Code Smells Detection and Visualization: A Systematic Literature Review

Code smells tend to compromise software quality and also demand more effort by developers to maintain and evolve the application throughout its life-cycle. They have long been catalogued with corresponding mitigating solutions called refactoring operations. Researchers have argued that due to the subjectiveness of the code smells detection process, proposing an effective use of automatic support for this end is a non trivial task. This systematic literature review (SLR) has a twofold goal: the first is to identify the main code smells detection techniques and tools discussed in the literature, and the second is to analyze to which extent visual techniques have been applied to support the former. Over eighty primary studies indexed in major scientific repositories were identified by our search string in this SLR. Then, following existing best practices for secondary studies, we applied inclusion/exclusion criteria to select the most relevant works, extract their features and classify them. We found that the most commonly used approaches to code smells detection are search-based (30.1%), metric-based (24.1%), and symptom-based approaches (19.3%). Most of the studies (83.1%) use open-source software, with the Java language occupying the first position (77.1%). In terms of code smells, God Class (51.8%), Feature Envy (33.7%), and Long Method (26.5%) are the most covered ones. Machine learning (ML) techniques are used in 35% of the studies, with genetic programming, decision tree, support vector machines and association rules being the most used algorithms. Around 80% of the studies only detect code smells, without providing visualization techniques. In visualization-based approaches several methods are used, such as: city metaphors, 3D visualization techniques, interactive ambient visualization, polymetric views, or graph models. This paper presents an up-to-date review on the state-of-the-art techniques and tools used for code smells detection and visualization. We confirm that the detection of code smells is a non trivial task, and there is still a lot of work to be done in terms of: reducing the subjectivity associated with the definition and detection of code smells; increasing the diversity of detected code smells and of supported programming languages; constructing and sharing oracles and datasets to facilitate the replication of code smells detection and visualization techniques validation experiments.