Abstract
GUI testing has been an effective means of validating Android apps. Meanwhile, it still faces a strong challenge about how to explore trails, i.e., unfrequented test sequences, as defects tend to reside on these unfrequented trails. This paper introduces FSMdroid, a novel, guided approach to GUI testing of Android apps. The essential idea of FSMdroid is to (1) construct an initial stochastic model for the app under test, (2) iteratively mutate the stochastic model and derive tests. The model mutations are guided by an MCMC sampling method such that the resulting test sequences can be diverse and also achieve high code coverage during testing. We have evaluated FSMdroid on 40 real-world Android apps. Compared with the traditional model- based testing approaches, FSMdroid enhances the diversity of test sequences by 85%, but reduces the number of them by 54%. Furthermore, we uncover 7 app bugs.