Dorylus: An Ant Colony Based Tool for Automated Test Case Generation

Abstract

Automated test generation to cover all branches within a program is a hard task. We present Dorylus, a test suite generation tool that uses ant colony optimisation, guided by coverage. Dorylus constructs a continuous domain over which it conducts independent, multiple objective search that employs a lightweight, dynamic, path-based input dependency analysis. We compare Dorylus with EvoSuite with respect to both coverage and speed using two corpora. The first benchmark contains string based programs, where our results demonstrate that Dorylus improves over EvoSuite on branch coverage and is 50% faster on average. The second benchmark consists of 936 Java programs from SF110 and suggests Dorylus generalises well as it achieves 79% coverage on average whereas the best performing of three EvoSuite algorithms reaches 89%.