Improving Evolutionary Test Data Generation with the Aid of Symbolic Execution

Recently, search based techniques have received great attention as a means of automating the test data generation activity. On the contrary, more traditional methods that automate the test data generation usually employ symbolic execution by incorporating a path generation phase and constraint solvers to produce the sought test data. In this paper, the benefits of both schools of thought are bridged in an attempt to investigate whether a mixed strategy approach could be employed when evaluating a coverage criterion. To this effect, a strategy that uses symbolic execution and dynamic domain reduction in order to enhance the initial population and approximately prune the search space considered by evolutionary based methods is proposed. This suggestion is also put under a number of tests which clearly show a dramatic improvement of its effectiveness. This suggests that the combination of evolutionary based and symbolic execution approaches can be beneficial toward, automating the generation of test data.