Research On Paralleling GUI Testing For Android Applications

Nowadays,widely used commercial Android applications(apps)usually have hundreds of activities and plenty of complex GUIs.It is challenging to test this kind of apps and guarantee their qualities.Existing automated testing tools focus on improving the test coverage with heuristics or even machine learning algorithms.However,these tools are time-consuming to reach their upper bound of coverage.They cannot com-plete the testing of an app with time-to-market pressure in a short time.To address such limitation,we consider scaling the testing of an app to multiple devices.We introduce the technique of parallel testing Android apps.In this paper,we make the following contributions:1.We propose an activity granularity parallel testing technique for Android apps,to improve the testing performance.We introduce the sequence minimization to re-move redundant events and reduce the time cost for state restoring.We also design a widget fuzzy match algorithm to alleviate the state inconsistency problem.2.We propose a parallel automated testing technique,which partitions a testing job into tasks in terms of GUI subtree,and dynamically allocates new tasks to idle devices.We introduce a DFS and BFS hybrid strategy,which reduces the cost in backtracking and replaying,and improves the efficiency of parallel testing.3.We implement two parallel testing techniques into two tools,ParaAim and ParaDroid,based on Mesos and Monkey.They are implemented on the same system architec-ture,and scale well.They can manage multiple Android devices distributed on several PCs.4.We evalute ParaAim and ParaDroid with 20 popular closed-source commercial apps on different number of devices and compared against Monkey and AimDroid.The results show that they scale well and evidently increases the average speed of ex-ploration.Especially,ParaDroid improves the efficiency of testing linearly.