Gungnir：用编码智能体演化 SQL 测试生成器

AI 摘要

论文提出 Gungnir，一种利用编码智能体（coding agents）演化 DBMS 测试生成器的方法。其核心是「生成见证」（generation witness）机制：智能体在修改生成器时需提供见证，记录生成目标 SQL 语句所需的数据库设置与语义选择，验证器通过重放这些选择来确认修改后的生成器能复现目标语句，从而在存在幻觉风险下保证新增 SQL 特性支持的正确性。作者在 SQLite、DuckDB 和 PostgreSQL 上测试 Gungnir，共发现 44 个被开发者确认的 bug，其中 34 个已修复。

正文节选

Agentic Evolution of SQL Test Generators Abstract. Automated testing has uncovered many logic bugs in Database Management Systems (DBMSs). Existing tools are typically limited to a few SQL dialects and a subset of these dialects’ features. Coding agents based on Large Language Models (LLMs) could encode the semantics of additional SQL features, which would then require no additional LLM calls during these tools’ run time, thus allowing high efficiency and low cost. However, a key challenge is to

发布时间：2026-10-14 12:00

抓取时间：2026-10-14 15:59

来源机构：arXiv