HOMI: Searching Higher Order Mutants for Software Improvement

This paper introduces HOMI, a Higher Order Mutation based approach for Genetic Improvement of software, in which the code modification granularity is finer than in previous work while scalability remains. HOMI applies the NSGAII algorithm to search for higher order mutants that improve the non-functional properties of a program while passing all its regression tests. Experimental results on four real-world C programs shows that up to 14.7% improvement on time and 19.7% on memory are found using only First Order Mutants. By combining these First Order Mutants, HOMI found further improvement in Higher Order Mutants, giving an 18.2% improvement on the time performance while keeping the memory improvement. A further manual analysis suggests that 88% of the mutation changes cannot be generated using line based ‘plastic surgery’ Genetic Improvement approaches.