Identifying strategies on god class detection in two controlled experiments

Context: "Code smell" is commonly presented as indicative of problems in design of object-oriented systems. However, some empirical studies have presented findings refuting this idea. One of the reasons of the misunderstanding is the low number of studies focused on the role of human on code smell detection. Objective: Our aim is to build empirical support to exploration of the human role on code smell detection. Specifically, we investigated strategies adopted by developers on god class detection. God class is one of the most known code smell. Method: We performed a controlled experiment and replicated it. We explored the strategies from the participant's actions logged during the detection of god classes. Result: One of our findings was that the observation of coupling is more relevant than the observation of attributes like LOC or complexity and the hierarchical relation among these. We also noted that reading source code is important, even with visual resources enha ncing the general comprehension of the software. Conclusion: This study contributes to expand the comprehension of the human role on code smell detection through the use of automatic logging. We suggest that this approach brings a complementary perspective of analysis in discussions about the topic.