Enhancing code clone detection using control flow graphs

Abstract

Code clones are syntactically or semantically equivalent code fragments of source code. Copy-and-paste programming allows software developers to improve development productivity, but it could produce code clones that can introduce non-trivial difficulties in software maintenance. In this paper, a code clone detection framework is presented with a feature extractor and a clone classifier using deep learning. The clone classifier is trained with true and false clones and then is tested with a test dataset to evaluate the performance of the proposed approach to clone detection. In particular, the proposed approach to clone detection uses Control Flow Graphs (CFGs) to extract features of a given code snippet. The selected features are used to compute similarity scores for comparing two code fragments. The clone classifier is trained and tested with similarity scores that quantify the degree of how similar two code fragments are. The experimental results demonstrate that using CFG features is a viable methodology in terms of the effectiveness of clone detection for both syntactic and semantic clones.