On the Prioritization of Design-Relevant Smelly Elements: A Mixed-Method, Multi-Project Study

• Anderson Oliveira PUC-Rio
• Leonardo Sousa PUC-Rio
• Willian Oizumi PUC-Rio
• Alessandro Garcia PUC-Rio

Abstract

Software systems are likely to face what is called design problems. Given the typical lack of design documentation, developers have to rely on implementation-level symptoms, the so-called code smells, to identify and remove design problems. A code smell is a microstructure in the program that can indicate the presence of a design problem. Large programs have hundreds or thousands of program elements (e.g., classes) in which a significant proportion may be affected by smells. Consequently, due to time constraints and the large number of elements, developers have to prioritize the designrelevant program elements, i.e., locate a shortlist of elements that are suspects of having design-relevant smells. However, this task is hard and time-consuming. Unfortunately, the literature fails to provide developers with effective heuristics that automate such prioritization task. The objective of this paper is to propose heuristics that effectively locate a shortlist of design-relevant smelly program elements. For this purpose, we report two studies. In the first one, we investigated the criteria that developers used in practice to accurately prioritize design-relevant smelly elements. Based on these criteria, we derived a preliminary suite of prioritization heuristics. Since we do not know if the heuristics are suitable for an effective prioritization across multiple projects, we performed a second study to evaluate the proposed heuristics. We found that two out of nine heuristics reached an average precision higher than 75% for the four projects we analyzed. Thus, our findings suggest these heuristics are promising to support developers in prioritizing design-relevant smelly elements.

Keywords: design problems, heuristics, prioritization