Mahadevan Subramaniam and Parvathi Chundi
View Full Paper
Software testing, data clustering, regression, observation testing
Clustering test proﬁles to retrieve relevant tests is a recurring theme in software validation. A novel clustering approach using a probabilistic notion of coverage among line-based test proﬁles is described. The approach automatically determines the number of clusters to generate a clustering and can potentially group together
tests to execute a few distinct lines of code. A simple method to identify tests aﬀected by program changes is developed and used to determine the retrieval eﬀort needed for cluster-based retrieval of aﬀected tests. The approach is applied to four unix utility programs from a popular testing benchmark. Our results show that comparing
original and new clusterings with respect to test cases is a promising criterion for deciding the potential re-clustering points in software evolution.