Automated Code Smell Detection using Metrics Based Approach

Abstract

Business now a days completely rely on software systems for this vary reason, software system are becoming more and more complex with time. And for this software system are continuously subjected to rapid change to implement various new feature. To meet this rapid change request developers ignore language standards and proper design patterns which later on leads to code smells. This rapid market requirements and various change patches were explored by Fowler specifying suboptimal coding decisions subjected by developers while writing source code. Many researches are conducted to explore these suboptimal choices which leads to the discovery of automatic techniques to point towards design errors in source code. Many of these techniques completely rely on structural characteristics (like method calls, LOC) extracted from the source code. In spite of researches hard work in the past, there is a gap for detecting code smells with respect to dynamic languages like python. There is still lack of evidence concerning (i) how to detect code smells between classes (ii) which code smell happens more frequently and these problems need to be addressed because it impact directly on the software maintainability. In our research we propose an algorithm to detect code smells between classes (considering only three code smells. We devise a metrics based proposition for detecting code smells that is based on Abstract Syntax Tree, i.e., parsing code and forming nodes and we evaluate our results by running the algorithm on five open source python projects.