The use of Data Flow Information for the Selection and Evaluation of Software Test Data

Candidate: Frankl,Phyllis G.

Abstract

Two families of software test data adequacy criteria, each based on data flow analysis, are defined for programs written in Pascal. Their formal properties are investigated and interactive software testing tools based on them are described. The first of these families, the data flow testing criteria, was previously defined for programs written in a simple language. We extend the definitions to apply to programs written in Pascal. The data flow testing criteria are based purely on the syntax of the program being tested. They require that the test data execute certain paths from program points at which variables are defined to program points at which those definitions are used. We describe the design and implementation of a software testing tool, ASSET, based on the data flow testing criteria. A serious weakness of the data flow testing criteria is that for some programs there exists no set of test data which is adequate for testing the program according to these criteria. This problem arises due to unexecutable paths in the program. The second family of criteria, the feasible data flow testing criteria, circumvent this problem by eliminating from consideration those definition-use associations which can never be exercised. We show that certain formal properties of the feasible data flow testing criteria differ significantly from those of the data flow testing criteria. Since it is undecidable whether a given set of test data satisfies a given feasible data flow testing criterion, feasible data flow testing cannot be fully automated. However, it can be partially automated. We describe a heuristic method, the path expression method, which attempts to determine whether a given definition-use association can be exercised. The path expression method is based on a combination of data flow analysis and symbolic evaluation. We introduce a new symbolic evaluation technique which is more general, but essentially no more expensive, than symbolic execution. The path expression method, along with ASSET, constitute a tool which partially automates feasible data flow testing.