Explorando métricas de código para a detecção de Long Envious Methods

Abstract

Currently, projects are constantly evolving to introduce new features and adapt them to different execution contexts. With this, some problems in the quality of the software project can be introduced, which may be related to the presence of code anomalies, usually called code smells or bad smells in English, which are poor code structures. Anomalies are categorized into types and some of these types are Long Method, when a method has several lines of code, and Feature Envy, when a method often uses features from other classes. As manual identification of these code anomalies is costly, several code anomaly tools have been proposed. These tools use detection strategies based on code metrics, but they focus on removing a single anomaly and a recent study indicated that a code element can have two or more code anomalies. Thus, these elements may have their metrics degraded and the thresholds used by these tools are not sufficient to indicate their severity. This monograph aims to explore code metrics of methods that are Long Method and Feature Envy. These methods are called the Long Envious Method, a new type of anomaly that can be caused by excessive lines of code and the implementation of two or more features. The goal is to explore which code metrics can be used to identify this method. More than 8000 instances of code anomalies were investigated in three software projects, where 4707 (54%) are Long Methods and 3910 (45.3%) methods are Long Envious Methods. That is, Long Envious Methods are frequent and tools often fail to detect them. It can also be seen that the metrics cyclomaticComplexity, NumberOfCatchStatements and couplingIntensity have a significant difference between Long Envious Methods and Long Methods. Finally, it is concluded that these metrics can be used to identify Long Envious Methods.