CRII: SHF: Regression Testing for Projects with Distributed Software Histories

开发人员实践回归测试--针对每个项目提交运行测试--以检查项目更改不会破坏任何功能。虽然回归测试很重要，但由于测试的数量和提交的数量，回归测试是昂贵的。回归测试选择（RTS）技术通过跳过不受最近更改影响的测试来加速回归测试，回归测试优先级（RTP）技术重新排序测试以更快地运行失败的测试。现有的回归技术只分析两个相邻的提交，有效地假设了线性的软件历史。然而，版本控制系统的革命改变了软件历史的形状。分布式软件历史是分支和合并的复杂图形，这与现有技术的简单视图不匹配。虽然现有的回归技术可以berun，如果分布式软件的历史是线性的，这些technologesunderutilize的数据在分布式software history.To加速回归测试，PI建议设计的technologesthat利用数据在整个分布式software history。本项目提出了四个改进RTS和RTP的任务：（1）非相邻提交重用--发现每次分析中可能使用的最优提交，而不是总是分析相邻提交;（2）多提交分析--设计分析两个以上提交的方法，而不是总是分析两个提交;（3）命令感知方法-专门化创建每个提交而不是命令感知的各种命令的方法;（4）统一实施和评估--共享技术间的实施和成果，并对开源和工业项目的技术进行评估。改进RTS和RTP的更广泛的影响是提高开发人员的生产力，减少测试阶段的资源使用。

Developers practice regression testing -- running tests against eachproject commit -- to check that project changes do not break anyfunctionality. While important, regression testing is expensive due tothe number of tests and the number of commits. Regression testselection (RTS) techniques speed up regression testing by skipping torun tests that are not affected by recent changes, and regression testprioritization (RTP) techniques reorder tests to run failing testsfaster. Existing regression techniques analyze only two adjacentcommits, effectively assuming a linear software history. However, therevolution in version-control systems changed the shape of softwarehistories. Distributed software histories are complex graphs ofbranches and merges, which do not match the simplistic view of theexisting techniques. Although existing regression techniques can berun if a distributed software history is linearized, these techniquesunderutilize the data available in the distributed software history.To speed up regression testing, the PI proposes to design techniquesthat utilize the data available in the entire distributed softwarehistory. This project proposes four tasks to improve RTS and RTP: (1)non-adjacent commit reuse - discover likely optimal commit to be usedin each analysis rather than always analyzing adjacent commits; (2)multi-commit analyses - design methods that analyze more than twocommits rather than always analyzing only two commits; (3)command-aware methods - specialize the methods for various commandsthat create each commit rather than be command-unaware; and (4)unified implementation and evaluation - share the implementation andresults among techniques, and evaluate the techniques on open-sourceand industrial projects. The broader impacts of improving RTS and RTPare to increase developers' productivity and reduce the resource usageduring testing phases.