Authors:
Dániel Vince
;
Renáta Hodován
and
Ákos Kiss
Affiliation:
Department of Software Engineering, University of Szeged, Dugonics tér 13, 6720 Szeged, Hungary
Keyword(s):
Spectrum-based Fault Localization, Test Case Reduction, Fuzz Testing.
Abstract:
Spectrum-based fault localization (SBFL) is a popular idea for automated software debugging. SBFL techniques use information about the execution of program elements, recorded on a suite of test cases, and derive statistics from them, which are then used to determine the suspiciousness of program elements, thus guiding the debugging efforts. However, even the best techniques can face problems when the statistics are unbalanced. If only one test case causes a program failure and all other inputs execute correctly, as is typical for fuzz testing, then it may be hard to differentiate between the program elements suspiciousness-wise. In this paper, we propose to utilize test case reduction, a technique to minimize unnecessarily large test cases often generated with fuzzing, to assist SBFL in such scenarios. As the intermediate results, or by-products, of the reduction are additional test cases to the program, we use these by-products when applying SBFL. We have evaluated this idea, and ou
r results show that it can help SBFL precision by up to 49% on a real-world use-case.
(More)