Disqover: Debugging via code sequence covers

Automated model-based test generation has seen an undeniable trend towards obtaining large numbers of test cases. However, the full benefits of this trend have not yet percolated to downstream activities, such as debugging. We present Disqover for automated software debugging based on code sequence covers that leverages execution traces, or alternatively, sequence covers of large numbers of failing test cases to quickly identify causes of test failures, thereby aiding debugging. We develop a new algorithm that efficiently extracts commonalities between sequence covers in the form of ordered subsequences and values of variables contained in these subsequences that contribute to each failure. The results of our experimental evaluation suggest that users of Disqover need only 30% of the time needed to identify faults compared to the baseline in a user study. Furthermore, we show that the number of inspected statements using our approach is smaller than that of other state-of-the-art systems by multiple orders of magnitude. Additionally, we show that increasing the number and diversity of test cases improves our results by further decreasing the length of output subsequences to be examined.