Enhancing Contexts for Automated Debugging Techniques

Most existing automated debugging techniques just focus on selecting a set of suspicious statements that may cause failures and ranking them in terms of suspiciousness. Therefore, these techniques always ignore the contextual information of how suspicious statements behave and propagate in the program. However, the contextual information is useful for discovering and understanding bugs. Hence, this paper proposes a novel approach to enhance contexts for automated debugging techniques. Based on localization results obtained from automated debugging techniques, our approach utilizes program slicing to classify suspicious statements into different contexts, and assigns different suspiciousness to the contexts and their elements. The experimental study shows that our approach can substantially improve debugging effectiveness. Keywords-automate debugging; program spectra; program slicing; statistical analysis.