Search based GUI Test Generation in Java - Comparing Code-based and EFG-based Optimization Goals

摘要

Modern software systems often communicate with their users by graphical user interfaces (GUI). While the underlying business logic may be fully covered by unit tests, the GUI mostly is not. Despite the widespread use of capture and replay tools, which leave the test generation of GUI tests to the user, recent research also focuses on automated GUI test generation. From the numerous approaches, which include symbolic execution, model-based generation, and random testing, search based test data generation seems to be the most promising. In this paper, we create GUI tests using hill climbing, simulated annealing and several genetic algorithms which deal differently with the sequence length and use multi or single objective algorithms. These different test data generators are compared in terms of runtime and coverage. All approaches are also compared using different optimization goals which are a high coverage of the event flow graph (EFG) of the GUIs and a high coverage of the underlying source code. The evaluation shows that the genetic algorithms outperform hill climbing and simulated annealing in terms of coverage, and that targeting a high EFG coverage causes the best runtime performance.