Some Remarks about Random Testing

The use of a pseudo-random number generator to produce test cases for software appears not to be widely used, although the method has been reported in the literature for many years [1, 2, 4]. This paper draws on experience with this technique in two areas: compilers and software components. The use of a pseudo-random number generator to produce test cases which are expected to reflect actual usage of the software is not discussed here (we would call this statistical testing). Although such testing would have the advantage that reliability data could be produced, in practice with most software of any complexity, it is not possible to model accurately enough the distribution of real input data. Hence the test case generation considered here is for conventional testing purposes, rather than predicting reliability. All the work here has exploited a portable and efficient random number generator [5]. Use of this generator has the advantage that one can reproduce test cases from the three integer seed values, which are input to the test case generator, no matter how complex the test cases are.