems testing. The technique (and tool) for Markovian model inference based on n-grams had several follow-up applications. It contributed to the field of behavioral reverse engineering of variability-intensive systems, a challenging task due to the widespread and continuous evolution of these systems. That same technique also proved useful to improve crash reproduction using evolutionary algorithms.

Perspectives