Towards Semantic Clone Detection, Benchmarking, and Evaluation

Developers copy and paste their code to speed up the development process. Sometimes, they copy code from other systems or look up code online to solve a complex problem. Developers reuse copied code with or without modifications. The resulting similar or identical code fragments are called code clones. Sometimes clones are unintentionally written when a developer implements the same or similar functionality. Even when the resulting code fragments are not textually similar but implement the same functionality they are still considered to be clones and are classified as semantic clones. Semantic clones are defined as code fragments that perform the exact same computation and are implemented using different syntax. Software cloning research indicates that code clones exist in all software systems; on average, 5% to 20% of software code is cloned. Due to the potential impact of clones, whether positive or negative, it is essential to locate, track, and manage clones in the source code. Considerable research has been conducted on all types of code clones, including clone detection, analysis, management, and evaluation. Despite the great interest in code clones, there has been considerably less work conducted on semantic clones. As described in this thesis, I advance the state-of-the-art in semantic clone research in several ways. First, I conducted an empirical study to investigate the status of code cloning in and across open-source game systems and the effectiveness of different normalization, filtering, and transformation techniques for detecting semantic clones. Second, I developed an approach to detect clones across .NET programming languages using an intermediate language. Third, I developed a technique using an intermediate language and an ontology to detect semantic clones. Fourth, I mined Stack Overflow answers to build a semantic code clone benchmark that represents real semantic code clones in four programming languages, C, C#, Java, and Python. Fifth, I defined a comprehensive taxonomy that identifies semantic clone types. Finally, I implemented an injection framework that uses the benchmark to compare and evaluate semantic code clone detectors by automatically measuring recall