Λεπτομέρειες βιβλιογραφικής εγγραφής
| Τίτλος: |
HisFuzz: 基于历史缺陷的通用模糊测试用例 生成方法. |
| Alternate Title: |
His Fuzz: general fuzzing test case generation method based on historical defects. |
| Συγγραφείς: |
谢丽霞1, 晏阳维2, 杨宏宇1,2 |
| Πηγή: |
Application Research of Computers / Jisuanji Yingyong Yanjiu. Jul2026, Vol. 43 Issue 7, p2119-2125. 7p. |
| Θεματικοί όροι: |
*Computer software testing, *Software failures, *Language models, *Software measurement, *Compilers (Computer programs) |
| Abstract (English): |
Fuzz testing explores potential defects in language processors by generating diverse test cases. However, existing approaches based on large language model (LLM) still exhibit limitations in generating anomalous programs and ensuring test case validity. To address these issues, this paper proposed a general fuzz test case generation method based on historical defect information (named His Fuzz) and implemented a prototype system. The proposed method leveraged historical defect data from multiple language processors and employed vectorized retrieval and targeted mutation strategies to guide LLM in generating test cases with anomalous characteristics. In addition, it incorporated syntax repair and validity checking mechanisms to improve the compilability and execution success rate of generated test cases. During the testing phase, differential testing techniques were used to analyze behavioral discrepancies among language processors. Experimental results on C, C++, Go, and Python language processors demonstrate that HisFuzz outperforms existing approaches in terms of test case validity and code coverage, achieving an average improvement of 20. 33% in validity and 32. 56% in coverage, while successfully identifying six design defects. These results indicate that HisFuzz can effectively enhance the performance of LLM-based fuzz testing. [ABSTRACT FROM AUTHOR] |
| Abstract (Chinese): |
模糊测试通过生成多样化测试用例发现语言处理器中的潜在缺陷, 但现有基于大语言模型 (large lan-guage model, LLM) 的方法在异常程序生成能力及测试用例有效性方面仍存在不足。针对上述问题, 提出一种基于历史缺陷信息的通用模糊测试用例生成方法 HisFuzz, 并实现原型工具。该方法利用多语言处理器的历史缺陷数据, 通过向量化检索与定向突变策略引导 LLM 生成具有异常特征的测试用例, 并结合语法修复与有效性检查机制, 提高测试用例的可编译性与执行成功率。在测试阶段, 采用差分测试技术分析语言处理器的行为差异。 实验结果表明, 在 C 、 C ++、Go 和 Python 语言处理器的评估中, HisFuzz 在测试用例有效性和代码覆盖率方面均优于现有方法, 平均有效性提升 20.33%, 覆盖率提升 32.56%, 并成功发现 6 个设计缺陷。实验表明, HisFuzz 能够有效提升基于 LLM 的模糊测试性能。 [ABSTRACT FROM AUTHOR] |
| Βάση Δεδομένων: |
Academic Search Index |