Stephen J. Mellor

What's really going on is that Executable UML is a concurrent specification language.

In its current form UML is designed to support a wide variety of different modelling techniques and formalisms. This is evident, for example, in the state machine formalism which allows both Moore and Mealy formalism with hierarchical states including concurrent sub-states and both synchronous and asynchronous calling semantics. The result of this is not only that almost any state modelling style can be supported but also that many combinations of elements have no defined execution semantics. It is now widely recognised within the UML community, however, that considerable benefit can be gained by forming subsets of the UML with well defined execution semantics. Such subsets can form an “executable UML” which would enable the simulation, execution, testing and ultimately translation of UML models into target code. As part of this movement, work is progressing under the auspices of the OMG towards the definition of “profiles” that define such subsets and towards the more detailed definition of the contents of “actions” including a more precise definition of the execution semantics of UML models.

We build models to increase productivity, under the justified assumption that it's cheaper to manipulate the model than the real thing. Models then enable cheaper exploration and reasoning about some universe of discourse . One important application of models is to understand a real, abstract, or hypothetical problem domain that a computer system will reflect. This is done by abstraction, classification, and generalization of subject-matter entities into an appropriate set of classes and their behavior.

An object in OOA represents a single typical but unspecified instance of something in the real world - any airplane, I don't care which one, as long as it is typical. The object-oriented analyst distinguishes this concept from that of a specified instance: Airplane number N2713A, Air Force One, or The Spirit of St. Louis, for example.

Executable UML is at the next higher layer of abstraction, abstracting away both specific programming languages and decisions about the organization of the software so that a specification built in Executable UML can be deployed in various software environments without change.

Creating a modeling language that is also an executable language has long been a goal of the software community. Many years ago, in 1968 to be exact, while working with software components to successfully develop a telecommunications system, we created a modeling language that was the forerunner to UML. To model components we used sequence diagrams, collaboration diagrams, and state transition diagrams (a combination of state charts and activity diagrams). Our modeling language then seamlessly translated the component models into code. Each code component was in its turn compiled into an executable component that was deployed in our computer system. The computer system was a component management system—thus we had "components all the way down."

While a small domain (consisting of fifty or fewer objects) can generally be analyzed as a unit, large domains must be partitioned to make the analysis a manageable task. To make such a partitioning, we take advantage of the fact that objects on an information model tend to fall into clusters: groups of objects that are interconnected with one another by many relationships. By contrast, relatively few relationships connect objects in different clusters.
When partitioning a domain, we divide the information model so that the clusters remain intact... Each section of the information model then becomes a separate subsystem. Note that when the information model is partitioned into subsystems, each object is assigned to exactly one subsystem.

MDA per se is relaxed about exactly what models it transforms, so long as the modeling language in which the models are expressed can be defined.

I assume that a precisely defined, verifiable, executable, and translatable UML is a Good Thing and leave it to others to make that case... In the summer of 1999, the UML has definitions for the semantics of its components. These definitions address the static structure of UML, but they do not define an execution semantics. They also address (none too precisely) the meaning of each component, but there are "semantic variation points" which allow a component to have several different meanings. Multiple views are defined, but there is no definition of how the views fit together to form a complete model. When alternate views conflict, there is no definition of how to resolve them. There are no defined semantics for actions...
To determine what requires formalization, the UML must distinguish clearly between essential, derived, auxiliary, and deployment views. An essential view models precisely and completely some portion of the behavior of a subject matter, while a derived view shows some projection of an essential view... All we need now is to make the market aware that all this is possible, build tools around the standards defined by the core, executable UML, and make it so...

Executable UML is designed to produce a comprehensive and comprehensible model of a solution without making decisions about the organization of the software implementation. It is a highly abstract thinking tool to aid in the formalization of knowledge, a way of thinking about and describing the concepts that make up an abstract solution to a client problem.

I was astonished to be invited to what became the meeting that originated the Agile Manifesto because my work had always been based around building models... The other signatories were kind enough, back in 2001, to write the manifesto using the word “software” (which can include executable models), not “code” (which is more specific.) As such I felt able, in good conscience, to become a signatory to the Manifesto while continuing to promote executable modeling. Ten years on we have a standard action language for agile modeling.

In the bad old days before MDA, (conceptual) models served only to facilitate communication between customers and developers and act as blueprints for construction. Nowadays, MDA establishes the infrastructure for defining and executing transformations between models of various kinds.

What's the point of having metamodels, and why should you care? Because models must be stated in a way that yields a common understanding among all involved parties, we need a way to specify exactly what a model means. Metamodels allow you to do just that: They specify the concepts of the language you're using to specify a model.