Can Pairwise Testing Perform Comparably to Manually Handcrafted Testing Carried Out by Industrial Engineers?

摘要

Testing is an important activity in engineering of industrial software. Forsuch software, testing is usually performed manually by handcrafting testsuites based on specific design techniques and domain-specific experience. Tosupport developers in testing, different approaches for producing good testsuites have been proposed. In the last couple of years combinatorial testinghas been explored with the goal of automatically combining the input values ofthe software based on a certain strategy. Pairwise testing is a combinatorialtechnique used to generate test suites by varying the values of each pair ofinput parameters to a system until all possible combinations of thoseparameters are created. There is some evidence suggesting that these kinds oftechniques are efficient and relatively good at detecting software faults.Unfortunately, there is little experimental evidence on the comparison of thesecombinatorial testing techniques with, what is perceived as, rigorous manuallyhandcrafted testing. In this study we compare pairwise test suites with testsuites created manually by engineers for 45 industrial programs. The testsuites were evaluated in terms of fault detection, code coverage and number oftests. The results of this study show that pairwise testing, while useful forachieving high code coverage and fault detection for the majority of theprograms, is almost as effective in terms of fault detection as manual testing.The results also suggest that pairwise testing is just as good as manualtesting at fault detection for 64% of the programs.