Unit Analysis and Testing

Inductive inference, the automatic syn-Budd80thesis of programs, bears certain ostensible rela-Budd, Timothy A., Richard A. DeMillo, Richardtionships with program testing. For inductive in-J. Lipton, and Frederick G. Sayward. “Theoreticalference, one must take a finite sample of the desiredand Empirical Studies on Using Program Mutationsinput/output behavior of some program and pro-to Test the Functional Correctness of Programs.”duce (synthesize) an equivalent program. In theConf. Record 7th Ann. ACM Symp. on Principles oftesting paradigm, one seeks a finite sample for aProg. Lang. New York: ACM, Jan. 1980, 220-233.function such that any program (in a given set) which computes something other than the objectThis paper presents little-known results on mutationfunction differs from the object function on the finitetesting, both theoretical and empirical. The theoret-sample. In both cases, the finite sample embodiesical section can be safely ignored, except for thesufficient knowledge to isolate the desired programanalysis of decision tables and straight-line Lispfrom all other possibilities. These relationships areprograms. The empirical results are more interest-investigated and general recursion theoreticing, since they provide insight into the mutantproperties of testable sets of functions are exposed.operators used and their success on buggy pro-grams.[Budd82] presents similar theory from a programming language theoretic view. [Rowland81] pre-The theoretical section is useful only for those whosents related theory for a narrower problem domain.wish to pursue mutation testing at an expert level.The empirical section is of some use in demonstrat-This paper requires a strong background in the nota-ing when mutation testing does and does not work.tions and conventions of computability theory. Forexperts only. Budd82Clarke76Budd, Timothy A., and Dana Angluin. “Two No-tions of Correctness and Their Relation to Testing.” Clarke, Lori A. “A System to Generate Test DataActa Informatica 18, 1 (1982), 31-45.and Symbolically Execute Programs.” IEEE Trans.Software Eng. SE-2, 3 (Sept. 1976), 215-222.Abstract: We consider two interpretations for what it means for test data to demonstrate correctness.Abstract: This paper describes a system that at-For each interpretation, we examine under whattempts to generate test data for programs written inconditions data sufficient to demonstrate correct-ANSI Fortran. Given a path, the system symboli-