Automated Mobile App Test Script Intent Generation via Image and Code Understanding
Shengcheng Yu,Chunrong Fang,Tongyu Li,Mingzhe Du,Xuan Li,Jing Zhang,Yexiao Yun,Xu Wang,Zhenyu Chen
DOI: https://doi.org/10.48550/arxiv.2107.05165
2021-01-01
Abstract: Testing is the most direct and effective technique to ensure software quality. However, it is a burden for developers to understand the poorly-commented tests, which are common in industry environment projects. Mobile applications (app) are GUI-intensive and event-driven, so test scripts focusing on GUI interactions play a more important role in mobile app testing besides the test cases for the source code. Therefore, more attention should be paid to the user interactions and the corresponding user event responses. However, test scripts are loosely linked to apps under test (AUT) based on widget selectors, making it hard to map the operations to the functionality code of AUT. In such a situation, code understanding algorithms may lose efficacy if directly applied to mobile app test scripts. We present a novel approach, TestIntent, to infer the intent of mobile app test scripts. TestIntent combines the GUI image understanding and code understanding technologies. The test script is transferred into an operation sequence model. For each operation, TestIntent extracts the operated widget selector and link the selector to the UI layout structure, which stores the detailed information of the widgets, including coordinates, type, etc. With code understanding technologies, TestIntent can locate response methods in the source code. Afterwards, NLP algorithms are adopted to understand the code and generate descriptions. Also, TestIntent can locate widgets on the app GUI images. Then, TestIntent can understand the widget intent with an encoder-decoder model. With the combination of the results from GUI and code understanding, TestIntent generates the test intents in natural language format. We also conduct an empirical experiment, and the results prove the outstanding performance of TestIntent. A user study also declares that TestIntent can save developers' time to understand test scripts.