Holmes is a statistical tool for automatically root causing failures from code coverage data. Holmes is particularly suited for a scenario where the application has a large test suite and the cause of the failing tests needs to be found. Holmes uses statistical analysis of coverage data to identify program features (blocks, arcs or paths) that strongly correlate with failure. The analysis is designed to detect and eliminate features that just happen to correlate with failure but are not the cause (such as error handling code). The final result of the analysis is a set of bug predictors, features that are likely to be the cause of failures.

Collaborators

• Trishul Chilimbi (MSR Redmond)
• Ben Liblit (Wisconsin)
• Aditya Nori (MSR India)