LLM-Assisted Repository-Level Generation with Structured Spec-Driven Engineering
State-of-the-art Large Language Models (LLMs) excel in code generation at the function level. However, the output quality significantly declines when scaling to repository-level systems. Current workflows relying only on natural language prompts suffer from inherent ambiguity and a lack of verifiability. To address this, we propose structured spec-driven engineering (SSDE), a paradigm that leverages structured artifacts to guide LLM generation. We argue that structured specifications as LLM inputs make high-quality, repository-level code generation a tangible goal, while at the same time offering superior verifiability, leading to significant potential for improvement. We first investigate the feasibility of this vision through a pilot study generating Model-View-Controller (MVC) business logic for three software systems using five LLMs, and then highlight the potential, challenges, and future roadmap for SSDE.
Wed 8 JulDisplayed time zone: Eastern Time (US & Canada) change
10:30 - 12:30 | Repo-Level Code GenerationResearch Papers / Ideas, Visions and Reflections / Industry Papers at MB 3.210 Chair(s): Zhijie Wang Concordia University | ||
10:30 20mTalk | RealBench: A Repo-Level Code Generation Benchmark Aligned with Real-World Software Development Practices Research Papers Jia Li Wuhan University, Hongyi Deng Peking University, Yiran Zhang Nanyang Technological University, Kechi Zhang Peking University, China, Tianqi Shao Peking University, Tiankuo Zhao Wuhan University, Weinan Wang Peking University, Zhi Jin Wuhan University, Ge Li Peking University, Yang Liu Nanyang Technological University, Yingtao Fang Wuhan University, Yihong Dong Peking University | ||
10:50 20mTalk | Do Not Treat Code as Natural Language: Implications for Repository-Level Code Generation and Beyond Research Papers Minh Le-Anh FPT Software AI Center, Huyen Nguyen Hanoi University of Science and Technology, Khanh An Tran Hanoi University of Science and Technology, Nam Le Hai Hanoi University of Science and Technology, Linh Ngo Van Hanoi University of Science and Technology, Nghi D. Q. Bui Google Research, Xuan-Bach D. Le University of Melbourne DOI Pre-print | ||
11:10 20mTalk | One Size Does Not Fit All: Revisiting Code Context Engineering for Repository-Level Code Generation Research Papers Yichen LI ByteDance, Qiye Lin Harbin Institute of Technology, Shenzhen, Yun Peng The Chinese University of Hong Kong, Zhihan Jiang The Chinese University of Hong Kong, Jinyang Liu Chinese University of Hong Kong, Chaozheng Wang The Chinese University of Hong Kong, Yintong Huo Singapore Management University, Singapore, Cuiyun Gao Harbin Institute of Technology, Shenzhen | ||
11:30 10mTalk | LLM-Assisted Repository-Level Generation with Structured Spec-Driven Engineering Ideas, Visions and Reflections Shuzhao Feng McGill University, Boqi Chen University of Ottawa, Brett Meyer McGill University, Gunter Mussbacher McGill University Pre-print | ||
11:40 20mTalk | In Line with Context: Repository-Level Code Generation via Context Inlining Research Papers Chao Hu Shanghai Jiao Tong University, Wenhao Zeng Shanghai Jiao Tong University, Yuling Shi Shanghai Jiao Tong University, Beijun Shen Shanghai Jiao Tong University, Xiaodong Gu Shanghai Jiao Tong University Pre-print | ||
12:00 20mTalk | RepoFuse: A Dual-Context Approach to Repository-Level Code Completion at Industrial Scale Industry Papers Ming Liang Ant Group, Xiaoheng Xie Ant Group, Gehao Zhang Ant Group, Xunjin Zheng Ant Group, Wei Jiang Ant Group, Chengpeng Wang National University of Singapore, Gang Fan Huawei Hong Kong Research Centre, Peng Di Kunlunxin & UNSW Sydney | ||