Weihuan Min’s research while affiliated with Central South University and other places

Code completion, which provides code suggestions by generating code snippets or structures, has become an essential feature of integrated development environments (IDEs). Recently, some studies have begun to use graph neural networks to complete AST-level code, and shown that it is promising to introduce GNNs into AST-level completion. However, these methods do not fully exploit the potential of reference codes with similar structures nor solve out-of-vocabulary (OOV). We propose Retrieval-Assisted Graph Code Completion (ReGCC) to enhance AST-level code completion further. ReGCC integrates a retrieval model that searches for similar code graphs to generate graph nodes and a completion model that leverages information from multiple domains. The key component of both the retrieval and completion models is the Multi-field Graph Attention Block, which consists of three layers of stacked attention: (1) Neighborhood Attention: preserves the heterogeneity and local dependency of the graph, enabling nodes to exchange information within their neighborhood. (2) Global & Memory Attention: addresses the long-distance dependency problem by providing nodes with a global view and the ability to extract information from the memory domain. (3) Reference Attention: lets nodes obtain valuable information from structurally similar reference code graphs. Furthermore, we tackle the OOV issue by employing feature matching and copying values from existing nodes. Specifically, we predict edges between nodes beyond the vocabulary, enabling effective information transfer. Experimental results demonstrate the superiority of our approach over state-of-the-art AST-level completion methods and generative language models.

The Internet of Things (IoT) is a revolutionary network of interconnected devices embedded with sensors and software that enables seamless communication, data sharing, and intelligent decision-making in the form of IoT services. To facilitate the efficient development of IoT services, code completion technique provides a promising solution by providing suggestions for missing code snippets. The development trend of IoT services is to support more mobile device terminals. Mobile devices are portable and easy to use, allowing IoT device operation and management anytime and anywhere. However, the current multi-token completion methods struggle to guarantee code generation quality under the constraints of low resources and low latency, making it difficult to fully support IoT service development. We propose a multi-token code completion framework, S2RCC, which completes code from skeleton to refinement with dual encoder and dual decoder. The framework consists of two phases: first, the code skeleton, which is the simplification of code containing structure-sensitive tokens, is predicted based on the semantics of the code context; second, the broken context is repaired with the predicted skeleton, and then parsed into the code structure so that the specific tokens can be generated combining the semantics and structure of context. Furthermore, we then provide an implementation of the framework, representing the repaired code as an improved Heterogeneous code graph and fusing the semantics and structure of code context by the three-layer stacked attention. We conducted experiments on multi-token completion datasets, showing that our model has achieved the state-of-the-art with the smallest possible scale and the fastest generation speed.