An Investigation of Compression Techniques to Speed up Mutation Testing

Mutation testing is widely considered as a high-end
test coverage criterion due to the vast number of mutants it
generates. Although many efforts have been made to reduce the
computational cost of mutation testing, in practice, the scalability
issue remains. In this paper, we explore whether we can use
compression techniques to improve the efficiency of strong mutation
based on weak mutation information. Our investigation is centred
around six mutation compression strategies that we have devised.
More specifically, we adopt overlapped grouping and Formal
Concept Analysis (FCA) to cluster mutants and test cases based
on the reachability (code covergae) and necessity (weak mutation)
conditions. Moreover, we leverage mutation knowledge (mutation
locations and mutation operator types) during compression. To
evaluate our method, we conducted a study on 20 open source
Java projects using manually written tests. We also compare
our method with pure random sampling and weak mutation.
The overall results show that mutant compression techniques
are a better choice than random sampling and weak mutation
in practice: they can effectively speed up strong mutation 6.3 to
94.3 times with an accuracy of >90%.