Title:

Constraint of Model-Based Test Generation Using Input-Output Analysis

Author:

Patrick J. Schroeder

Address:

Illinois Institute of Technology, Chicago, IL., USA

Abstract:

Automating the software testing process can result in better software products. Test automation can improve the quality of the software, and reduce its cost and time to market. Many tasks in the testing process may be automated, including test generation, test execution, and results evaluation. The focus of this research is on test generation, or the automated creation of tests for a software program. Specifically, we are investigating the generation of black-box, or functional, tests from a model derived from the software's specification.

A current problem for this type of test generator is that they generate too many tests, far exceeding the capacities of manual or automated test execution [Fews99]. To use a test generator effectively, controls must be put in place to limit, or "constrain," the number of tests generated. Currently, techniques for constraining these generators rely on heuristics and statistical methods to reduce the number of tests created. In our experience, constraining test generators using these techniques can have a detrimental impact on the fault detection capability of the generated test set. Important tests may be inadvertently omitted, and seemingly redundant tests can be included in the test set. The question we are researching is: Can we find a way to constrain automatic test generators while maintaining the fault detection capability of the test set? Our initial experience with constraining test generation using information from the software's implementation looks promising for the specific class of program and problem type we have identified.