Abstract: Automatic software fault localization techniques try to identify the codes obtaining potential faults by comparing the difference on behaviors of correct and false running. Each method has its own adaptive preconditions or environments. Considering these preconditions and environments will be helpful to improve the selection processes and the efficiency of methods. By analyzing the impact of class proportion in test suites on the efficiency of methods deeply, this paper selects ten spectrum-based fault localization methods, including Tarantula, Zoltar and so on, as experimental objects to locate the faults in statement-level existing in programs such as space and flex. Moreover, this paper proposes a metric to evaluate the impact on the efficiency of fault localization with the changing of class proportions in test suites. Experimental results show that the impact of class proportion on the efficiency of methods is multiform. Some methods can keep stable efficiency in a range of balance proportions, but others represent highly sensitivity on the efficiency with the changing of balance proportions.

Key words: fault localization, spectrum, class proportion, localization stability

摘要： 软件缺陷定位技术通过比较程序正确与错误运行过程中的不同行为特征，找出导致错误运行行为的代码。任意一种软件缺陷定位算法都有其适用的环境或条件，如果能够在其设计与选择过程中考虑到这些因素，将对提升算法的使用效果起到积极作用。在已有研究的基础上，选择了Tarantula、Zoltar等10种具有代表性的基于程序谱的缺陷定位算法作为研究对象，并通过语句级别的插桩设置，在space、flex等常用程序集上展开实验研究，提出了一种度量指标来评估不同算法受测试用例集类别比例变动的影响。实验结果表明，不同算法受类别比例变动的影响有很大差异，一些算法的定位效率受类别比例变动的影响较小，而其他算法的定位效率在类别比例变动时则呈现出跳变特性。

关键词: 缺陷定位, 程序谱, 类别比例, 定位稳定性