Generating Reproducible and Replayable Bug Reports from Android Application Crashes

Manually reproducing bugs is time-consuming and tedious. Software maintainers routinely try to reproduce unconfirmed issues using incomplete or no informative bug reports. Consequently, while reproducing an issue, the maintainer must augment the report with information - such as a reliable sequence of descriptive steps to reproduce the bug - to aid developers with diagnosing the issue. This process encumbers issue resolution from the time the bug is entered in the issue tracking system until it is reproduced. This paper presents Crash Droid, an approach for automating the process of reproducing a bug by translating the call stack from a crash report into expressive steps to reproduce the bug and a kernel event trace that can be replayed on-demand. Crash Droid manages trace ability links between scenarios' natural language descriptions, method call traces, and kernel event traces. We evaluated Crash Droid on several open-source Android applications infected with errors. Given call stacks from crash reports, Crash Droid was able to generate expressive steps to reproduce the bugs and automatically replay the crashes. Moreover, users were able to confirm the crashes faster with Crash Droid than manually reproducing the bugs or using a stress-testing tool.