The application of Evolutionary Algorithms to structural test data generation, known as Evolutionary Testing, has to date largely focused on programs with input-output behavior. However, the existence of state behavior in test objects presents additional challenges for Evolutionary Testing, not least because certain test goals may require a search for a sequence of inputs to the test object. Furthermore, state-based test objects often make use of internal variables such as boolean flags, enumerations and counters for managing or querying their internal state. These types of variables can lead to a loss of information in computing fitness values, producing coarse or ... [more] flat fitness landscapes. This results in the search receiving less guidance, and the chances of finding required test data are decreased.This paper proposes an extended approach based on previous works. Input sequences are generated, and internal variable problems are addressed through hybridization with an extended Chaining Approach. The basic idea of the Chaining Approach is to find a sequence of statements, involving internal variables, which need to be executed prior to the test goal. By requiring these statements are executed, information previously unavailable to the search can be made use of, possibly guiding it into potentially promising and unexplored areas of the test object’s input domain. A number of experiments demonstrate the value of the approach.

DOI

Bibtex Entry

@inproceedings{McMinn2005,
  author    = "McMinn, Phil and Holcombe, Mike",
  title     = "Evolutionary Testing of State-Based Programs",
  booktitle = "Genetic and Evolutionary Computation Conference (GECCO 2005)",
  pages     = "1013--1020",
  year      = "2005",
  publisher = "ACM"
}