Alternating GUI Test Generation and Execution

Users of today's software perform tasks by interacting with a graphical user interface (GUI) front-end via sequences of input events. Due to the flexibility offered by most GUIs, the number of event sequences grows exponentially with length. One ubiquitous challenge of GUI testing is to selectively generate those sequences that lead to potentially problematic states. This paper presents ALT, a new technique that generates GUI test cases in batches, by leveraging GUI run-time information from a previously run batch to obtain the next batch. Each successive batch consists of "longer" test cases that expand the state space to be explored, yet prune the "unimportant" states. The "alternating" nature of ALT allows it to enhance the next batch by leveraging certain relationships between GUI events (e.g., one enables the other, one alters the other's execution) that are revealed only at run-time and non-trivial to infer statically. An empirical study on four fielded GUI-based applications demonstrates that ALT is successful at identifying complex failure-causing interactions between GUI events.