Using Genetic Algorithms to Generate Test Suites for FSMs

It is unaffordable to apply all the possible tests to an implementation in order to assess its correctness. Therefore, it is necessary to select relatively small subsets of tests that can detect many errors. In this paper we use different approaches to select these test suites. In order to decide how good a test suite is, we confront it with a set of mutants, that is, small variations of the specification of the system to be developed. The goal is that our algorithms build test suites that kill as many mutants as possible. We compare the different approaches (consider all the possible subsets up to a given number of inputs, intelligent greedy algorithm and different genetic algorithms) and discuss the obtained results. The whole framework has been fully implemented and the tool is freely available.