ISSN 1000-1239 CN 11-1777/TP

• 软件技术 •

### 基于切片谱的错误定位框架影响因素分析

1. 1(中国矿业大学计算机科学与技术学院 江苏徐州 221116);2(南通大学计算机科学与技术学院 江苏南通 226019) (xiaolinju@gmail.com)
• 出版日期: 2014-12-01
• 基金资助:
基金项目：国家自然科学基金项目(61202006,61340037)；中央高校基本科研业务费专项资金资助项目(2013QNB17)；南通市应用研究计划基金项目(BK2014055)；江苏省高校自然科学研究基金项目(12KJB520014)；江苏省研究生培养创新工程基金项目(CXZZ12_0935)

### Factor Analysis of Influence for Fault Localization Framework Based on Slice Spectrum

Ju Xiaolin1,2, Jiang Shujuan1, Chen Xiang2, Zhang Yanmei1, Shao Haoran2

1. 1(School of Computer Science and Technology, China University of Mining and Technology, Xuzhou, Jiangsu 221116); 2(School of Computer Science and Technology, Nantong University, Nantong, Jiangsu 226019)
• Online: 2014-12-01

Abstract: Fault localization is an important task of program debugging. The statistical fault localization techniques, based on slice spectrum, can reduce the effort of fault localization by analyzing the program slices. However, the effectiveness of these techniques depends on slice selection criteria and suspiciousness computing formulas. Thus, we propose a fault localization framework to evaluate the influence on the effectiveness of fault localization by the above two factors. Firstly, we compute the full slices of failed runs and the execution slices of passed runs, respectively. Secondly, we give a definition of the similarity between slice spectrums, and develop a set of slices selection criteria to construct a hybrid slice spectrum. Finally, we choose a suspiciousness evaluation formula and then generate a fault location report. To investigate the impact of the two factors (i.e., the similarity between slices, and the evaluation formulas) on the effectiveness of fault localization, we conduct empirical study on several classical Java benchmarks consisting of more than 90 faults. The experimental result suggests that the performance of Wong, Russel&Rao, and Binary cannot be influenced by the similarity of slice spectrum. However, our proposed formula (HSS), Tarantula, DStar, Naish1, and Naish2 can perform better on slice spectrum of lower similarity.