Large database systems (e.g., federations, warehouses) are multi-layer—i.e., a combination of base databases and (virtual or physical) view databases*. Smaller systems use views for layers that hide detailed physical and conceptual structures. We argue that most databases would be more effective if they were more user-centered—i.e., if they allowed users, administrators, and application developers to work mostly within their native view. To do so, we need first class views—views that support most of the metadata and operations available on source tables.
First class views could also make multi-tier object architectures (based on objects in multiple tiers of servers) easier to build and maintain. The views modularize code for data services (e.g., query, security) and for coordinating changes with neighboring tiers. When data in each tier is derived declaratively, one can generate some of these methods semi-automatically.
Much of the functionality required to support first class views can be generated semi-automatically, if the derivations between layers are declarative (e.g., SQL, rather than Java). We present a framework where propagation rules can be defined, allowing the flexible and incremental specification of view semantics, even by non-programmers. Finally, we describe research areas opened up by this approach.
*The term "layer" was chosen for its naturalness in these familiar architectures. However, in the general case, derivation edges can form an arbitrary digraph.
