Dynamically Testing Graphical User Interfaces

Software test generation for GUIs is a hard problem. The goal of this thesis is to investigate different methods for dynamically generating tests for GUIs. We introduce the concept of an event-pair graph, which is used to represent and measure test suites, and show how it can be used to generate tests and measure GUI coverage. Before we can begin generating tests, we first want to determine which is better: a small test suite with a few long tests or a large test suite with many short tests. Therefore, we designed and conducted a study to determine which is more effective. We found that moderate to long tests perform better than short tests. We then move on to discuss seven test generation algorithms. Two are based on random selection, two are based on greedy selection, one is based on Q-Learning, and the last two are based on ant colony optimization. We conducted a study in order to compare the performance of each algorithm. We measured code coverage, GUI coverage, time to run, and faults found. The results show that the greedy algorithms performed the best. Finally, we conducted a study in order to determine if any of the GUI coverage metrics can be used to predict code coverage, and we conducted a study to determine if any of the coverage metrics can be used to predict the faults found. The results show that event pairs are good at predicting code coverage, and that predicting faults is difficult