On mutation

Program mutation has been advanced by DeMillo et. al. as a powerful program testing technique. The present work is an empirical study of several factors important in the feasibility and reliability of mutation analysis. In the course of investigating the coupling effect, over a million complex mutants were subjected to test data sufficient for first order mutation. All but 45 of these were either eliminated by the test data or proved to be equivalent to the original programs. Most of the 45 could be addressed within the framework of mutation analysis by the addition of a new mutagenic operator. An experiment on the human evaluation of equivalence (of mutants to original programs) produced inconclusive results, but indicates the need for caution in the practice of mutation analysis in the step of judging mutants equivalent. A study of mutagenic operator efficiencies indicates that mutation analysis can be performed on compiled programs, rather than being limited to interpretive execution.