On Sufficiency of Mutants

Mutation is the practice of automatically generating possibly faulty variants of a program, for the purpose of assessing the adequacy of a test suite or comparing testing techniques. The cost of mutation often makes its application infeasible. The cost of mutation is usually assessed in terms of the number of mutants, and consequently the number of "mutation operators" that produce them. We address this problem by finding a smaller subset of mutation operators, called "sufficient", that can model the behaviour of the full set. To do this, we provide an experimental procedure and adapt statistical techniques proposed for variable reduction, model selection and nonlinear regression. Our preliminary results reveal interesting information about mutation operators.