Opportunities and challenges for deep constraint languages

Structural models are often augmented with additional well-formedness constraints to rule out unwanted configurations of instances. These constraints are usually written in dedicated constraint languages specifically tailored to the conceptual framework of the host modeling language, the most well-known example being the OCL constraint language for the UML. Many multi-level modeling languages, however, have no such associated constraint language. Simply adopting the OCL for such multi-level languages is not a complete strategy, though, as the OCL was designed to support the UML's two-level class/instance dichotomy, i.e., it can only define constraints which restrict the properties of the immediate instances of classes, but not beyond. The OCL would consequently not be able to support the definition of deep constraints that target remote or even multiple classification levels. In fact, no existing constraint language can address the full range of concerns that may occur in deep modeling using the Orthogonal Classification Architecture (OCA) as an infrastructure. In this paper we consider what these concerns might be and discuss the syntactical and pragmatic issues involved in providing full support for them in deep modeling environments.