We study the question of consistency of multi-view models in UML and OCL. We first critically survey the large amount of literature that already exists. We find that only limited subsets of the UML/OCL have been covered so far and that consistency checks mostly only cover structural aspects, whereas only few methods also address behaviour. We also give a classification of different techniques for multi-view UML/OCL consistency: consistency rules, the system model approach, dynamic meta-modelling, universal logic, and heterogeneous transformation. Finally, we elaborate cornerstones of a comprehensive distributed semantics approach to consistency using OMG's Distributed Ontology, Model and Specification Language (DOL). consistency problem has already been stated in the early UML days [5,13,14], and has been addressed quite intensively in the literature. In particular, several categorisations for partitioning the consistency problem have been designed: Engels et al. [c26] suggest to distinguish between horizontal (or intra-model) and vertical (or inter-model) consistency, i.e., whether the views are on the same level of abstraction; as well as syntactic (structural well-formedness of the abstract syntax) and semantic consistency (compatibility of behaviour). Mens et al. [29] focus more on the intention of sub-languages and give a classification into structural vs. behavioral diagrams and their use on the specification vs. instance level. Allaki et al.[1] combine these schemes into a typological frame of monovs. multi-diagram, specification vs. instance, and syntactic vs. semantic, and furthermore add a taxonomy of consistency problems in a terminological dimension, mentioning incompleteness, ambiguity, contradiction, incompatibility, and anomaly. From a verification perspective, Hilken et al.[18] present a list of structural and behavioural verification tasks for UML models considering besides consistency the categories of consequence, independence, executability, and reachability. A structural verification task considers a single (integrated) system state only, whereas a behavioural task pertains to a sequence of states. In contrast to [c46], here "[c]onsistency problems are structural problems and do not involve behaviour" [18, p. 122].The large number of approaches to multi-view consistency in the literature has also been reviewed and summarised [3,20,28,40,41,42]. In particular, Torre et al. [40,41] systematically survey existing consistency rules. They find that most rules are syntactic (88.21% in [40] and 81.89% according to the more comprehensive [41]), and that most of the rules are related to class diagrams (71.58%), sequence diagrams (47.37%), and state machine diagrams (42.11%). Moreover, they deplore that "it appears that researchers tend to discuss very similar consistency rules, over and over again", and conclude that "much more work is needed to develop consistency rules for all 14 UML diagrams, in all dimensions of consistency (e.g., semantic and syntactic on the one hand, horizontal, vertical and e...
