Building a Better Backtrace: Techniques for Postmortem Program Analysis

After a program has crashed, it can be difficult to reconstruct why the failure occurred, or what actions led to the error. We propose a family of analysis techniques that use the evidence left behind by a failed program to build a time line of its possible actions from launch through termination. Our design can operate with zero run time instrumentation, or can flexibly incorporate a wide variety of artifacts such as stack traces and event logs for increased precision. Efficient demanddriven algorithms are provided, and the approach is well suited for incorporation into interactive debugging support tools.