Correlating Adaptation Eeort with Functional Distance

While it is abundantly clear that black box reuse ooers the best return on investment in terms of quality gains and productivity gains, it is equally clear (on the basis of empirical evidence) that black box reuse accounts for a small percentage of code that is developed with reuse. Applications that are developed with reuse typically include no more than 15 % of black box-reused code, vs 60 to 65 % of white box-reused code. White box reuse raises two crucial questions, which we attempt to address in this paper: rst, given a speciication and a set of candidate components, how can we identify those components that minimize adaptation eeort (to satisfy the speciication); second , given that we have identiied an optimal adaptation candidate, how can we determine whether it is best to adapt the selected component to satisfy the given speciication or to develop a solution to the given speciication from scratch. To address these questions we deene a set of semantic (functional) measures of distance , and attempt to correlate them to adaptation eeort; this paper reports on the current status of our research, which is in progress. Black box reuse consists of reusing software artifacts verbatim, without inspecting their internal representation; it is believed that on average black box reuse costs about 20% of new development, hence saves 80% of development cost, in addition to providing quality gains 9]. White box reuse costs considerably more than black box reuse, typically in the 60% range, and provides less quality assurance , but has a non-negligible advantage: it accounts for a larger percentage of the volume of any application, typically in the 60% range, vs. 15% for black box reuse 9]. Hence white box reuse is expected to remain an important part of software reuse practice. The eecient conduct of white box reuse is contingent upon two key decisions: Given a speciication K that we wish to satisfy and two candidate components C and C 0 that are known not to satisfy K, which of C or C 0 costs less to adapt in order to satisfy K? Given a speciication K and an adaptation candidate C, does it cost less to adapt C to satisfy K or to develop a solution to K from scratch. In order to adequately support these decisions, we need means to estimate the manpower required to carry out an adaptation. Traditional software …