The Problem of Justification of Empirical Hypotheses in Software Testing

摘要

This paper takes part in the methodological debate concerning the nature and the justification of hypotheses about computational systems in software engineering by providing an epistemological analysis of Software Testing, the practice of observing the programs' executions to examine whether they fulfil software requirements. Property specifications articulating such requirements are shown to involve falsifiable hypotheses about software systems that are evaluated by means of tests which are likely to falsify those hypotheses. Software Reliability metrics, used to measure the growth of probability that given failures will occur at specified times as new executions are observed, is shown to involve a Bayesian confirmation of falsifiable hypotheses on programs. Coverage criteria, used to select those input values with which the system under test is to be launched, are understood as theory-laden principles guiding software tests, here compared to scientific experiments. Redundant computations, fault seeding models and formal methods used in software engineering to evaluate test results are taken to be instantiations of some epistemological strategies used in scientific experiments to distinguish between valid and non-valid experimental outcomes. The final part of the paper explores the problem, advanced in the context of the philosophy of technology, of defining the epistemological status of software engineering by conceiving it as a scientifically attested technology.