Investigating performance issues in mobile apps

The world is moving at a dynamic pace, and this has led to the technological advancement of mobile applications. This rise in the advancement of the mobile application comes with critical concerns to end-users in terms of the performance, especially when implementing high intensive features. Moreover, enjoyable user experience in terms of performance is often considered as the main parameter to measure the success of any app. Poor implementation of source code, lack of developers knowledge, and time constraints on resolving performance issues are few of the major potential performance drawbacks in Android applications. To overcome these performance issues, in this dissertation, we focus on investigating the performance-related issues in open-source Android application (mainly apps from GitHub). Our thesis can be divided into four key research objectives: (i) initially we investigate on the extent to which developers consider performance issues in their commits (while maintaining their apps) and how they document it, (ii) to complement this study, we conduct an experiment to study the evolution of Android specific performance issues detected by Android Lint, and based on the obtained results, (iii) we introduce an Eclipse plugin that can be used to automatically resolve seven types of performancerelated issues detected by Android Lint; in addition to this, we performed a survey-based study to analyze the self assessed performance refactoring code of the proposed tool from the developers’ perspective; and (iv) we design and conduct a measurement-based study to examine the impact of performance violations at run-time. The key contributions of this thesis are (i) a taxonomy of developers’ concerns about performance, obtained by applying card sorting technique on a dataset of commit messages extracted from GitHub, (ii) the empirical research considering seven types of performance issues identified by Lint tool resulted: (a) a taxonomy for different kinds of evolution patterns of Android performance issues emerged by tracing the history of Android apps, (b) a catalog of documented performance issues resolved by Android developers, (iii) an automatic refactoring tool to address the seven types of performancerelated issues of Android Lint, (iv) developers perspective related to refactoring and non-refactoring code in the form of survey responses, and (v) a measurement-based study to analyze run-time performance of Android apps. These results provide developers a base to take the next leap in solving performance-related issues in the mobile apps of the future.