Towards a parallel approach for test data generation for branch coverage with genetic algorithm using the extended path prefix strategy

Abstract

In this paper we present a proposal for an approach to test data generation for branch coverage with a structured genetic algorithm (GA) using the extended path prefix strategy. The structured GA implements a parallel master-slave distributed model in which each slave implements an elitist panmictic GA. Branches to be covered are selected by the master using the extended path prefix strategy and then dispatched to slaves. The slaves then conduct search for test data to cover the assigned target branch. The extended path prefix strategy ensures that each time a branch is selected for coverage, the sibling branch is already covered and that individuals are available that traverse the sibling. The strategy also permits a variable number of slaves to be used which can help speed up the test data generation process. Experiments on two programs with real inputs indicate that significant improvements are achieved over a simple panmictic GA in terms of number of generations and the coverage achieved.