An empirical study of Android behavioural code smells detection

Mobile applications (apps) are developed quickly and evolve continuously. Each development iteration may introduce poor design choices, and therefore produce code smells. Code smells complexify source code and may impede the evolution and performance of mobile apps. In addition to common object-oriented code smells, mobile apps have their own code smells because of their limitations and constraints on resources like memory, performance and energy consumption. Some of these mobile-specific smells are behavioural because they describe an inappropriate behaviour that may negatively impact software quality. Many tools exist to detect code smells in mobile apps, based specifically on static analysis techniques. In this paper, we are especially interested in two tools: Paprika and aDoctor. Both tools use representative techniques from the literature and contain behavioural code smells. We analyse the effectiveness of behavioural code smells detection in practice within the tools of concern by performing an empirical study of code smells detected in apps. This empirical study aims to answer two research questions. First, are the detection tools effective in detecting behavioural code smells? Second, are the behavioural code smells detected by the tools consistent with their original literal definition? We emphasise the limitations of detection using only static techniques and the lessons learned from our empirical study. This study shows that established static analysis methods deemed to be effective for code smells detection are inadequate for behavioural mobile code smells detection.