Introducing probability into model-based software debugging

摘要

This paper introduces probability into a formal framework for software debugging to isolate the most likely causes of a detected misbehavior by using a probabilistic model that describes a working piece of software. Our approach is based on dynamic dependencies between statements and statistics of program runs. The well known Siemens benchmark is used for empirical purpose. We conclude that our approach has considerable improvements over available model-based software debugging in terms of diagnosis quality, and is statistically more effective in fault localization than competitors like Tarantula, SOBER, CT and PPDG.