This chapter is from the book
This chapter is from the book
This chapter is from the book
UML
The Unified Modeling Language, or UML, was first published in 1997 by the Object Management Group. UML unified three earlier approaches for graphically depicting software systems: the Booch method, the Object Modeling Technique, and the Object-Oriented Software Engineering method. The advantage of the UML was that it provided a standard set of notational conventions for describing aspects of a software system. Before the UML was published, different authors used different graphical elements to mean the same thing. Three examples are shown in Figure 1-19. The method described in Grady Booch's 1990 book, Object-Oriented Analysis and Design with Applications, represented a class by a cloud; the OMT method described in the 1991 book, Object-Oriented Modeling and Design, by James Rumbaugh and his colleagues, represented a class by a rectangle; and the 1992 book, Object-Oriented Software Engineering: A Use Case Driven Approach, by Ivar Jacobson and his colleagues, advocated representing a class by a little circle and distinguished diagrammatically between entity classes, controller classes, and interface classes. Many other approaches also existed at that time. UML succeeded in eliminating this "Tower of Babel"—almost all competing diagramming approaches vanished rapidly from the marketplace when UML appeared.
)
Figure 1-19 Different representations for a class
On publication, UML became increasingly popular as a technique for documenting the early phases of software development, especially those using object-oriented technologies. Class diagrams, use case diagrams, and sequence diagrams were especially popular for documenting the results of object-oriented analysis and object-oriented design.
Figures 1-20 through Figure 1-22 show how to use UML to analyze the operation of a very simplified public library.
)
Figure 1-20 Use case diagram for simple library
)
Figure 1-21 Class diagram for simple library
)
Figure 1-22 Sequence diagrams for simple library
The meaning of these diagrams is relatively informal. Being an analysis model, this set of diagrams does not exactly represent anything that happens in the software system. Instead, it helps the developer to make some early decisions about what information will be represented in the software and how that information may be collected together and flow around when the system interacts with its environment. Translating the analysis model into an exact design for the actual software involves working out many details, such as the design of the database, the design of the classes that represent the business logic, the mapping between business logic and database classes, the design of the user interface, the messages that flow between clients and servers, and so on. Traces of the analysis model will be found in the design, but the detailed correspondence between the analysis model and the eventual programs, schemas, and definitions that constitute the running software will be complex.
When UML emerged during the 1990s, mainstream thinking about object-oriented development assumed that there would be a relatively simple continuity between an object-oriented analysis and a corresponding object-oriented design. Several methodologies proposed that the way to get from the analysis to the design was simply to add detail while retaining the basic shape of the analysis. For simple examples, where there is a single computer implementing a simple non-distributed application, this can work, especially when no data persistence is involved.
The design of UML itself is actually based on this concept of adding implementation detail. The UML specification defines the ability to express the kind of detail found in an object-oriented programming language; for example, class members can be marked with the Java-inspired visibility values of public, private, protected, or package, and operations can have detailed signatures and so on. This helps to map a UML model to program code, especially if the programming language is Java. Note that there are many inconsistencies between the details of UML and Microsoft's Common Language Runtime, which make it more difficult to map UML effectively to the popular .NET languages Visual Basic and C#. When UML is used for a more abstract purpose such as analysis, these implementation details have to be ignored, because they are meaningless.
UML does offer limited extension facilities, called profiles, stereotypes, tagged values, and constraints. Stereotypes, tagged values, and constraints are mechanisms that add labels and restrictions to UML models to indicate that a UML concept is being used to represent something else. So, for example, a UML class could be labeled as a <<resource>>, or even as a <<webpage>>—the symbols <<>> are conventionally used to indicate that a stereotype is being used. But labeling a UML concept does not change anything else about it—a class still has attributes and operations, inheritance, and the rest of the built-in features.
A UML Profile is a packaged set of stereotypes, tagged values, and constraints that can be applied to a UML model. A tool can make use of the profile information to filter or hide elements but may not delete unwanted elements; a profile is effectively a viewing mechanism. These facilities do allow a limited amount of customization of UML for particular domains, and of course individual UML tool vendors can go beyond the published standard to provide increased levels of customization.
However, the world has moved on apace since UML was defined. The Internet and World Wide Web have matured, most of the computers in the world are connected together, and a multitude of new standards and technologies has emerged, especially XML and Web Services. In 2007 and beyond, the likely platform for implementing a business system will involve many distributed components executing in different computers. Logic and data are replicated for scalability and load balancing. Legacy systems are accessed on mainframes and servers. Firewalls and routers are configured to maintain security and connectivity. Browsers and smart clients are distributed to many different devices and appliances. Common artifacts in this world, such as Web Service Definition Language (WSDL) or configuration files, have no standard representations in UML. Although stereotypes and profiles can be used to apply UML in domains for which it was not designed, such an approach gives cumbersome results. In such a world, the transformation from a simple object-oriented analysis to a detailed system design is far too complex to be thought of simply as "adding detail." Different approaches are needed.
If UML is not convenient to be used directly, what happens if we open up the definition of UML, remove all of the parts we don't need, add new parts that we do need, and design a language specifically tailored for the generation task that we want to accomplish? In short, what would happen if we had an environment for constructing and manipulating graphical modeling languages? The answer is that we would eliminate the mismatches and conceptual gaps that occur when we use a fixed modeling language, and we would make our development process more seamless and more efficient. That is the approach adopted in DSL Tools.
Instead of thinking about UML as a single language, we prefer to think of it as a set of reusable diagrammatic conventions, each of which can be applied to a particular kind of situation that we might encounter during software development. For example, sequence charts such as those in Figure 1-22 might be used to describe the flow of messages between applications in a distributed system, the flow of invocations between objects in an application, or even information interchange between departments in an organization. In the first case, the vertical lines on the diagram represent applications, in the second case they represent objects, and in the third case they represent departments.
Note also that it is not only end users that benefit from clean domain-specific abstractions. Developers who build tools that generate code and other artifacts from models and keep models coordinated with one another, need to access model data; providing APIs that work directly in terms of the abstractions of the problem domain is critical to productivity for developers. Developers want the API for the logical data center to give them direct access to the properties of an IIS server or a SQL Server database. Similarly, they want the API for the sequence charts to talk directly about applications, objects, or departments. They'd like to write strongly typed code, such as this:
foreach (Department dept in message.Receiver.SubDepartments)
{
// generate some artifacts
}
This contrasts with having to reinterpret a model intended for other purposes (such as a UML model), which can give rise to code like this:
Lifeline lifeline = message.Receiver;
if (lifeline.Object.Label = "Department")
{
Department receiver = lifeline.Object.Element as Department;
if (receiver != null)
{
foreach (Department dept in receiver.SubDepartments)
{
// generate some artifacts
}
}
}
else
{
// handle errors
}