CrashAutomata: an approach for the detection of duplicate crash reports based on generalizable automata

Crash reporting systems are useful tools that allow users to report system failures, subsequently contacting the appropriate support group for resolution. As a software system grows and becomes more versatile, the number of crashes increases. A large software company receives typically thousands of crashes a day, which make it difficult for software engineers to address these reports in a timely manner. Fortunately, not all reports are new; many of them are duplicates of previously reported crashes. Research has shown that early detection of duplicate reports can reduce the effort and time it takes to handle crash reports. In this paper, we propose a new approach for detecting duplicate crash reports, called CrashAutomata. CrashAutomata builds a model from historical crash reports (more precisely their stack traces) that is used to classify an incoming report. The model is based on variedlength n-grams and automata. Unlike existing techniques, CrashAutomata takes advantage of the generalization aspect of automata, making it possible to build a representative model of crash reports, reducing the number of false positives. When applied to crash reports of the Firefox system, CrashAutomata results in very high precision and recall. It also outperforms CrashGraph, a leading technique in the detection of duplicate crash reports.