Experimental evaluation of a fuzzy-set based measure of software correctness using program mutation

Experimental evaluation of software reliability models that depend on the source code of the target program is expensive due to the need for a large sample of programs. The authors have used program mutation to generate many versions of one of the more complex components comprising a hypothetical but realistic nuclear reactor safety control program. Trivial mutants were filtered by using branch and path testing. These programs were used to assess a fuzzy set based measure of program correctness. The results confirmed that the model is conservative. In addition, the experiments provided new insights into the model, including reassessment of its assumptions and directions for refining it