Bio-Inspired Optimization of Test Data Generation for Concurrent Software

Concurrent software includes a number of key features such as communication, concurrency, and non-determinism, which increase the complexity of software testing. One of the main challenges is the test data generation. Techniques of search-based software can also benefit concurrent software testing. To do so, this paper adopts a bio-inspired approach, called BioConcST, to support the automatic test data generation for concurrent programs. BioConcST uses a Genetic Algorithm (GA) and an evolutionary strategy adapted to accept genetic information from some bad individuals (test data) in order to generate better individuals. Structural testing criteria for concurrent programs are used to guide the evolution of test data generation. An experimental study was carried out to compare BioConcST with an elitist GA strategy (EGA) in terms of adequacy of testing criteria for message-passing and share-dmemory programs. Twelve concurrent Java programs were included and the results suggest BioConcST is a promising approach, since in all the testing criteria evaluated, it achieved a better coverage and the effect-size measure was large in most cases.