Search-based multi-paths test data generation for structure-oriented testing

This paper presents a new fitness function to generate test data for a specific single path, which is different from the predicate distance applied by most test data generators based on genetic algorithms (GAs). We define a similarity between the target path and execution path to evaluate the quality of the populations. The problem of the most existing generators is to search only one target data a time, wasting plenty of available interim data. We construct another fitness function combined with the single path function, which can drive GA to complete covering multi-paths to avoid the reduplicate searching and utilize the interim populations for different paths. Several experiments are taken to examine the effectiveness of both the single path and multi-path fitness functions, which evaluate the functions' performance with the convergence ability and consumed time. Results show that the two functions perform well compared with other two typical path-oriented functions and the multi-paths approach retrenches the searching actually.