Effective mutant reduction using fine-grained mutation operators

Abstract

Although mutation analysis is important for various software analysis tasks, there exists no practical mutation tools for modern, complex, real-world C programs. I have developed MUSIC (MUtation analySIs tool with high Configurability and extensibility) which generates mutants for modern complex real-world C programs. I have conducted a case study on Siemens benchmark programs and a modern real-world C program cURL to compare MUSIC with Milu, Proteum in terms of applicability and number of stillborn (i.e. syntactically illegal) mutants generated. In this case study, MUSIC successfully generates mutants without any stillborn mutants. Another serious obstacle for mutation analysis is the huge cost of running test suites on a large number of mutants. To resolve this problem, I have proposed a new mutation operator-based mutant reduction technique REFINER which applies cost-considerate linear regression (i.e., CLARS) on finegrained mutation operators. Also, I have applied REFINER to predict hard-to-kill mutation score which is more valuable to measure test suite quality than commonly used mutation score. The experiment results show that, while sustaining accurate prediction power to estimate hardto-kill mutation score, REFINER selects far fewer mutants than CLARS on the traditional mutation operators (i.e., 2.0% vs. 16.5%). Also, REFINER predicts hard-to-kill mutation score 4.5, 4.4, and 4.3 times more accurate than mutant reduction techniques that use random selection, Offutt’s four mutation operators selection, and, only SSDL mutation operator respectively.