Incorporating record subtyping into a relational data model

Most of the current proposals for new data models support the construction of heterogeneous sets. One of the major challenges for such data models is to provide strong typing in the presence of heterogenity. Therefore the inclusion of as much as possible information concerning legal structural variants is needed. We argue that the shape of some part of a heterogeneous scheme is often determined by the contents of some other part of the scheme. This relationship can be formalized by a certain type of integrity constraint we have called attribute dependency. Attribute dependencies combine the expressive power of general sums with a notation that fits into relational models. We show that attribute dependencies can be used, besides their application in type and integrity checking, to incorporate record subtyping into a relational model. Moreover, the notion of attribute dependency yields a stronger assertion than the traditional record subtyping rule as it considers some refinements to be caused by others.
To examine the differences between attribute dependencies and traditional record subtyping and to be able to predict how attribute dependencies behave under transformations like query language operations we develop an axiom system for their derivation and prove it to be sound and complete. We further investigate the interaction between functional and attribute dependencies and examine an extended axiom system capturing both forms of dependencies.