Nested and specialized associations

Abstract

UML supports specialization of associations, but with sparse semantics. It is recognized that specialization is different from subsetting of association end sets, and UML indicates that it has a meaning similar to specialization for classes (or classifiers in general). Recent efforts on the introduction of associations into programming languages have claimed, with convincing examples, that specialization of associations can not be defined similar to specialization for classes. In this paper we demonstrate that these examples really call for the notion of nested associations; in addition it is demonstrated that other examples call for real specialization (as for classes) of associations. The two notions: nested associations and specialization of associations have been implemented in a prototype tool for an experimental modeling language, and together they open for rich modeling of associations similar to what we have for classes.