Repository Context
Repository context, encompassing the intricate relationships within a software project's codebase, is crucial for improving code generation and completion tasks. Current research focuses on integrating repository-level information, such as dataflow analysis and cross-file dependencies, into large language models (LLMs) through techniques like dataflow-guided retrieval and iterative refinement with compiler feedback. These advancements aim to enhance the accuracy and efficiency of code generation tools by providing LLMs with a more comprehensive understanding of the project's structure and context, ultimately leading to improved developer productivity and software quality. The development of new benchmarks and datasets specifically designed to evaluate these models under realistic conditions is also a key area of focus.
Papers
Repository-Level Graph Representation Learning for Enhanced Security Patch Detection
Xin-Cheng Wen, Zirui Lin, Cuiyun Gao, Hongyu Zhang, Yong Wang, Qing Liao
ContextModule: Improving Code Completion via Repository-level Contextual Information
Zhanming Guan, Junlin Liu, Jierui Liu, Chao Peng, Dexin Liu, Ningyuan Sun, Bo Jiang, Wenchao Li, Jie Liu, Hang Zhu