Investigating the Impact of Multiple Dependency Structures on Software Defects
Investigating the Impact of Multiple Dependency Structures on Software Defects
复制标题
DOI：
10.1109/icse.2019.00069
复制
发表时间：
2019-05
期刊：
影响因子：
--
通讯作者：
Di Cui;Ting Liu;Yuanfang Cai;Q. Zheng;Qiong Feng;Wuxia Jin;Jiaqi Guo;YunHuan Qu
中科院分区：
文献类型：
--
作者：
Di Cui;Ting Liu;Yuanfang Cai;Q. Zheng;Qiong Feng;Wuxia Jin;Jiaqi Guo;YunHuan Qu
在过去的几十年中，提出了许多方法来帮助从业者预测或定位有缺陷的文件。这些技术通常使用句法依赖性，历史共同变化或语义相似性。问题在于，尚不清楚这些不同的依赖关系在缺陷预测和本地化方面是否会呈现相似的准确性。在本文中，我们从软件体系结构的角度介绍了对这个问题的系统调查。将涉及每种依赖类型类型的文件视为单个设计空间，我们使用一个DRSPACE对这样的设计空间进行建模。我们为117个Apache开源项目中的每一个都得出了3个DRSpaces，共有643,079个修订提交和101,364个错误报告，并计算了它们与有缺陷文件的交互。实验结果令人惊讶：三种依赖性类型具有明显不同的架构视图，并且它们与有缺陷的文件的相互作用也截然不同。直观地，当用于缺陷预测/本地化时，它们扮演着完全不同的角色。好消息是，这些结构的组合有可能提高缺陷预测/本地化的准确性。总而言之，我们的工作提供了一个新的观点，该观点涉及哪种类型的关系用于缺陷预测/本地化的任务。这些定量和定性的结果也提高了我们对使用不同依赖类型形成的软件质量和建筑视图之间关系的了解。
Over the past decades, numerous approaches were proposed to help practitioner to predict or locate defective files. These techniques often use syntactic dependency, history co-change relation, or semantic similarity. The problem is that, it remains unclear whether these different dependency relations will present similar accuracy in terms of defect prediction and localization. In this paper, we present our systematic investigation of this question from the perspective of software architecture. Considering files involved in each dependency type as an individual design space, we model such a design space using one DRSpace. We derived 3 DRSpaces for each of the 117 Apache open source projects, with 643,079 revision commits and 101,364 bug reports in total, and calculated their interactions with defective files. The experiment results are surprising: the three dependency types present significantly different architectural views, and their interactions with defective files are also drastically different. Intuitively, they play completely different roles when used for defect prediction/localization. The good news is that the combination of these structures has the potential to improve the accuracy of defect prediction/localization. In summary, our work provides a new perspective regarding to which type(s) of relations should be used for the task of defect prediction/localization. These quantitative and qualitative results also advance our knowledge of the relationship between software quality and architectural views formed using different dependency types.