Software crash reproduction is a necessary first step for debugging. Unfortunately, crash reproduction is often difficult and labor intensive. To automate crash reproduction, many approaches have been proposed including record-replay approaches and post-failure-process approaches. Record-replay approaches record software executions and reliably reproduce the recorded executions. However, they usually incur substantial performance overhead, thus, not commonly deployed in practice. Alternatively, post-failure-process approaches perform analysis on crashes only after they have occurred. Therefore they do not incur performance overhead. However, existing post-failure-process approaches still cannot reproduce many crashes in practice due to scalability issues and object creation chall...[ Read more ]

Software crash reproduction is a necessary first step for debugging. Unfortunately, crash reproduction is often difficult and labor intensive. To automate crash reproduction, many approaches have been proposed including record-replay approaches and post-failure-process approaches. Record-replay approaches record software executions and reliably reproduce the recorded executions. However, they usually incur substantial performance overhead, thus, not commonly deployed in practice. Alternatively, post-failure-process approaches perform analysis on crashes only after they have occurred. Therefore they do not incur performance overhead. However, existing post-failure-process approaches still cannot reproduce many crashes in practice due to scalability issues and object creation challenge. In this paper, we propose an automatic crash reproduction framework using collected crash stack traces. The proposed approach combines an efficient backward symbolic execution and a novel method sequence composition approach to generate unit test cases that can reproduce the original crashes without incurring additional runtime overhead. Our evaluation study shows that our approach successfully exploited 38 (59.4%) out of 64 crashes in four open source projects. Among these exploitable crashes, 25 (39.1%) are useful reproductions of the original crashes to reveal the crash triggering bugs. A comparison study also demonstrates that our approach can effectively outperform existing crash reproduction approaches.