- 08 Oct 2007
Improving Automatic Software Fault Localization
To keep software systems dependable, debugging is essential. The diagnosis of software faults, and especially automated software fault localization can contribute to the efficiency of this process. In this research we will mainly focus on single-fault programs, when we constitute a similarity coefficient which is applied in an approach to software fault localization where wemake use of line hit spectra. This new similarity coefficient equation delivers better results than the Tarantula and Ochiai equation, as our evaluation shows by testing them on the Siemens Test Suite benchmark set. We also show, that combining static program slicing with this software fault localization approach, improves the accuracy of the faulty diagnosis even further. Analyses are presented which show why the software coefficient works well in most cases and why it fails to perform in some particular cases. Our equation combined with program slicing requires 42% less code to be inspected than with the Ochiai coefficient and 57% less than with the Tarantula coefficient.