The all-du-paths software testing criterion is the most discriminating of the data ow testing criteria of Rapps and Weyuker. Unfortunately, in the worst case, the criterion requires an exponential number of test cases. To investigate the practicality of the criterion, we develop tools to count the number of complete program paths necessary to satisfy the criterion. This count is an estimate of the number of test cases required. In a case study of an industrial software system, we nd that in eighty percent of the subroutines the all-du-paths criterion is satis ed by testing ten or fewer complete paths. Only one subroutine out of 143 requires an exponential number of test cases.
James M. Bieman, J. Schultz