Research Trends, Detection Methods, Practices, and Challenges in Code Smell: SLR

Context: A code smell indicates a flaw in the design, implementation, or maintenance process that could degrade the software’s quality and potentially cause future disruptions. Since being introduced by Beck and Fowler, the term code smell has attracted several studies from researchers and practitioners. However, over time, studies are needed to discuss whether this issue is still interesting and relevant. Objective: Conduct a thorough systematic literature review to learn the most recent state of the art for studying code smells, including detection methods, practices, and challenges. Also, an overview of trends and future relevance of the topic of code smell, whether it is still developing, or if there has been a shift in the discussion. Method: The search methodology was employed to identify pertinent scholarly articles from reputable databases such as ScienceDirect, IEEE Xplore, ACM Digital Library, SpringerLink, ProQuest, and CiteSeerX. The application of inclusion and exclusion criteria serves to filter the search results. In addition, forward and backward snowballing techniques are employed to enhance the comprehensiveness of the results. Results: The inquiry yielded 354 scholarly articles published over the timeframe spanning from January 2013 to July 2022. After inclusion, exclusion, and snowballing techniques were applied, 69 main studies regarding code smells were identified. Many researchers focus on detecting code smells, primarily via machine learning techniques and, to a lesser extent, deep learning methods. Additional subjects encompass the ramifications of code smells; code smells within specific contexts, the correlation between code smells and software metrics, and facets about security, refactoring, and development habits. Contexts and types of code smells vary in the focus of the study. Some tools used are Jspirit, aDoctor, CAME, and SonarQube. The study also explores the concept of design smells and anti-pattern detection. While a singular dominating technique to code smell detection has yet to be thoroughly investigated, other aspects of code smell detection remain that still need to be examined. Conclusion: The findings underscore scholarly attention’s evolution towards code smells over the years. This study identified significant journals and conferences and influential researchers in this field. The detection methods used include empirical, machine learning, and deep learning. However, challenges include subjective interpretation and limited contextual applicability.