Automated testing of mobile apps has received significant attention in recent years from researchers and practitioners alike. In this paper, we report on the largest empirical study to date, aimed at understanding the test automation culture prevalent among mobile app developers. We systematically examined more than 3.5 million repositories on GitHub and identified more than 12,000 non-trivial and real-world Android apps. We then analyzed these non-trivial apps to investigate (1) the prevalence of adoption of test automation; (2) working habits of mobile app developers in regards to automated testing; and (3) the correlation between the adoption of test automation and the popularity of projects. Among others, we found that (1) only 8% of the mobile app development projects leverage automated testing practices; (2) developers tend to follow the same test automation practices across projects; and (3) popular projects, measured in terms of the number of contributors, stars, and forks on GitHub, are more likely to adopt test automation practices. To understand the rationale behind our observations, we further conducted a survey with 148 professional and experienced developers contributing to the subject apps. Our findings shed light on the current practices and future research directions pertaining to test automation for mobile app development.
Figure below depicts the flow of data collection and analysis in our study. This study consisted of the following steps: (1) we first collected a large list of GitHub repositories from the GHTorrent database; (2) we set filtering criteria to identify the repositories representing non-trivial Android apps; (3) we further analyzed the identified repositories to collect their meta-data and information about automated tests and popularity; (4) we evaluated the collected dataset to answer research questions about the test automation culture prevalent among mobile app developers; and finally (5) we conducted a survey with the developers of the subject apps to get a deeper understanding of the underlying reasons for our observations from the dataset.
The artifacts are available for download from here.