Toward Generalization of Mutant Clustering Results in Mutation Testing

Abstract

Mutation testing is effectively used for evaluation of test case quality but suffers from high cost required for its realization. Mutated programs are injected with program changes specified by various mutation operators. One of the methods applied to the reduction of mutant number is mutant clustering. Instead of using all generated mutants, special mutant groups are distinguished and group representatives are used in further evaluation of tests. Mutant clustering gave some promising results for C programs. In case of object-oriented programs with standard and object-oriented operators the results were positive but not superior to other cost reduction techniques. An open issue is interpretation of mutant clustering results and their generalization to other projects in terms of used mutation operators. In this paper, three metrics are proposed to comprehend mutation clustering. Experimental results are analyzed toward usefulness of mutants created by various operators, their frequency, and dependency. The evaluation result confirms applicability of the metrics, and the practical guidelines about the mutation operators are concluded from the experimental data.