Statistics‐Based Techniques for Software Fault Localization
Zhenyu Zhang, W. Eric Wong · 2023
In this chapter, the authors revisit Tarantula and other spectrum-based fault localization techniques. Tarantula ranks all the statements in a program in descending order of suspiciousness and uses the confidence values to resolve ties. Many kinds of statistics-based fault-localization techniques have been proposed. A key insight is based on the assumption that certain dynamic features of program entities are more sensitive to the differences between the set of failed runs and the set of all runs. Besides choosing a meaningful feature for the desired application domain and developing more accurate coefficient formulas, purifying input is also a research focus. A way to address the issue of coincidental correctness is to completely abandon the use of passed runs. A conventional statistic-based fault localization technique outputs a list of suspicious program elements, the order of which is based on the degree that each is deemed related to a fault.