The goal of automated unit test generation tools is to create a set of test cases for the software under test that achieve the highest possible coverage for the selected test quality criteria. The most effective approaches for achieving this goal at the present time use meta-heuristic optimization algorithms to search for new test cases using fitness functions defined on existing sets of test cases and the system under test. Regardless of how their search algorithms are controlled, however, all existing approaches focus on the analysis of exactly one implementation, the software under test, to drive their search processes, which is a limitation on the information they have available. In this paper we investigate whether the practical effectiveness of white box unit test generation tools can be increased by giving them access to multiple, diverse implementations of the functionality under test harvested from widely available Open Source software repositories. After presenting a basic implementation of such an approach, DivGen (Diversity-driven Generation), on top of the leading test generation tool for Java (EvoSuite), we assess the performance of DivGen compared to EvoSuite when applied in its traditional, mono-implementation oriented mode (MonoGen). The results show that while DivGen outperforms MonoGen in 33% of the sampled classes for mutation coverage (+16% higher on average), MonoGen outperforms DivGen in 12.4% of the classes for branch coverage (+10% higher average).
Dieser Eintrag ist Teil der Universitätsbibliographie.
Search Authors in
