TACO: test suite augmentation for concurrent programs

Abstract

The advent of multicore processors has greatly increased the prevalence of concurrent programs to achieve higher performance. As programs evolve, test suite augmentation techniques are used in regression testing to identify where new test cases are needed and then generate them. Prior work on test suite augmentation has focused on sequential software, but to date, no work has considered concurrent software systems for which regression testing is expensive due to large number of possible thread interleavings. In this paper, we present TACO, an automated test suite augmentation framework for concurrent programs in which our goal is not only to generate new inputs to exercise uncovered changed code but also to explore new thread interleavings induced by the changes. Our technique utilizes results from reuse of existing test inputs following random schedules, together with a predicative scheduling strategy and an incremental concolic testing algorithm to automatically generate new inputs that drive program through affected interleaving space so that it can effectively and efficiently validate changes that have not been exercised by existing test cases. Toward the end, we discuss several main challenges and opportunities of our approach.