Class-Graph Inference for Adaptive Programs

Abstract

Software generators can adapt components to changes in the architectures in which the components operate. The idea is to keep the architecture description separate and let the software generator mix it with specications of each component. Adaptation is done by regeneration: when the architecture changes, the components are regenerated. A software component will usually be written with a particular architecture in mind. This raises the question: how much has it committed to the particular structure of that architecture? To put it in a nutshell: How exible is a given software component? In this paper we study this question in the setting of Lieberherr’s adaptive programming. Lieberherr uses class graphs as the architecture and so-called adaptive programs as the software components. We present a polynomial-time class-graph inference algorithm for adaptive programs. The algorithm builds a representation of the set of class graphs with which a given adaptive program can work. It also decides if the set is non-empty, and if so it computes a particularly simple graph in the solution set. Several toy programs have been processed by a prototype implementation of the algorithm.