Exploring decision drivers on god class detection in three controlled experiments

Abstract

Context: Code smells define potential problems in design of software. However, some empirical studies on the topic have shown findings in opposite direction. The misunderstanding is mainly caused by lack of works focusing on human role on smell detection. Objective: Our aim is to build empirical support to exploration of the human role on smell detection. specifically, we investigated what issues in make a human identify a as a smell. We called these issues decision drivers. Method: We performed a controlled experiment and replicated it twice. We asked participants to detect god (one of the most known smell) on different software, indicating what decision drivers they adopted. Results: The stronger drivers were class is high complex and method is misplaced. We also found the agreement on drivers' choice is low. Another finding is: some important drivers are dependent of alternative support. In our case, dependency was an important driver only when visual resources were permitted. Conclusion: This study contributes with the comprehension of the human role on smell detection through the exploration of decision drivers. This perception contributes to characterize what we called the code smell conceptualization problem.