On the Impact of Multiple Source Code Representations on Software Engineering Tasks -- An Empirical Study

Efficiently representing source code is essential for various software engineering tasks such as code classification and code clone detection. Existing approaches for representing source code primarily use AST, and only a few works focus on semantic graphs such as CFG and PDG, which contain essential information about source code that AST does not have. Even though some works tried to utilize multiple representations, they do not provide any insights about the costs and benefits of using multiple representations against a single appropriate representation for the task. Moreover, they use hand-crafted program features to solve a specific task and have limited use cases. The primary goal of this paper is to discuss the implications of utilizing multiple code representations, specifically AST, CFG, and PDG, and how each of them affects the performance of a task. In this process, we use an approach that can use program features from multiple code graphs while not specifically coupling this approach to a specific task or a language. Our approach stems from the idea of modeling AST as a set of paths and using a learning model to capture program properties. We modify an existing AST path-based approach to accept multiple code representations as input. We do this since it allows us to measure the performance boost provided by additional representations over AST. We evaluate our approach on three tasks: Method Naming, Program Classification, and Code Clone Detection. Our approach increases the performance on these three tasks by 11% (F1), 15.7% (Accuracy), and 9.3% (F1), respectively, over the baseline. We discuss the impact of semantic features from the CFG and PDG paths on performance and the additional overheads incurred through our approach. We envision this work providing researchers with a lens to evaluate combinations of source code representations for various tasks.