Detecção de anomalias de código usando métricas de software

Abstract

Metrics have traditionally have been used to evaluate the maintainability of software programs by supporting identification of symptoms of bad smells. Recently, concern metrics have also been proposed with this purpose. While traditional metrics quantify properties of sotware modules, concern metrics quantify concern properties, such asthe scattering and tangling of concerns realized in a program. Despite being increasingly used in experimental studies, although lack empirical knowledge as to their effectiveness in identifying bad smells. This work investigates whether concern metrics may provide useful indicators for detecting five bad smells: Divergent Change, Shotgun Surgery, God Class, Feature Envy and God Method. For this, two experimentalstudies were performed. In the first study, we used a set of 54 participants from two different institutions for detecting three bad smells in classes in two systems. In the second study, based on detection of bad smells in methods, we used a set of 47 participants from two institutions in order to detect two bad smells in of one system. Inboth studies, participants analyzed traditional and concern metrics to assist bad smell detection. The results indicated that the concern metrics support developers detect these bad smells. In addition, our results showed that the Number of Concern per Component metric is a good indicator for the Divergent Change. However, elaborated joint analysis of traditional metrics and concern is often necessary to detect God Classand Shotgun Surgery. Regarding the results of bad smells in methods, they indicated that the Number of Concern per Operations metric is a good indicator for the God Method. Based on the results of these two studies, we propose a quantitative method for supporting a automated detection bad smells