Finding and breaking test dependencies to speed up test execution

Abstract

Software testing takes up the major part of the build time, which hinders developers' ability to promptly identify and fix problems. Test parallelization is an effective means to speed up test executions, hence improving software development. Effective and sound test parallelization requires that tests are independent or that test dependencies are known in advance. However, current techniques to detect test dependencies are either precise but slow, or fast but inaccurate. Further, available algorithms for test parallelization either over-constraint test executions, which reduces their level of parallelism, or re-execute the same tests multiple times, which increases the execution effort. This research addresses both sides of the problem of speeding up test execution: it aims to devise a practical test detection technique that can suitably balance efficiency and accuracy, and develop a novel technique to break test dependencies which allows both sound and efficient test executions.