Adaptive search techniques applied to software testing

Abstract

An experiment was performed in which executable assertions were used in conjuction with search techniques in order to test a computer program automatically. The program chosen for the experiment computes a position on an orbit from the description of the orbit and the desired point. Errors were interested in the program randomly using an error generation method based on published data defining common error types. Assertions were written for program and it was tested using two different techniques. The first divided up the range of the input variables and selected test cases from within the sub-ranges. In this way a “grid” of test values was constructed over the program's input space. The second used a search algorithm from optimization theory. This entailed using the assertions to define an error function and then maximizing its value. The program was then tested by varying all of them. The results indicate that this search testing technique was as effective as the grid testing technique in locating errors and was more efficient. In addition, the search testing technique located critical input values which helped in writing correct assertions.