January 93 - Object-Oriented Software Engineering
Object-Oriented Software Engineering
Addison-Wesley ISBN 0-201-54435-0
Sometimes I feel software methodologies pop up like mushrooms after a rainy day. For a while the common complaint in the object–oriented field of computing was the lack of formal analysis methodologies, while the structured programming field had it's SA/SD, SADTs and RDDs. And lo and behold, one year later we were presented with the Booch methodology named OOD (Object Oriented Design), Hierarchical Object Oriented Design (HOOD), Coad and Yourdon's Object Oriented Analysis (OOA), Solution Based Modeling (SBM) by Goldstein and Alger, and many more.
We will eventually enter a stage in the object–oriented paradigm where a clash of titans will shake out the winners and the losers. Still, the main problem with many of the OO-labeled methodologies is the lack of a comprehensive view concerning a software project. Design is nice, but you also need an overall methodology that spawns the analysis stage, design itself, implementation and, most importantly, testing as well.
At this year's OOPSLA Jacobson's Object-Oriented Software Engineering book was a popular book, so I didn't resist the temptation to pick up a copy. Jacobson's methodology, OOSE, tries to cover many of the earlier mentioned aspects of a software project.
What is OOSE?
In general, unless you can understand a methodology within a paragraph or two, my personal opinion is that we are dealing with snake oil.
In order to understand OOSE as a methodology we have to take a look at industrial engineering environments that work productively according to Jacobson. One such example he presents is the building construction industry.
In this engineering environment we have:
- an architecture, a defined approach selected from a universe of approaches, such as how to construct a sauna;
- a method, how to apply the selected architecture concepts (like how to put together the sauna walls);
- a process, how to scale up the method to industrial activity (how to mass produce sauna walls);
- and finally tools that support the architecture stage, methods and processes.
Jacobson wants to create an overall methodology that would provide a more engineering and stage-oriented process to produce software. The methodology tries to keep alive the creative part of software analysis/design, and at the same time provide an approach for making software development a rational industrial process. Most importantly, the bias for this is based on system changes that typical industrial process focuses on (compare with Japanese car manufacturers). Let's take a look at each of the following stages of software construction using Jacobson's methodology: architecture, analysis, design, and implementation.
A very important property of a software system is its internal structure-the architecture. A good architecture makes the system easy to understand, change, test and maintain. So the properties of the architecture will determine how the system will behave during its lifetime.
Note, however, that architecture also contains the following elements: the analysis, the design and the implementation models. A method is a planned procedure by which a specified goal is achieved (step by step). A basic requirement for a good method is that it simplifies the development of systems of a particular architecture. This has been for a long time the sore point in the object oriented computing world. And still today many OO methodologies treat this requirement quite superficially according to Jacobson. He states that such methodologies assume that the objects can easily be found directly from the activity.
Often we do have such cases, but as many of us realize it requires a strong background in object modeling before 'finding the objects' becomes second nature. I'm afraid I didn't find a good solution to this problem in the book. Jacobson hints at a high level construct called a process that is usable for scaling up the design at a later stage. However, it seems the issue of finding natural objects and object interactions is something that is hard to easily mechanize, even in OOSE.
Object Oriented Analysis
As with any other analysis stage, the idea is to obtain an understanding of the application based on functional requirements. Object–oriented analysis can be characterized as an iteration between analyzing the system's behavior and its information. Here the methodology should provide help with:
- finding the objects
- organizing the objects
- describing how the objects interact
- defining the operations of the objects
- defining the objects internally
Jacobson states that objects can be found as naturally occurring entities in the application domain. He's using the noun paradigm where a noun that exists in the domain is often a good start to learn the terminology for the object domain. When we learn more about the domain, the real objects will pop up. This is where I do think we are really dealing with an imperfect scientific methodology; it is questionable whether this can formalized into an exact procedure due to the classical problems of categorization.
Jacobson suggests organizing objects based on different criteria (active/passive, physical/conceptual, temporary/permanent/persistent, part/whole, generic/specific and so on...) as a means to conceptualize the naturally formed objects. This is just another method to learn about the domain in order to find the right objects.
The next step is to organize the objects based on various criteria. For instance, the similarity between objects might resolve into a hierarchy of objects. Another classification of objects may be based on how objects work together with other objects (for instance a house is built of doors, windows, walls…).
Booch's keynote at this year's OOPSLA was related to the art of classification, and he said that tomorrow's designers will spend more and more time working with categorization of objects.
Object interaction is defined from the way objects fit into the system. This is where the OOSE use case technique comes into picture (more about this later). The object's interface can be decided from such scenarios.
Operations on objects come naturally when we look at the object's interface. The operations can be identified directly from the application, when we consider what we want to do with the items we model. We have primitive operations (Create, Add, Delete…) as well as complex ones such as putting together a report of information based on various objects.
Finally the object should be defined internally, which includes defining the information that each object must hold.
OOSE and Object Oriented Analysis
OOSE has the concept of actors and use cases. In order to map a requirement specification to the model, we have actors that play various use case scenes, and these use cases will form the operations.
The actors represent interactions with the system, they represent everything that needs to exchange information with the system. Actors are actually outside the system, so we don't need to define them explicitly. Note that a user (an end user) could have various actor roles while using the system. This actor does a number of various operations which will trigger a sequence of transactions in a dialogue with the system. A single sequence is called a use case. A transaction is finished when the use case instance requires input from an actor.
The system model will be use case driven: when we want to change the system behavior, we remodel the appropriate actor and use case. The whole system architecture will be determined by what the user wishes to do with the total system.
NOTE: This is one of the core ideas behind OOSE
This modeling view fits very well with prototyping, user interface driven application work and re-specification of projects. Also, we can talk with the users directly and find out their requirements and preferences, eliminating the dreadful gap between specifications and the actual implementation (a sickness that has killed thousands of software projects).
Now, the use case model will control the formation of all other OOSE models, such as domain models, analysis, design, implementation and testing. The functionality specified by the use cases is then structured into a logical, implementation-level environment and further re-defined in the design model so it is resistant to change. The use cases will be implemented by the source code and should also give us tools when testing the system. Also, the use case should give us support when writing manuals and other operational instructions.
This is closely related to the notion of patterns and pattern languages for documentation use. For instance Ralph Johnson presented a paper at this year's OOPSLA-92 concerning how to structure the documentation of a framework by using a set of patterns that describes the purpose of the framework without having to know in detail how everything works.
Object Oriented Design and OOSE
During the design stage we create a design model that is a refinement of the analysis one. The analysis model was based on ideal conditions. Now we deal with reality.
We didn't want to introduce the environment earlier because we didn't want it to affect the basic structure of the system, Neither did we want the problem blurred by the complexity usually triggered when looking at the implementation environment.
So the design model is a formalization of the analysis space, where we adapt the analysis part so it works with the environment. It's a question of refining the model so that it is straight-forward to write the actual code. However, we still have issues related to changes in the model (such as future performance requirements, new features, computer language issues), so we need to develop a new model.
So when does the analysis model work end and the design work starts (a classical dilemma)? There's not a good answer; it really depends on each specific application. The goal is to not redo any work at a later phase that has been already done.
OOSE uses the concept of blocks that will describe how the code should be produced (see Figure 2). These blocks are design objects, and they are drawn as rectangles. One block normally aims at implementing one analysis object. However, the key issue is that these blocks are not the same objects as the analysis objects in order to keep the ideal analysis model intact.
The blocks are abstractions of the actual implementation of the system. We need to refine how the blocks interact or communicate during execution. The OOSE term is called stimulation. A stimulus is sent from one block to another to trigger an execution in that block.
To describe a sequence of stimuli, OOSE uses interaction diagrams. These describe how several blocks communicate by sending stimuli to each other (see Figure 3 on the next page). These diagrams describe in detail, for each use case, which stimuli will be sent how and in what order. A use case becomes a sequence of stimuli sent between the blocks.
In addition, the methodology uses the concept of subsystems in order to manage the design. Subsystems can contain several objects, which might be other subsystems. So we might end up with a hierarchical structuring of the system in order to manage the complexity.
This step implies that the analysis model has been designed and it is time for coding. The analysis has (we hope) provided us with an ideal structure which we shall now try to keep intact as long as possible. If everything works well the objects identified during the analysis should also appear in the design. This is called in OOSE traceability. OOSE uses the concept of components in order to map the analysis stage objects into real-life source code based entities.
Note also that this stage actually does not require an object oriented language, even if it helps in mapping the design to the implementation. If the block design was done properly, it is very easy to code into objects. And by reusing objects in forms of components we could achieve more powerful constructs (aggregates).
An important message Jacobson states in his book is the issue of systems development as part of a larger activity. This work does not take place in isolation. For instance in the banking industry data processing is just one sub-part of a larger activity providing a certain functionality set.
The output from systems development is a set of descriptions that includes the source code, diagrams, flow charts, and so on. All of these descriptions must be self-explanatory to facilitate reusability. Also, the knowledge of how to organize and manage projects must be documented so it could be made reusable (SBM is also targeting this interesting new concept of reusable design and project plans).
Using these concepts, we might eventually attain a rational approach of software development. A software house might create a library of these configuration sets, and each sub-project will receive a special combination of service packages with relevant information which has been assembled from the library of descriptions. New releases could reuse descriptions from previous releases in a controlled manner.
All this demonstrates the need for high level CASE tools and systems.
OK, OOSE made my day, but will it help me?
Well, it depends what you want from a methodology. There are two schools of thought concerning mix-and-match of various methodologies. The CASE manufacturers/providers kick and scream and say that one needs to stick to one single methodology, theirs.
The academic field generally holds that it is better to stick to one methodology in order to avoid any Byzantine solutions based on convoluted and mixed design methods.
The practical programmer is keen on creating his/her own methodology, by taking (or stealing) the best parts from various design systems (though sometimes it's darned hard to plot out all kinds of cloud figures with MacDraw without using the CASE tool from ACME Inc).
Then large corporations issue decrees that every software department will take a holy trip to the West/East coast for three days of training, and the defined methodology should be used till the end of the world.
Me? I've been looking around for a simple, nice, understandable and highly flexible methodology that could be used for personal computer application development work. I'm not stating that OOSE is the answer. However, many of the basic OOSE ideas are valid. For instance, the user is in control of an application, and the use cases based on actors are highly suitable for analysis of both what the GUI provides and what the application should achieve for the user. The book provides a lot of good ideas that the interested designer could try to implement, maybe in a pilot project, and if the end results are positive then the methodology is suitable for future work.
As is the case of most methodologies, refinement is the key-if you find a working formula, refine it instead of jumping to the next buzzword in the computing field.
I'm sure that all the other OO CASE and methodology vendors will step up and suggest that their methodology works where OOSE fails. And they might be right or wrong, I'm not the judge in this issue. Hey, I'm just a book reviewer. Anyway, would I recommend reading the book? As there are an unlimited number of OO books today, and time is precious when you want to ship your product tomorrow, I would suggest that if you are shopping around for a methodology, read the book for good ideas about the issues related to selecting a useful methodology. OOSE should also work for small scale projects easily-without the purchase of expensive UNIX workstation CASE tools.
The other reason I recommend that programmers read the book has to do with the background of OOSE/Objectory. It has been used in large-scale real-time system design, and 20+ years of blood, sweat and tears have certainly added a lot of understanding and insight about issues related to software construction. Admittedly, we are dealing with experience that sometimes does not map 1:1 to the personal computer application design. However, as Adele Goldberg lately stated, PC developers are discovering that the principles of object modeling will make an important difference because they have 'discovered' the need to design before coding. This is because it's maybe the first time the PC industry has to build systems instead of smaller applications.
Finally, the book contains a nice model of actors and use cases that could be used with smaller application work. I've recently seen many engineering specifications where use cases have been used to map the issues related to the analysis and design of the product. 1
- Documenting Frameworks using Patterns, Ralph E. Johnson, OOPSLA'92 Conference Proceeedings
- Object-Oriented Software for the Macintosh, Goldstein & Alger
- "Wishful Thinking", Adele Goldberg, Object Magazine, Nov-Dec 1992