一句话:提出结合形式验证反馈与检索增强生成来迭代修复 Dafny 代码并提升形式规约准确性的工具。
在 Dafny 等支持形式验证的语言中,程序正确性依赖精确的形式规约,而规约中常存在不完整或错误,导致验证失败。人工修复这些规约与代码需要理解验证器反馈并具备领域知识,成本高昂。DafnyRAG 将形式验证反馈作为修复信号,在迭代循环中结合检索增强生成:先从已有 Dafny 代码库中检索相似规约与证明片段,再由生成模型产生候选修复,并依据验证结果决定是否继续迭代。该工具的目标是自动修复 Dafny 代码,提高形式规约的准确性,最终使程序通过验证。摘要未报告具体实验数据与基准结果。
A system that iteratively repairs Dafny code by leveraging formal verification feedback and retrieval-augmented generation to improve formal specification accuracy.
@inproceedings{paperbot12412,
title = {DafnyRAG: A Retrieval-Augmented and Verification-Driven Tool for Automated Repair of Dafny Code},
author = {Yanzhen Liu and Xu Lu and Bin Yu and Haoxiang Zhang and Chu Chen and Meng Wang},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834610}
}