PCR-Chain: Partial Code Reuse Assisted by Hierarchical Chaining of Prompts on Frozen Copilot

Qing Huang, Jiahui Zhu, Zhilong Li, Zhenchang Xing, Changjing Wang, Xiwei Xu

Abstract: API documentation, technical blogs and programming Q&A sites contain a large amount of partial code that can be reused in programming tasks. However, due to unresolved simple names and last-mile syntax errors, such partial code is frequently not compilable. To facilitate partial code reuse, we develop PCR-Chain for resolving FQNs and fixing last-mile syntax errors in partial code based on a giant pre-trained code model (e.g., Copilot). Methodologically, PCR-Chain is backed up by the underlying global-level prompt architecture (which combines three design ideas: hierarchical task breakdown, prompt composition including sequential and conditional structures, and a mix of prompt-based AI and non-AI units) and the local-level prompt design. Technically, we propose PCR-Chain, which employs in-context learning rather than supervised fine-tuning with gradient updates on downstream task data. This approach enables the frozen, giant pre-trained code model to learn the desired behavior for a specific task through behavior-describing prompts and imitate it to complete the task. Experimental results show that PCR-Chain automatically resolves the FQNs and fixes last-mile syntax errors in 50 partial code samples collected from Stack Overflow with high success rates, without requiring any program analysis. The correct execution of the unit, module, and PCR-Chain demonstrates the effectiveness of the prompt design, prompt composition, and prompt architecture. Website:https://github.com/SE-qinghuang/PCR-ChainDemoVideo: https://youtu.be/6HGRNc2JE