超越规则式变异测试：基于大语言模型的测试感知变异体生成
原标题：Beyond Rule-Based Mutation Testing: Test-Aware Mutant Generation Using Large Language Models
AI 摘要
卡尔加里大学的研究者提出「测试感知的变异体生成」方法，让大语言模型在单次提示中同时接收问题描述、标准解法和基础单元测试，生成能通过基础测试但被扩展测试捕获的非平凡变异体。在 HumanEval 和 MBPP 基准上对 Gemini 3.1 Pro、Gemini 3 Flash、GPT 5.1 Codex Mini、GPT 4.1 Mini、Qwen3-32B 五个模型评估，验证故障率分别达 87.7% 和 79.1%，远高于测试盲提示（12.2%、23.0%）和传统规则工具 mutmut（4.4%、5.7%）。该方法将变异体生成从无目标注入缺陷转向有效发现测试套件弱点，并降低了每个验证故障的计算成本。
正文节选
[Page 1] Beyond Rule-Based Mutation Testing: Test-Aware Mutant Generation Using Large Language Models Nils Kiele # Department of Electrical and Software Engineering, University of Calgary, AB, Canada Zainab Saad Department of Electrical and Software Engineering, University of Calgary, AB, Canada Zirui Wang Department of Electrical and Software Engineering, University of Calgary, AB, Canada Steve Drew Department of Electrical and Software Engineering, University of Calgary, AB, Canada Samira