TABLE OF CONTENTS It is obvious that the accuracy of system calculations made from handbook data for components depends heavily on the subdomain breakdown chosen by the component designers. For example, if only a few subdomains are used, the properties measured for those subdomains will be averages over data that may have high variance, and the composition calculations using those averages can be inaccurate. In the extreme case of taking the entire input domain as a single subdomain, the theory reduces to combining single numbers that do not represent the components, as indicated in Section 6.3.
The insight that led testing theorists to look at subdomains in the first place was that a subdomain should group together inputs that are in some sense the same. If this is done for one of the applications in which a property is approximated by a step function, then the approximation is perfect. [8] For example, if the run time of a component within each subdomain S is the same for all points of S, the step function is the actual runtime function. Evidently, the design-from-components paradigm will work better if the handbook entry for a component uses subdomains on which the property of interest does not vary much. One collection of well known subdomains has the most promise: the path subdomains.
Path subdomains represent the strongest intuitive breakdown of a program s input domain. Two inputs are in the same path subdomain if for them...
