Automated Test Generation for Object-Oriented Programs with Multiple Targets

Software testing is costly. In particular, testing object-oriented programs is complicated and burdensome because of the difficulty in generating method sequences for creating objects and changing their states appropriately to achieve high branch coverage. Automated test generation based on static and dynamic analysis is not only an effective approach to saving time and reducing the burden of testing, but also an efficient way to find bugs. Seeker is an implementation for automated test generation that involves the generation of method sequences using static and dynamic analysis. However, when we want to change some values of variables to increase the branch coverage, the system fails to generate method sequences to achieve the desired states of objects with multiple targets (meaning fields or variables). In this paper, we extend the functionalities of the system for automated test generation so that it can handle multiple targets to cover. Our approach identifies all targets involved in uncovered branches and evaluates method sequences according to a fitness function, while applying a search strategy to suppress combinatorial explosion. The experimental results with several open source projects show that our extension achieves higher branch coverage than the original system and the effectiveness of the extension tends to vary according to the specific features of the projects.