Data Flow Analysis Based Defect Correlation

Static analysis has become an important procedure of the whole software testing process. It contains two major parts-defect detection and defect reviewing. As an inherent shortcoming of static analysis, defect detecting tools cannot be both sound and reliable. Decrease of false negatives usually brings about increase of false positives, which thus requires careful distinguishment during manual review process. For large projects, the manual review process could be quite consuming and therefore reduce the defect detection efficiency.In this paper, we examined the warnings reported by DTS and found a large number of defects dependent to each other. Identifying redundant warnings could recude manul efforts. We propose a Region-based Symbolic Memory Model (RSMM) to compute symbolic expressions of a certain Inspecting Point (IP) and further cluster different IPs based on the EOI’s symbolic expressions and their last modification points. Then we select one dominant warning for each cluster in a data flow analysis approach. If the dominant warning is a false positive, so are all the warnings in its cluster. Experimental results show20.5%warnings are redundant in the review process and would on average lead to a corresponding15-25%reduction in manual review efforts.