UML Applied

UML Applied
Object Oriented Analysis and Design Using the UML

A Course Companion

2 UML Applied - Object Oriented Analysis and Design using the UML

Authors and Contacts

Please contact info@ariadnetraining.co.uk, or see the website at
www.ariadnetraining.co.uk for further details about Ariadne’s supporting
training courses. Comments and feedback are welcome.

ã2001 Ariadne Training Limited www.ariadnetraining.co.uk


Contents

AN INTRODUCTION TO THE UML 7

What is the UML? 7
A Common Language 7
Summary 9

THE UML WITHIN A DEVELOPMENT PROCESS 10

The UML as a Notation 10
The Waterfall Model 10
The Spiral Model 12
Iterative, Incremental Frameworks 13
Inception 13
Elaboration 14
Construction 14
Transition 15
How Many Iterations? How Long Should They Be? 15
Time Boxing 16
Typical Project Timings 16
The Rational Unified Process 17
Summary 18

OBJECT ORIENTATION 19

Structured Programming 19
The Object Orientated Approach 22
Encapsulation 23
Objects 23
Terminology 24
The Object Oriented Strategy 24
Summary 25

AN OVERVIEW OF THE UML 26

The Use Case Diagram 27
The Class Diagram 28
Collaboration Diagrams 29
Sequence Diagram 30
State Diagrams 31
Package Diagrams 32
Component Diagrams 33
Deployment Diagrams 34
Summary 34

THE INCEPTION PHASE 35



THE ELABORATION PHASE 37

Deliverables 37
Summary 38

USE CASE MODELLING 39

Actors 39
The Purpose of Use Cases 40
Use Case Granularity 41
Use Case Descriptions 43
Use Cases at the Elaboration Phase 43
Finding Use Cases 44
Joint Requirements Planning Workshops (JRP) 44
Brainstorming Advice 45
Summary 45

CONCEPTUAL MODELLING 46

Finding Concepts 47
Extracting Concepts From Requirements 47
The Conceptual Model in the UML 48
Finding Attributes 49
Guidelines for Finding Attributes 50
Associations 50
Possible Cardinalities 51
Building the Complete Model 51
Summary 53

RANKING USE CASES 54

Summary 55

THE CONSTRUCTION PHASE 56

Construction 56
Summary 57

THE CONSTRUCTION PHASE : ANALYSIS 58

Back to the Use Cases 58
1. Pre-Conditions 59
2. Post Conditions 59
3. Main Flow 59
Alternate Flows 60
Exception Flows 60
The Complete Use Case 61
The UML Sequence Diagram 61
Summary 63



THE CONSTRUCTION PHASE : DESIGN 64

Design - Introduction 64
Collaboration of Objects in Real Life 65
Collaboration Diagrams 66
Collaboration Syntax : The Basics 66
Collaboration Diagrams : Looping 68
Collaboration Diagrams : Creating new objects 68
Message Numbering 68
Collaboration Diagrams : Worked Example 69
Some Guidelines For Collaboration Diagrams 72
Chapter Summary 73

DESIGN CLASS DIAGRAMS 74

Crediting and Debiting Accounts 74
Step 1 : Add Operations 75
Step 2 : Add Navigability 75
Step 3 : Enhance Attributes 75
Step 4 : Determine Visibility 76
Aggregation 76
Composition 77
Finding Aggregation and Composition 77
Summary 77

RESPONSIBILITY ASSIGNMENT PATTERNS 78

The GRASP Patterns 78
What is a pattern? 78
Grasp 1 : Expert 78
Grasp 2 : Creator 80
Grasp 3 : High Cohesion 81
Grasp 4 : Low Coupling 83
Grasp 5 : Controller 86
Summary 87

INHERITANCE 88

Inheritance – the basics 88
Inheritance is White Box Reuse 90
The 100% Rule 91
Substitutability 91
The Is-A-Kind-Of Rule 92
Example - Reusing queues through inheritance 92
Problems With Inheritance 94
Visibility of Attributes 95
Polymorphism 96
Abstract Classes 97
The Power of Polymorphism 98
Summary 99



SYSTEM ARCHITECTURE - LARGE AND COMPLEX SYSTEMS 100

The UML Package Diagram 100
Elements Inside a Package 101
Why Packaging? 101
Some Packaging Heuristics 102
Expert 102
High Cohesion 102
Loose Coupling 102
Handling Cross Package Communication 102
The Facade Pattern 104
Architecture-Centric Development 105
Example 105
Handling Large Use Cases 106
The Construction Phase 107
Summary 107

MODELLING STATES 108

Example Statechart 108
State Diagram Syntax 109
Substates 110
Entry/Exit Events 111
Send Events 111
Guards 111
History States 112
Other Uses for State Diagrams 112
Summary 113

TRANSITION TO CODE 114

Synchronising Artifacts 114
Mapping Designs to Code 115
Defining the Methods 117
Step 1 118
Step 2 118
Step 3 119
Step 4 119
Mapping Packages into Code 119
In Java 119
In C++ 120
The UML Component Model 120
Ada Components 121
Summary 121

BIBLIOGRAPHY 123



Chapter 1
An Introduction to the UML

What is the UML?
The Unified Modelling Language, or the UML, is a graphical modelling language that
provides us with a syntax for describing the major elements (called artifacts in the
UML) of software systems. In this course, we will explore the main aspects of the
UML, and describe how the UML can be applied to software development projects.

Through to its core, UML leans towards object oriented software development, so in
this course, we will also explore some of the important principles of object
orientation.

In this short chapter, we’ll look at the origins of the UML, and we’ll discuss the need
for a common language in the software industry. Then we will start to look at how to
exploit the UML on a software project.

A Common Language
Other industries have languages and notations, which are understood by every
member of that particular field.

Figure 1 - A Mathematical Integral

Although the picture above is a fairly simple drawing (a stylised "S" figure),
mathematicians the world over recognise instantly that I am representing an integral.
Although this notation is simple, it masks a very deep and complicated topic (though
perhaps not as deep as the concept represented by the figure of eight on its side!) So
the notation is simple, but the payoff is that mathematicians all around the world can
clearly and unambiguously communicate their ideas using this, and a small collection



of other symbols. Mathematicians have a common language. So do musicians,
electronic engineers, and many other disciplines and professions.

To date, Software Engineering has lacked such a notation. Between 1989 and 1994, a
period referred to as the “method wars”, more than 50 software modelling languages
were in common use – each of them carrying their own notations! Each language
contained syntax peculiar to itself, whilst at the same time, each language had
elements which bore striking similarities to the other languages.

To add to the confusion, no one language was complete, in the sense that very few
software practitioners found complete satisfaction from a single language!

In the mid 1990’s, three methods emerged as the strongest. These three methods had
begun to converge, with each containing elements of the other two. Each method had
its own particular strengths:
• Booch was excellent for design and implementation. Grady Booch had worked
extensively with the Ada language, and had been a major player in the
development of Object Oriented techniques for the language. Although the Booch
method was strong, the notation was less well received (lots of cloud shapes
dominated his models - not very pretty!)
• OMT (Object Modelling Technique) was best for analysis and data-intensive
information systems.
• OOSE (Object Oriented Software Engineering) featured a model known as Use
Cases. Use Cases are a powerful technique for understanding the behaviour of an
entire system (an area where OO has traditionally been weak).

In 1994, Jim Rumbaugh, the creator of OMT, stunned the software world when he left
General Electric and joined Grady Booch at Rational Corp. The aim of the partnership
was to merge their ideas into a single, unified method (the working title for the
method was indeed the "Unified Method").

By 1995, the creator of OOSE, Ivar Jacobson, had also joined Rational, and his ideas
(particularly the concept of "Use Cases") were fed into the new Unified Method - now
called the Unified Modelling Language1. The team of Rumbaugh, Booch and
Jacobson are affectionately known as the "Three Amigos".

Despite some initial wars and arguments, the new method began to find favour
amongst the software industry, and a UML consortium was formed. Heavyweight
corporations were part of the consortium, including Hewlett-Packard, Microsoft and
Oracle.

The UML was adopted by the OMG2 in 1997, and since then the OMG have owned
and maintained the language. Therefore, the UML is effectively a public, non-
proprietary language.

1
Officially, the spelling is "modeling", but I favour the English spelling
2
The OMG are the Object Management Group, an industry wide, non profit making standards body.
See www.omg.org for full details.



Summary
The UML is a graphical language for capturing the artifacts of software
developments.

The language provides us with the notations to produce models.

The UML is gaining adoption as a single, industry wide language.

The UML was originally designed by the Three Amigos at Rational Corp.

The language is very rich, and carries with it many aspects of Software Engineering
best practice.



Chapter 2
The UML within a Development Process

The UML as a Notation
The Three Amigos, when developing the UML, made a very clear decision to remove
any process based issues from the language. This was because processes are very
contentious - what works for company A might be a disaster for company B. A
defence company requires much more documentation, quality and testing than (say)
an e-commerce company. So the UML is a generic, broad language enabling the key
aspects of a software development to be captured on "paper".

In other words, the UML is simply a language, a notation, a syntax, whatever you
want to call it. Crucially, it does not tell you how to develop software.

To learn how to use the UML effectively, however, we will follow a simple process
on this course, and try to understand how the UML helps at each stage. To start with,
let's have a look at some common software processes.

The Waterfall Model

Figure 2 - The traditional “Waterfall” model



The waterfall model prescribes that each stage must be complete before the next stage
can commence.

This simplistic (and easy to manage) process begins to break down as the complexity
and size of the project increases. The main problems are:

• Even large systems must be fully understood and analysed before progress can be
made to the design stage. The complexity increases, and becomes overwhelming
for the developers.

• Risk is pushed forward. Major problems often emerge at the latter stages of the
process – especially during system integration. Ironically, the cost to rectify errors
increase exponentially as time progresses.

• On large projects, each stage will run for extremely long periods. A two-year long
testing stage is not necessarily a good recipe for staff retention!

Figure 3 –Over time on the waterfall, both the risks and the cost to rectify errors
increase

Also, as the analysis phase is performed in a short burst at the outset of the project, we
run a serious risk of failing to understand the customer’s requirements. Even if we
follow a rigid requirements management procedure and sign off requirements with the
customer, the chances are that by the end of Design, Coding, Integration and Testing,
the final product will not necessarily be what the customer wanted.

Having said all the above, there is nothing wrong with a waterfall model, providing
the project is small enough. The definition of "small enough" is subjective, but
essentially, if the project can be tackled by a small team of people, with each person
able to understand every aspect of the system, and if the lifecycle is short (a few



months), then the waterfall is a valuable process. It is much better than chaotic
hacking!

In summary, the waterfall model is easy to understand and simple to manage. But the
advantages of the model begin to break down once the complexity of the project
increases.

The Spiral Model
An alternative approach is the spiral model. In this approach, we attack the project in
a series of short lifecycles, each one ending with a release of executable software:

Figure 4 - a spiral process. Here, the project has been divided into five phases,
each phase building on the previous one and with a running release of software
produced at the end of each phase

With this approach:
• The team are able to work on the entire lifecycle (Analysis, Design, Code, Test)
rather than spending years on a single activity
• We can receive early and regular feedback from the customer, and spot potential
problems before going too far with development
• We can attack risks up-front. Particularly risky iterations (for example, an iteration
requiring the implementation of new and untested technology) can be developed
first
• The scale and complexity of work can be discovered earlier
• Changes in technology can be incorporated more easily
• A regular release of software improves morale
• The status of the project (eg – “how much of the system is complete”) can be
assessed more accurately

The drawbacks of a spiral process are



• The process is commonly associated with Rapid Application Development, which
is considered by many to be a hacker's charter.

• The process is much more difficult to manage. The Waterfall Model fits in closely
with classic project management techniques such as Gantt charts, but spiral
processes require a different approach.

To counteract the drawbacks of the spiral technical, let's look at a similar, but more
formal approach called an Iterative, Incremental Framework.

Philippe Kruchten’s Whitepaper (reference [5], available from Rational
Software’s website) explores the traps many managers are likely to face on
their first iterative development.

Iterative, Incremental Frameworks
The Iterative, Incremental Framework is a logical extension to the spiral model, but is
more formal and rigorous. We will be following an Iterative, Incremental Framework
through the rest of this course.

The framework is divided into four major phases: Inception; Elaboration;
Construction and Transition. These phases are performed in sequence, but the
phases must not be confused with the stages in the waterfall lifecycle. This section
describes the phases and outlines the activities performed during each one.

Figure 5 - the four phases of an Iterative, Incremental Framework

Inception
The inception phase is concerned with establishing the scope of the project and
generally defining a vision for the project. For a small project, this phase could be a
simple chat over coffee and an agreement to proceed; on larger projects, a more
thorough inception is necessary. Possible deliverables from this phase are:
• A Vision Document
• An initial exploration of the customer’s requirements
• A first-cut project glossary (more on this later)
• A Business Case (including success criteria and a financial forecast, estimates of
the Return on Investment, etc)
• An initial risk assessment



• A project plan

We’ll explore the inception phase in a little detail when we meet the case study in
Chapter 4.

Elaboration
The purpose of elaboration is to analyse the problem, develop the project plan further,
and eliminate the riskier areas of the project. By the end of the elaboration phase, we
aim to have a general understanding of the entire project, even if it is not necessarily a
deep understanding (that comes later, and in small, manageable chunks).

Two of the UML models are often invaluable at this stage. The Use Case Model helps
us to understand the customer’s requirements, and we can also use the Class Diagram
to explore the major concepts our customer understands. More on this shortly.

Construction
At the construction phase, we build the product. This phase of the project is not
carried our in a linear fashion – rather, the product is built in the same fashion as the
spiral model, by following a series of iterations. Each iteration is our old friend, the
simple waterfall.3 By keeping each iteration as short as possible, we aim to avoid the
nasty problems associated with waterfalls.

Figure 6 - The Construction Phase consists of a series of mini waterfalls

3
Note that at the inception and elaboration phases, prototypes can be built. These prototypes can be
developed in exactly the same way – as a series of mini waterfall iterations. However, for this course,
we will keep the inception and elaboration phases simple and use the waterfalls for construction only.



At the end of as many iterations as possible, we will aim to have a running system
(albeit, of course, a very limited system in the early stages). These iterations are called
Increments, hence the name of the framework!

Transition
The final phase is concerned with moving the final product across to the customers.
Typical activities in this phase include:
• Beta-releases for testing by the user community
• Factory testing, or running the product in parallel with the legacy system that the
product is replacing
• Data takeon (ie converting existing databases across to new formats, importing
data, etc)
• Training the new users
• Marketing, Distribution and Sales
The Transition phase should not be confused with the traditional test phase at the end
of the waterfall model. At the start of Transition, a full, tested and running product
should be available for the users. As listed above, some projects may require a beta-
test stage, but the product should be pretty much complete before this phase happens.

How Many Iterations? How Long Should They Be?
A single iteration should typically last between 2 weeks and 2 months. Any more than
two months leads to an increase in complexity and the inevitable “big bang”
integration stage, where many software components have to be integrated for the first
time.

A bigger and more complex project should not automatically imply the need for
longer iterations – this will increase the level of complexity the developers need to
handle at any one time. Rather, a bigger project should require more iterations.

Some factors that should influence the iteration length include: (see Larman [2],
pp447-448).
• Early development cycles may need to be longer. This gives developers a chance
to perform exploratory work on untested or new technology, or to define the
infrastructure for the project.
• Novice staff
• Parallel developments teams
• Distributed (eg cross site) teams [note that Larman even includes in this category
any team where the members are not all located on the same floor, even if they are
in the same building!]
To this list, I would also add that a high ceremony project will generally need longer
iterations. A high ceremony project is one which might have to deliver a lot of project
documentation to the customer, or perhaps a project which must meet a lot of legal
requirements. A very good example would be any defence related project. In this case,
the documentary work will extend the length of the iteration – but the amount of



software development tackled in the iteration should still be kept to a minimum to
avoid our chief enemy, complexity overload.

Time Boxing
A radical approach to managing an iterative, incremental process is Time Boxing. This
is a rigid approach which sets a fixed time period in which a particular iteration must
be completed by.

If an iteration is not complete by the end of the timebox, the iteration ends anyway.
The crucial activity associated with timeboxing is the review at the end of iteration.
The review must explore the reasons for any delays, and must reschedule any
unfinished work into future iterations.

Larman (ref [2]) gives details on how to implement timeboxing. One of his
recommendations is that the developers be responsible for (or at least, have a large say
in) setting which requirements are covered in each iteration, as they are the ones who
will have to meet the deadlines.

Implementing timeboxing is difficult. It requires a culture of extreme discipline
through the entire project. It is extremely tempting to forgo the review and overstep
the timebox if the iteration is “99%” complete when the deadline arrives. Once a
project succumbs to temptation and one review is missed, the whole concept begins to
fall apart. Many reviews are missed, future iterations planning becomes sloppy and
chaos begins to set in.

Some managers assume that timeboxing prevents slippage. It does not. If an iteration
is not complete once the timebox has expired, then the unfinished work must be
reallocated to later iterations, and the iteration plans are reworked – this could include
slipping the delivery date or adding more iterations. However, the benefits of
timeboxing are:
• The rigid structure enforces planning and replanning. Plans are not discarded once
the project begins to slip
• If timeboxes are enforced, there is less of a tendency for the project to descend
into chaos once problems emerge, as there is always a formal timebox review not
too far away
• If panic sets in and developers start to furiously hack, the hacking is stemmed
once the review is held
Essentially, timeboxing allows the entire project to regularly “stand back” and take
stock. It does not prevent slippage, and requires strong project management to work.

Typical Project Timings
How long should each of the four phases last? This is entirely up to individual
projects, but a loose guideline is 10% inception, 30% elaboration, 50% construction
and 10% transition.



Figure 7 - Possible timings for each phase. This example shows the length of each
phase for a two year project.

The Rational Unified Process
The Rational Unified Process (the RUP) is the most famous example of an Iterative,
Incremental Lifecycle in use at the moment. The RUP was developed by the same
Three Amigos that developed the UML, so the RUP is very complementary to the
UML.

Essentially, Rational appreciate that every project is different, with different needs.
For example, for some projects, a tiny Inception Phase is appropriate, whereas for
defence projects, the Inception phase could last years.

To this end, the RUP is tailorable, and enables each phase of the process to be
customised. The RUP also defines the roles of everyone on the project very carefully
(in the shape of so-called Workers - again, these are tailorable to the project's needs).

Rational Corp produce a product to help projects work with the RUP. Full details can
be found at www.rational.com. Essentially, the RUP project is an on-line, hypertext
guide to every aspect of the RUP. Rational provide 30 day trials of the product.




Figure 8 - Screenshot from RUP 2000 ( Rational Corp)

The precise advantages and disadvantages of the RUP are beyond the scope of this
course. However, the core of the RUP, the Iterative, Incremental Lifecycle will be
followed throughout this course to illustrate the key aspects of the UML models.

For more details on the RUP, Philippe Kruchten’s book The Rational Unified
Process– An Introduction (ref 1) covers the subject in detail.

Summary
An Iterative, Incremental Framework offers many benefits over traditional processes.

The Framework is divided into four phases - Inception, Elaboration, Construction,
Transition.

Incremental development means to aim for running code at the end of as many
iterations as possible.

Iterations can be timeboxed - a radical way of scheduling and reviewing iterations.

The rest of this course will focus on the Framework, and how the UML supports the
deliverables of each phase in the Framework.



Chapter 3
Object Orientation

In this chapter we will look at the concept of Object Orientation4 (OO). The Unified
Modelling Language has been designed to support Object Orientation, and we'll be
introducing Object Oriented concepts throughout this course. Before we begin looking
at the UML in depth, however, it is worth introducing OO and looking at the
advantages that OO can offer to Software Development.

Structured Programming
First of all, let's examine (in very rough terms) how software systems are designed
using the Structured (sometimes called Functional) approach.

In Structured Programming, the general method was to look at the problem, and then
design a collection of functions that can carry out the required tasks. If these
functions are too large, then the functions are broken down until they are small
enough to handle and understand. This is a process known as functional
decomposition.

Most functions will require data of some kind to work on. The data in a functional
system was usually held in some kind of database (or possibly held in memory as
global variables).

As a simple example, consider a college management system. This system holds the
details of every student and tutor in the college. In addition, the system also stores
information about the courses available at the college, and tracks which student is
following which courses.

A possible functional design would be to write the following functions:

add_student5
enter_for_exam
check_exam_marks
issue_certificate
expel_student

4
I'll use the phrase Object Orientation to denote Object Oriented Design and/or Object Oriented
Programming
5
I'm using underscores to highlight the fact that these functions are written in code.



We would also need a data model to support these functions. We need to hold
information about Students, Tutors, Exams and Courses, so we would design a
database schema to hold this data.6

Figure 9 - Simple Database Schema. The dotted lines indicate where one set of
data is dependent on another. For example, each student is taught by several
tutors.

Now, the functions we defined earlier are clearly going to be dependent on this set of
data. For example, the add_student function will need to modify the contents of
Students. The issue_certificate function will need to access the Student data (to
get details of the student requiring the certificate), and the function will also need to
access the Exam data.

6
Note that throughout this chapter, I am not using a formal notation to describe the concepts



The following diagram is a sketch of all the functions, together with the data, and
lines have been drawn where a dependency exists:

Figure 10 - Plan of the functions, the data, and the dependencies

The problem with this approach is that if the problem we are tackling becomes too
complex, the system becomes harder and harder to maintain. Taking the example
above, what would happen if a requirement changes that leads to an alteration in the
way in which Student data is handled?

As an example, imagine our system is running perfectly well, but we realise that
storing the Student's date of birth with a two digit year was a bad idea. The obvious
solution is to change the Date of Birth field in the Student table, from a two-digit
year to a four-digit year.

The serious problem with this change is that we might have caused unexpected side
effects to occur. The Exam data, the Course data and the Tutors data all depend (in
some way) on the Student data, so we might have broken some functionality with our
simple change. In addition, we might well have broken the add_student,
enter_for_exams, issue_certificate and expel_student functions. For example,
add_student will certainly not work anymore, as it will be expecting a two digit year
for date of birth rather than four.

So we have a large degree of potential knock-on problems. What is far, far worse is
that in our program code, we cannot easily see what these dependencies actually are.

How many times have you changed a line of code in all innocence, without realising
that you've inadvertently broken apparently unrelated functionality?

The costly Year 2000 problem (The Millennium Bug) was caused by exactly this
problem. Even though the fix should be simple (make every year occupy four digits



instead of two), the potential impacts of these minor changes had to be investigated in
detail.7

The Object Orientated Approach
OO tries to lessen the impact of this problem by simply combining related data and
functions into the same module.

Looking at Figure 10 above, it is apparent that the data and functions are related. For
example, the add_student and expel_student functions are clearly very closely related
to the Student data.

The following figure shows the full grouping of the related data and functions, in the
form of modules:

Figure 11 - Related data and functions placed in modules

A couple of points to note about this new modular system of programming:

• More than one instance of a single module can exist when the program is running.
In the college system, there would be an instance of Student for every student
that belongs to the college. Each instance would have its own values for the data
(certainly each would have a different name).

• Modules can talk to other modules by calling each other's functions. For
example, when the add function is called in Student, a new instance of the
Student module would be created, and then the add_attendee function would be
called from the appropriate instances of the Course module.

7
This doesn't mean that I am implying that all non-OO Cobol systems are a load of rubbish, by the
way. There is nothing wrong with structured programming. My suggestion in this chapter is that OO
provides a method of building more robust software as our systems get larger and more complex.



Encapsulation
Crucially, only the instance that owns an item of data is allowed to modify or read it.
So for example, an instance of the Tutor module cannot update or read the age data
inside the Student module.

This concept is called Encapsulation, and enables the structure of the system to be far
more robust, and avoid the situation as described previously, where a small change to
a data member can lead to rippling changes.

With Encapsulation, the programmer of (say) the Student module can safely make
changes to the data in the module, and rest assured that no other modules are
dependent upon that data. The programmer might well need to update the functions
inside the module, but at least the impact is isolated to the single module.

Objects
Throughout this chapter, I have referred to these collections of related data and
functions as being modules. However, if we look at the characteristics of these
modules, we can see some real world parallels.

Objects in the real world can be characterised by two things: each real world object
has data and behaviour. For example, a television is an object and posses data in the
sense that it is tuned to a particular channel, the scan rate is set to a certain value, the
contrast and brightness is a particular value and so on. The television object can also
do things. The television can switch on and off, the channel can be changed, and so
on.

We can represent this information in the same way as our previous software
modules:

Figure 12 - The data and behaviour of a television

In some sense, then, real world objects can be modelled in a similar way to the
software modules we discussed earlier.

For this reason, we call the modules Objects, and hence we have the term Object
Oriented Design/Programming.



Since our software systems are solving real world problems (whether you are working
on a College booking system, a Warehouse Management System or a Weapons
Guidance System), we can identify the objects that exist in the real world problem,
and easily convert them into software objects.

In other words, Object Orientation is a better abstraction of the Real World. In
theory, this means that if the problem changes (ie the requirements change, as they
always do), the solution should be easier to modify, as the mapping between the
problem and solution is easier.

Terminology
The data for an object are generally called the Attributes of the object. The different
behaviours of an object are called the Methods of the object. Methods are directly
analogous to functions or procedures in programming languages.

The other big jargon term is Class. A class is simply a template for an object. A class
describes what attributes and methods will exist for all instances of the class. In the
college system we described in this chapter, we had a class called Student.

The attributes of the Student Class were name, age, etc. The methods were add() and
expel(). In our code, we would only need to define this class once. Once the code is
running, we can create instances of the class - ie, we can create objects of the class.

Each of these objects will represent a student, and each will have its own set of values
of data.

The Object Oriented Strategy
Although this chapter has briefly touched on the benefits of Object Orientation (ie
more robust systems, a better abstraction of the real world), we have left many
questions unanswered. How do we identify the objects we need when we're designing
a system? What should the methods and attributes be? How big should a class be? I
could go on! This course will take you through a software development using Object
Orientation (and the UML), and will answer all these questions in full.

One significant weakness of Object Orientation in the past has been that while OO is
strong at working at the class/object level, OO is poor at expressing the behaviour of
an entire system. Looking at classes is all very well, but classes are very low-level
entities and don't really describe what the system as a whole can do. Using classes
alone would be rather like trying to understand how a computer works by examining
the transistors on a motherboard!

The modern approach, strongly supported by the UML is to forget all about objects
and classes at the early stages of a project, and instead concentrate on what the system
must be able to do. Then, as the project progresses, classes are gradually built to
realise the required system functionality. Through this course, we will follow these
steps from the initial analysis, all the way through to class design.



Summary
• Object Orientation is a slightly different way of thinking from the structured
approach

• We combine related data and behaviour into classes

• Our program then creates instances of the class, in the form of an object

• Objects can collaborate with each other, by calling each other’s methods

• The data in an object is encapsulated - only the object itself can modify the data



Chapter 4
An Overview of the UML

Before we begin to look at the theory of the UML, we are going to take a very brief
run through some of the major concepts of the UML.

The first thing to notice about the UML is that there are a lot of different diagrams
(models) to get used to. The reason for this is that it is possible to look at a system
from many different viewpoints. A software development will have many
stakeholders playing a part – for example:
• Analysts
• Designers
• Coders
• Testers
• QA
• The Customer
• Technical Authors
All of these people are interested in different aspects of the system, and each of them
require a different level of detail. For example, a coder needs to understand the design
of the system and be able to convert the design to a low level code. By contrast, a
technical writer is interested in the behaviour of the system as a whole, and needs to
understand how the product functions. The UML attempts to provide a language so
expressive that all stakeholders can benefit from at least one UML diagram.

Here’s a quick look at some of the most important diagrams. Of course, we will look
in detail at each of them as the course progresses:



The Use Case Diagram

Figure 13 - The Use Case Diagram

A Use Case is a description of the system’s behaviour from a user’s viewpoint. This
diagram is a valuable aid during analysis – developing Use Cases helps us to
understand requirements.

The diagram is deliberately simple to understand. This enables both developers
(analysts, designers, coders, tests) and the customer to work with the diagram.

However, do not be tempted to overlook Use Cases as being “too simple to bother
with”. We shall see that Use Cases can drive an entire development process, from
inception through to delivery.



The Class Diagram

Figure 14 - The UML Class Diagram

Drawing Class Diagrams is an essential aspect of any Object Oriented Design method,
so it isn’t surprising that the UML provides us with the appropriate syntax. We’ll see
that we can use the Class Diagram at the analysis stage as well as design – we’ll use
the Class Diagram syntax to draw a plan of the major concepts our customer
understands (and we’ll call this the Conceptual Model). Together with Use Cases, a
Conceptual Model is a powerful technique in requirements analysis



Collaboration Diagrams

Figure 15 - The UML Collaboration Diagram

As we are developing object-oriented software, anything our software needs to do is
going to be achieved by objects collaborating. We can draw a collaboration
diagram to describe how we want the objects we build to collaborate.

Here is a good example of why the UML is “just” a syntax rather a true software
development process. We will see that the UML notation for the diagram is simple
enough, but designing effective collaborations, (that is to say “designing software
which is easy to maintain and robust”), is very difficult indeed. We shall be devoting
an entire chapter for guidelines on good design principles, but much of the skill in
design comes from experience.



Sequence Diagram

Figure 16 - A UML Sequence Diagram

The sequence diagram is, in fact, directly related to the collaboration diagram and
displays the same information, but in a slightly different form. The dotted lines down
the diagram indicate time, so what we can see here is a description of how the objects
in our system interact over time.

Some of the UML modelling tools, such as Rational Rose, can generate the sequence
diagram automatically from the collaboration diagram, and in fact that is exactly how
the diagram above was drawn – directly from the diagram in Figure 15.



State Diagrams

Figure 17 - A State Transition Diagram

Some objects can, at any particular time, be in a certain state. For example, a traffic
light can be in any one of the following states:

Off, Red, Amber, Green

Sometimes, the sequence of transitions between states can be quite complex – in the
above example, we would not want to be able to go from the “Green” state to the
“Red” state (we’d cause accidents!).

Although the traffic light may seem like a trivial example, being sloppy with states
can cause serious and embarrassing faults to occur in our software.

Take as a case in point – a gas bill is sent out to a customer who died four years ago –
it really happens and it is because a programmer somewhere has not taken care with
their state transitions.

As state transitions can be quite complex, the UML provides a syntax to allow us
model them.



Package Diagrams

Figure 18 - The UML Package Diagram

Any non-trivial system needs to be divided up in smaller, easier to understand
chunks, and the UML Package Diagram enables us to model this in a simple and
effective way. We'll be looking in detail at this model when we explore large systems
in the System Architecture chapter.



Component Diagrams

Figure 19 - The UML Component Diagram

The Component Diagram is similar to the package diagram - it allows us to notate
how our system is split up, and what the dependencies between each module is.
However, the Component Diagram emphasises the physical software components
(files, headers, link libraries, executables, packages) rather than the logical
partitioning of the Package Diagram. Again, we'll look at this diagram in more detail
in the System Architecture chapter.



Deployment Diagrams

Figure 20 - A UML Deployment Diagram

The UML provides a model to allow us to plan how our software is going to be
deployed. The diagram above shows a simple PC configuration, for example.

Summary
The UML provides many different models of a system. The following is a list of
them, with a one sentence summary of the purpose of the model:
• Use Cases - “How will our system interact with the outside world?”
• Class Diagram - “What objects do we need? How will they be related?”
• Collaboration Diagram - “How will the objects interact?”
• Sequence Diagram - “How will the objects interact?”
• State Diagram - “What states should our objects be in?”
• Package Diagram - “How are we going to modularise our development?”
• Component Diagram - “How will our software components be related?”
• Deployment Diagram - “How will the software be deployed?”



Chapter 5
The Inception Phase

For the rest of this course, we are going to concentrate on a case study to describe
how the UML is applied on real projects. We shall use the process outlined in Chapter
1, as in the diagram below:

Inception Elaboration Construction Transition

Short Use Cases
Conceptual Model
Prototypes

Full Use Cases Analysis Analysis

Class Diagrams
Interaction Diagrams Design Design
State Diagrams
Coded Use Cases Code Code

Tested Use Cases Test Test

Figure 21 - The process for our case study

In the diagram, I have included the name of each model we will produce at each stage.
For example, at the design stage we will produce Class Diagrams, Interaction
Diagrams and State Diagrams. Of course, we’ll explore these diagrams throughout
this course.

To recap the Inception Phase, the key activities in the phase are:
• Specify the vision for the product
• Produce a business case
• Define the scope of the project
• Estimate the overall cost of the project

The size of the phase depends upon the project. An ecommerce project may well need
to hit the market as quickly as possible, and the only activities in Inception might be
to define the vision and get finance from a bank via the business plan.



By contrast, a defence project could well require requirements analysis, project
definition, previous studies, invites to tender, etc, etc. It all depends on the project.
On this course, we assume the inception phase is already complete. A business study
has been produced (see separate document) that details our customer's initial
requirements and a description of their business model.



Chapter 6
The Elaboration Phase

In the Elaboration Phase, we are concerned with exploring the problem in detail,
understanding the customer’s requirements and their business, and to develop the plan
further.

We must get in to the correct frame of mind to attack this phase correctly. We must
try not to get bogged down with too much detail – especially implementation details.

We need to have a very broad view of the system and understand system-wide issues.
Kruchten (ref [1]) calls this a mile wide and inch deep view.

Prototyping
A key activity in the Elaboration Phase is the mitigation of risks. The sooner risks are
identified and shot down, the lesser their impact will be on the project.

Prototyping difficult or problematic areas of the project are a tremendous help in the
mitigation of risks. Given that we don't want to get bogged down in implementation
and design at this phase, the prototypes should be very focussed, and explore just the
area of concern.

Prototypes can be thrown away at the end of the exercise, or they can be reused during
the construction phase.

Deliverables
Apart from prototypes, we are going to develop two UML models to help us towards
our goal of understanding the problem as a whole.

The first model is the Use Case Model. This will help us to understand what the
system needs to do, and what it should look like to the outside world (ie the users,
or perhaps the systems it must interface to).

The second model is the Conceptual Model. This model allows us to capture, using
UML, a graphical statement of the customer's problem. It will describe all of the
major concepts in the customer's problem, and how they are related. To build this,
we'll use the UML Class Diagram. We will use this Conceptual Model in the
Construction Phase to build our software classes and objects.

We'll cover these two models, in depth, in the next two chapters.



Figure 22 - Two UML models built during Elaboration

Summary
The Elaboration Phase is concerned with developing an understanding of the problem,
without worrying about deep design details (except where risks are identified and
prototypes are required).

Two models will help us with this phase: The Use Case Model and the Conceptual
Model.



Chapter 7
Use Case Modelling

A very powerful UML tool is the Use Case. A Use Case is simply a description of a
set of interactions between a user and the system. By building up a collection of Use
Cases, we can describe the entire system we are planning to create, in a very clear and
concise manner.

Use cases are usually described using verb/noun combinations – for example, “Pay
Bills”, “Update Payroll”, or “Create Account”.

For example, if we were writing a missile control system, typical Use Cases for the
system might be “Fire Missiles”, or “Issue Countermeasures”.

Along with the name of the use case, we will provide a full textual description of the
interactions that will occur between the user and the system. These textual
descriptions will generally become quite complicated, but the UML provides an
astoundingly simple notation to represent a Use Case, as follows:

Figure 23 - Use Case Notation

Actors

A Use Case cannot initiate actions on its own. An actor is someone who can initiate a
Use Case. For example, if we were developing a banking system, and we have a Use
Case called “withdraw money”, then we would identify that we require customers to
be able to withdraw money, and hence a customer would become one of our actors.
Again, the notation for an actor is simple:



Figure 24 - UML Notation for an Actor

Going further, actors can be more than just people. An actor can be anything external
to the system that initiates a Use Case, such as another computer system. An actor
could also possibly be a more abstract concept such as time, or a specific date. For
example, we may have a use case called “Purge Old Orders” in an order processing
system, and the initiating actor could be “Last Working Day”.

As we have noted, actors are related to Use Cases, in the sense that it is an actor that
will initiate a particular use case. We can represent this on a Use Case diagram by
connecting the actor to the use case:

Figure 25 - an Actor's relationship to a Use Case

Clearly, for most systems, a single actor can interact with many use cases, and a
single use case can be initiated by many different actors. This leads to the full use case
diagram, an example of which follows:

Figure 26 - A complete system described using actors and use cases

The Purpose of Use Cases
Given the simple definition of “Use Case” and “Actor”, together with the simple
visualisation of Use Cases through the UML model, we could be forgiven for thinking



that Use Cases are simple – almost too simple to worry about. Wrong. Use Cases are
immensely powerful.
• Use Cases define the scope of the System. They enable us to visualise size and
scope of the entire development.
• Use Cases are very similar to requirements, but whilst requirements tend to be
vague, confusing, ambiguous and poorly written, the tighter structure of Use
Cases tend to make them far more focused
• The “sum” of the use cases is the whole system. That means that anything not
covered by a use case is outside the boundary of the system we are developing.
So the Use Case diagram is complete, with no holes.
• They allow for communication between the customer and developers (since the
diagram is so simple, anyone can understand it)
• Use Cases guide the development teams through the development process – we
shall see that Use Cases are the backbone of our development, and we refer to
them in everything we do
• We’ll see that Use Cases provide a method for planning our development work,
and allow us to estimate how long the development will take
• Use Cases provide the basis for creating system tests
• Finally, Use Cases help with the creation of user guides!
It is often claimed that Use Cases are simply an expression of the system
requirements. Anyone making this claim are clearly missing the point of Use Cases!8

Use Case Granularity
It can be difficult to decide upon the granularity of use cases – in a particular scenario,
should each user-system interaction be a use case, or should the use case encapsulate
all of the interactions? For example, let us consider the example of the ATM machine.
We need to build the ATM system to allow a user to withdraw money. We might have
the following series of common interactions in this scenario:
• enter card
• enter pin number
• select amount required
• confirm amount required
• remove card
• take receipt
Should each of these steps – for example, “enter pin number” be a use case?

8
Though, of course, Use Cases are closely related to requirements. See reference [9] for an excellent
treatment on requirements through Use Cases



Figure 27 - A Useful Use Case Diagram?

This is a classic mistake in the construction of Use Cases. Here, we have generated a
large number of small, almost inconsequential use cases. In any non-trivial system, we
would end up with a huge number of Use Cases, and the complexity would become
overwhelming.

To handle the complexity of even very large systems, we need to keep the Use Cases
at a fairly “high level”. The best way to approach a Use Case is to keep the following
rule-of-thumb in mind:

A Use Case should satisfy a goal for the actor

Applying this simple rule to our example above, we can ask the question “Is take
receipt”, for example, the goal for our customer? Well, not really. It wouldn’t be the
end of the world if the receipt wasn’t dispensed.

Apply the rule to the other Use Cases, and you’ll find that really, none of them
describe the goal of the user. The goal of the user is to withdraw money, and that
should be the use case!

Figure 28 - A more focused Use Case

This approach can feel painful at first, as we are used to performing “functional
decomposition”, where complex tasks are broken down into smaller and smaller tasks.



We will see later that Use Cases can be decomposed, but we must leave that step until
we begin construction.

Use Case Descriptions
Each Use Case contains a full set of textual details about the interactions and
scenarios contained within it.

The UML does not specify what the structure and contents of this document should be
– this is up to individual projects/companies to specify9. We shall use the following
template:

Use Case: Use Case Name

Short Description: A Brief Description of the Use Case

Pre-Conditions: A description of the conditions that must be satisfied before the
use case is invoked

Post-Conditions : A description of what has happened at the end of the use case

Main Flow: A list of the system interactions that take place under the most
common scenario. For example, for “withdraw money”, this
would be “enter card, enter pin, etc...”

Alternate Flow(s): A description of possible alternative interactions.

Exception Flow(s): A description of possible scenarios where unexpected or
unpredicted events have taken place

Figure 29 - Template for a Use Case Description

Use Cases at the Elaboration Phase
Our main job at the elaboration phase is to identify as many of the potential Use Cases
as possible. Bearing in mind the “mile wide and inch deep” principle, our aim is to
provide sketchy details of as many Use Cases as possible – but without the need to
provide the full detail of each Use Case. This will help us to avoid complexity
overload.

At this stage, a Use Case diagram (with actors and Use Cases), plus a brief description
of each Use Case, will suffice. We can revisit the full details of the Use Cases during
the construction phase. Once we have identified the use cases, we can cross reference
Use Cases to requirements and ensure that we have caught all of the requirements.

If we identify some very risky Use Cases at this phase, however, it will be necessary
to explore the details of the risky Use Cases. The production of prototypes at this
stage will help mitigate the risks.

9
An excellent example of the UML providing the syntax, but deliberately not specifying how to use the
syntax



Finding Use Cases
One approach to finding Use Cases is via interviews with the potential users of the
system. This is a difficult task, given that two people are likely to give two completely
different views on what the system should do (even if they work for the same
company)!

Certainly, most developments will involve some degree of direct one-to-one user
communication. However, given the difficulty of gaining a consistent view of what
the system will need to do, another approach is becoming more popular – the
workshop.

Joint Requirements Planning Workshops (JRP)
The workshop approach pulls together a group of people interested in the system
being developed (the stakeholders). Everyone in the group is invited to give their
view of what the system needs to do.

Key to the success of these workshops is the facilitator. They lead the group by
ensuring that the discussion sticks to the point, and that all the stakeholders are
encouraged to put their views across, and that those views are
captured. Good facilitators are priceless!

A scribe will also be present, who will ensure that everything is
documented. The scribe might work from paper, but a better
method is to connect a CASE tool or drawing tool to a
projector and capture the diagrams “live”.

The simplicity of the use case diagram is critical here – all
stakeholders, even non-computer literate stakeholders, should
be able to grasp the concept of the diagram with ease.

A simple method of attacking the workshop is:

1) Brainstorm all of the possible actors first

2) Next, brainstorm all of the possible Use Cases

3) Once brainstorming is complete, as a group, justify each Use Case through by
producing a simple, one line/paragraph description

4) Capture them on the model

Steps 1) and 2) can be reversed if desired.

Some good advice on the workshop:
• don’t work too hard trying to find every single Use Case and Actor! It is natural
that some more use cases will emerge later on in the process.
• If you can’t justify the Use Case at step 3), it might not be a use case. Feel free to
remove any Use Cases you feel are incorrect or redundant (they’ll come back later
if they’re needed!)



The above advice is not a license to be sloppy, but remember the benefit of iterative
processes is that everything doesn’t have to be 100% correct at every step!

Brainstorming Advice
Brainstorming is not as easy as it sounds; in fact I have rarely seen a well-executed
brainstorm. The key things to remember when participating in a brainstorming session
are:
• Document ALL ideas, no matter how outrageous or nonsensical they seem. Stupid
ideas might turn out to be very sensible ideas after all
• Also, silly ideas may trigger off more sensible ideas in other people’s mind – this
is called leapfrogging ideas
• Never evaluate or criticise ideas. This is a very hard rule to observe – we are
trying to break human nature here!

“mmm. No, that won’t work. We won’t bother to document that!”

The facilitator should keep on their toes and ensure that all ideas are captured, and
that all of the group participate.

On the course, a Use Case Workshop will be carried out alongside our client.

Summary
Use Cases are a powerful way of modelling what the system needs to do.

They are an excellent way of expressing the system’s scope (What’s in = Sum of the
Use Cases; what’s out = The Actors).

We need to keep an eye on the granularity of the Use Cases to contain complexity.

The best way of building Use Cases is with the customer, in a workshop.



Chapter 8
Conceptual Modelling

Conceptual Modelling (sometimes called Domain Modelling) is the activity of finding
out which concepts are important to our system. This process helps us to understand
the problem further, and develop a better awareness of our customer’s business.

Once again, the UML does not tell us how or when to do Domain Modelling, but it
does provide us with the syntax to express the model. The model we are going to use
is the class diagram.

Figure 30 - A UML Class Diagram

Developing a Class Diagram is key to any object oriented design process. The Class
Diagram will essentially provide the structure of the eventual code we produce.

At this stage, however, we are not yet interested in the system design (we are still
analysing), so the class diagram we produce at this stage will be quite sketchy, and
will not contain any design decisions.

For example, we would not want to add a “LinkList” class at this stage, as that is tying
us down to a particular solution far too early.

We are essentially producing an Analysis Class Diagram. Many practitioners prefer
to completely distinguish their Analysis Class Diagram from the Design Class
Diagram by calling the Analysis Diagram the Conceptual Model – a term I shall use
for the rest of this course.



On the conceptual model, we aim to capture all of the concepts or ideas that the
customer recognises. For example, some good examples of concepts would be:
• Lift in a lift control system
• Order in a home shopping system
• Footballer in a football transfers system (or a PlayStation football game!)
• Trainer in a stock management system for a shoe shop
• Room in a room booking system
Some very bad examples of concepts are:
• OrderPurgeDaemon the process that regularly deletes old orders from the system
• EventTrigger – the special process that waits for 5 minutes and then tells the
system to wake up and do something
• CustomerDetailsForm – the window that asks for details of the new customer in
a shopping system
• DbArchiveTable – the database table holding a list of all old orders
These are bad concepts, because they are focussing on design – the solution, and not
the problem. In the DbArchiveTable example, we are already tying ourselves down to
a relational database solution. What if it turns out later that it is more efficient,
cheaper, and perfectly acceptable to use a simple text file?

The best rule of thumb here is:

If the customer doesn’t understand the concept, it probably isn’t a concept!

Designers hate the conceptual step – they cannot wait to crack on with design. We
shall see, however, that the conceptual model will slowly transform into a full design
class diagram as we move through the construction phase.

Finding Concepts
I recommend a similar approach to finding Use Cases – a workshop is best – once
again, with as many interested stakeholders as possible.

Brainstorm suggestions for concepts, capture all the suggestions. Once brainstorming
is complete, work as a group to discuss and justify each suggestion. Apply the rule of
thumb that the customer must understand the concept, and discard any that don’t
apply to the problem, and discard any that are touching on design.

Extracting Concepts From Requirements
The requirements document is a good source of concepts. Craig Larman (ref [2])
suggests the following candidate concepts from the requirements:
• Physical or tangible objects
• Places
• Transactions
• Roles of People (eg Customer, Sales Clerk)
• Containers for other Concepts
• Other Systems external to the system (eg Remote Database)



• Abstract Nouns (eg Thirst)
• Organisations
• Event (eg Emergency)
• Rules/Policies
• Records/Logs

A couple of points here. First of all, gathering concepts in a mechanical manner is
poor practise. The above list are good suggestions, but it would be wrong to think that
it is enough to run through the requirements document with a highlighter pen, pulling
out some phrases and setting them as concepts. You must have input from the
customer.

Secondly, many practitioners suggest extracting noun phrases from documents. This
approach has been is common usage for almost 20 years, and although there is
nothing inherently wrong with it, I hate the implication that mechanically searching
for nouns will result in a good list of concepts/classes. Sadly, the English language is
far too ambiguous to allow for such a mechanical approach. I’ll say it again – input
from the customer is essential!

The Conceptual Model in the UML

Now that we’ve seen how to discover the concepts, we need to look how to capture
the concepts in the UML. We’ll use the core aspects of the class diagram.

We represent our concept in a simple box, with the title of the concept (by convention,
capitalised) at the top of the box.

Figure 31 - The Race concept captured in UML (for a Horse Racing System)

Notice that inside the large box are two smaller, empty boxes. The box in the middle
will be used shortly, to capture the attributes of the concept – more on this in a
moment. The bottom box is used to capture the behaviour of the concept – in other
words, what (if anything) the concept can actually do. Deciding on the behaviour of
the concept is a complicated step, and we defer this stage until we are in the design
stage of construction. So we needn’t worry about behaviour for now.



Runner
Name
Age

Figure 32 - The UML captures the attributes and behaviour of a concept

In the example above, we have decided that every runner will have two attributes –
“Name” and “Age”. We leave the bottom area blank until later, when we decide what
a “Runner” is able to do.

Finding Attributes
We need to determine what the attributes of each concept are – and again, a
brainstorming session with the stakeholders is probably the best way to achieve this.

Often, arguments arise over whether or not an attribute should become a concept on
its own. For example, lets say we are working on a Staff Management system. We
have identified that one concept would be “Manager”. A suggestion for an attribute
might be “salary”, as follows:

Manager
salary

Figure 33 - Manager concept, with the attribute Salary

That looks fine, but someone might argue that “Salary” is also a concept. So, should
we make promote it from an attribute to a concept?

I have seen many modelling sessions descend into chaotic arguments over issues like
this one, and my advice is to simply not worry about it: if in doubt, make it another
concept. These kind of problems usually resolve themselves later on anyway, and it
really isn’t worth wasting valuable modelling time on arguments!

Manager Salary

Figure 34 - Manager and Salary, two separate concepts



Guidelines for Finding Attributes
The following rules of thumb may be helpful when deciding between attributes and
concepts – but heed the advice above and don’t worry too much about the distinction.
If in doubt, make it a concept!
• Single valued strings or numbers are usually attributes10
• If a property of a concept cannot do anything, it might be an attribute - eg for the
manager concept, “Name” clearly sounds like an attribute. “Company Car” sounds
like a concept, because we need to store information about each car such as the
registration number and colour.

Associations
The next step is to decide how our concepts are related. In any non-trivial system, at
least some of the concepts are going to have some kind of conceptual relationship
with other concepts. For example, back to our Staff Management system, given the
following two concepts:

Company Car
Manager registration
name engine capacity
colour

Figure 35 - Manager and Company Car Concepts

These concepts are related, because in the company we are developing a system for,
each Manager drives a Company Car.

We can express this relationship in the UML by connecting the two concepts together
with a single line (called an association), as follows:

Company Car
Manager drives registration
name engine capacity
1 1 colour

Figure 36 - Manager and Company Car related by association

Two important things to note about this association. First of all, the association has a
descriptive name – in this case, “drives”. Secondly, there are numbers at each end of
the association. These numbers describe the cardinality of the association, and tell us
how many instances of each concept are allowed.

10
But not always – this is a rule of thumb and shouldn’t be followed slavishly.



In this example, we are saying that “Each Manager Drives 1 Company Car”, and
(going from right to left) “Each Company Car is driven by 1 Manager”.

Manager manages Staff Member
name
1 1..*

Figure 37 - Another Association Example

In the above example, we see that “Each Manager manages 1 or more staff members”;
and (going the other way), “Each Staff Member is managed by 1 Manager”.

Every association should work like this – when reading the association back in
English, the English sentence should make perfect sense (especially to the customer).
When deciding upon association names, avoid weak names like “has” or “is
associated to” – using such weak language could easily hide problems or errors that
would otherwise have been uncovered if the association name was more meaningful.

Possible Cardinalities
Basically, there are no restrictions on the cardinalities you are able to specify. The
following diagram lists some examples, although this list is by no means exhaustive.
The * notation denotes many. Note the slight difference between * and 1..*. The
former is a vague many, meaning that perhaps any number of concepts are allowed,
or maybe we haven't made the decision yet. The latter is more concrete, implying that
one or more are allowed.

Figure 38 - Example Cardinalities

Building the Complete Model
Finally, let’s look at a methodical system for determining the associations between the
concepts. Assume we have completed the brainstorm session and uncovered several
concepts for the Staff Management system. The set of concepts are in the figure below
(I’ve omitted the attributes for clarity).



Figure 39 - Set of Concepts for Staff Management

The best way to proceed is to “fix” one concept, say “Manager” and consider every
other concept in turn. Ask yourself “are these two concepts related?”, and if so,
immediately decide on the name of the association, and the cardinality…

Are Manager and Staff Member related? Yes, Each Manager manages 1 or

“ more staff members.
Manager and Company Car? Yes, Each Manager drives 1 company car.
Manager and Pension? Yes, Each Manager contributes to 1 pension” ”
And so on until the model is complete. A common mistake at this stage is to decide
two concepts are related, draw a line on the diagram and leave off the association
name until later. This is making extra work for yourself – you’ll find that once you’ve
finished adding lines, you’ll have no idea what any of them mean (and they’ll usually
look like spaghetti), and you have to start all over again!



Figure 40 - The simple conceptual model, completed

When building the model, it is important to remember that associations are less
important than attributes. Any missing associations will be easily picked up during
design, but it is harder to spot missing attributes.

Furthermore, it can be tempting to overspecify the map of associations just in case,
and end up with quite a confusing and complex diagram. So a good rule of thumb is to
concentrate on concepts and attributes, and try to fix the most obvious associations.

At the end of the modelling, the diagram should make sense to the customer when you
read back the diagram in English.

Summary
Conceptual Models provide a powerful way of investigating the problem further.

Later on, we’ll expand our model into the design aspects.

This model will eventually be one of the key inputs when we build the code.

To build the model, use a workshop technique as with the Use Cases.



Chapter 9
Ranking Use Cases

We have a lot of work ahead – how can we divide up the work into the simple,
manageable iterations that we described in the early stages of the course?

The answer is by focusing on our Use Cases. In each iteration, we design, code and
test just a few Use Cases. So effectively, we have already decided how we are going
to divide up the work into iterations – the only thing we haven’t done yet is to decide
upon the order in which we will attack them.

To plan the order, we allocate each Use Case a rank. The rank is simply a number that
indicates which iteration the Use Case will be developed in. Any good Case Tool
should allow the rank to be captured as part of the model.

There are no hard and fast rules on how to allocate ranks. Experience and knowledge
of software development plays a large part in setting the rank. Here are some
guidelines on which Use Cases should be allocated a higher rank (ie to be developed
earlier rather than later):

• Risky Use Cases
• Major Architectural Use Cases
• Use Cases exercising large areas of system functionality
• Use Cases requiring extensive research, or new technologies
• “Quick Wins”
• Large payoffs for the customer

Some Use Cases will need to be developed over several iterations. This could be
because the Use Case is simply too big to design, code and test in one iteration, or it
could be because the Use Case depends upon many other Use Cases being complete
(“Start Up” is a classic example of this).

This shouldn’t cause too many problems – simply break the use case down into
several versions. For example, here is a large Use Case, which is to be developed over
three iterations. At the end of each iteration, the Use Case can still perform a useful
task, but to a limited degree.

“Fire Torpedoes” Use Case:

Version 1a allows the user to set a target (Rank : 2)
Version 1b allows the user to prime the weapons (Rank : 3)
Version 1c allows the user to discharge the weapon (Rank : 5)



Summary
Use Cases allow us to schedule the work across our multiple iterations

We rank the Use Cases to determine the order in which they are attacked

Ranking Use Cases is based on your own knowledge and experience.

Some rules of thumb will help in your early days.

Some Use Cases will span several iterations.



Chapter 10
The Construction Phase

In this short chapter, we’ll take stock of what we’ve done, and what needs to be done
next.

In the Elaboration Phase, we needed to understand the problem as fully as possible,
without going into too much detail. We built a Use Case Model, and created as many
Use Cases as possible. We did not fill in the complete details of the Use Cases,
instead we supplied a very brief description of each one.

The next step was to build a conceptual model, where we captured the concepts
driving our development. This conceptual model will provide us with the foundations
of the design.

We then ranked each of our Use Cases, and in doing so, we have planned the order of
the Use Case development.

This completes our Elaboration Phase. A complete review of the phase would be held,
and a Go/No Go decision needs to be made. After it, we may have discovered during
Elaboration that we really cannot provide a solution for our customer – better to find
out now than at the end of coding!

Construction
Now that we are in the Construction Phase, we need to build the product, and take the
system to the state where it can be delivered to the user community.

Recall that our general plan of attack is to follow a series of short waterfalls, with a
small number of Use Cases developed in each iteration. At the end of each iteration,
we will review progress, and preferably timebox the iteration.

Ideally, we will aim to achieve a running (albeit, of course, limited) system at the end
of each iteration.

Each stage of the waterfall will produce a set of documents or UML models.

• In Analysis, we will produce some Expanded (or Full) Use Cases

• In Design, we will produce Class Diagrams, Interaction Models and State
Diagrams

• In Code, we will produce running and unit tested code



Then the iteration is tested (ie all of the use cases need to be demonstrably working),
and then we reach the review.

Summary
We have completed Elaboration, and we are now ready to begin construction. We will
look at each model in turn and see how it benefits the construction exercise.



Chapter 11
The Construction Phase : Analysis

The first stage of the construction phase is Analysis. We need to revisit the Use Cases
we are building in this iteration, and enhance and expand those Use Cases. By
concentrating on the full detail of only a few Use Cases per iteration, we are reducing
the amount of complexity we have to manage at any one time.


Short Use Cases
Conceptual Model
Prototypes


Class Diagrams
State Diagrams


Figure 41 - Analysis Phase of Construction

Remember that even though we are now in construction, we are still at the analysis
stage – albeit a more detailed analysis than the analysis we did in elaboration.
Therefore, we must bear in mind that we are only concerned with the problem, and
not the solution. So we are looking at what the system has to do, without worrying
about how it will do it.

Back to the Use Cases
During Elaboration, we produced Short Use Cases and decided to defer the full details
(ie Main Flow, Alternate Flow, Pre and Post Conditions) until construction phase. The
time has come to complete the full details (but only for the Use Cases we are dealing
with in this iteration).

Use Case Place Bet
Short Description:
The user places a bet on a particular horse after choosing a race
Actors: Gambler



Requirements R2.3; R7.1
Pre-Conditions:
Post-Conditions
Main Flow:
Alternate Flow(s):
Exception Flow(s):
Figure 42 - The Short Use Case, Place Bet

The diagram above shows an example Short Use Case. Each of the headings needs to
be filled in. The best way to explain how to fill in the headings is with a specific
example, so lets have a look at the Place Bet Use Case:

1. Pre-Conditions
This section describes the system conditions which must be satisfied before the Use
Case can actually take place. In the Place Bet, example, a good pre-condition could
be:

“The User Has Successfully Logged In”.

Clearly, the betting system needs to validate customers before they can start
gambling. However, the validation of the user is not part of this Use Case, so we must
ensure that this condition has been satisfied before the betting takes place.

2. Post Conditions
The post conditions describe the state the system will be in at the end of the Use Case.
The post-condition is conventionally written in past tense language. So in our Place
Bet example, the post condition would be:

“The User placed a bet and the bet was recorded by the system”

There can be more than one post condition, depending on the outcome of the Use
Case. These different post conditions are described using “if then” language”. Eg “If a
new customer, then a customer account was created. If an existing customer, then the
customer details were updated”.

3. Main Flow
The main flow section describes the most likely, or the most conventional, flow of
events through the Use Case. Clearly, in the Place Bet Use Case, many things can go
wrong. Perhaps the user cancels the transaction. Maybe the user has insufficient funds
to place a bet. These are all events we have to consider, but really, the most
conventional flow through this use case is going to be a user successfully placing a
bet.

In the main flow, we need to detail the interactions between the actor and the system.
Here is the main flow for “Place Bet”:



(1) On initiation of Place Bet by the gambler, a list of the day’s races are requested
from the system, and (2) the list of races are displayed

(3) The Gambler chooses the race to bet on [A1] and (4) the system presents a list of
the runners for that race

(5) The Gambler chooses the horse to bet on [A1] and enters the required stake [E1]

(6) The User Confirms the transaction and (7) the system displays a confirmation
message

Notice that every actor/system interaction is broken down into steps. In this case,
there are seven steps in the main flow of the Use Case. The [A1] and [E1] notation
will be explained in a moment, when we look at Alternate Flows and Exception
Flows.

Alternate Flows
Alternate flows are simply less common (but legitimate) flows through the Use Case.
The alternate flow will typically share many steps with the main flow, so we can
notate the point in the main flow where the alternate flow takes over. We have done
this in step (3) of the main flow above, through the [A1] notation. This is because
when the user chooses the race to bet on, they can cancel the transaction. They can
also cancel the transaction at step 5, when they are required to enter the stake.

“(A1) The User Cancels the Transaction
Post Condition - No bets were placed”

In this case, the Alternate flow has resulted in a change to the post condition – no bets
were placed.11

Exception Flows
Finally, the exception flow describes exceptional situations. In other words, a flow
where an error has occurred, or an event that couldn’t have otherwise been predicted.

In our place bet example, we could have the following exception:

“(E1) The users credit is not sufficient to fund the bet. The User is informed and the
Use Case terminates”

When we move to program code. the items under Exception Flow should map to
exceptions in the program - if your target language supports exceptions. Many modern
languages do support them - Java, C++, Delphi and Ada to name but four.

11
Some UML practitioners prefer to say that an alternate flow will always result in the same post
conditions as the main flow. Another example where the UML can be applied in many different ways. I
prefer to allow the Alternate flow to be any legitimate but less common flow, resulting in any post
condition you want.



The Complete Use Case

Use Case Place Bet
Short Description: The user places a bet on a particular horse after choosing a race
Actors: Gambler
Requirements R2.3; R7.1
Pre-Conditions: The User has successfully logged in
Post-Conditions: A bet was placed and the bet was recorded by the system
Main Flow:
(1) On initiation of Place Bet by the gambler, a list of the day’s races are requested
from the system, and (2) the list of races are displayed
(3) The Gambler chooses the race to bet on [A1] and (4) the system presents a list of
the runners for that race
(5) The Gambler chooses the horse to bet on [A1] and enters the required stake [E1]
(6) The User Confirms the transaction and (7) the system displays a confirmation
message
Alternate Flow(s):
(A1) The gambler cancels the transaction.
Post Condition - No bets were placed
Exception Flow(s):
(E1) The user’s credit is not sufficient to fund the bet. The user is informed and the
Use Case Terminates

Figure 43 - Full Use Case Description

The UML Sequence Diagram
Producing the Use Case Descriptions is difficult. Many people find the distinction
between analysis and design especially difficult – often the Use Case descriptions
become littered with design decisions.

Here’s an example from the Place Bet Use Case:

“The User selects the race to bet on. The system interrogates the race database and
compiles an array of runners for the race.”

This is a poor Use Case description. By talking about the race database and
introducing arrays, we are tying ourselves down to specific design decisions.

When building the Use Cases, we need to treat the system as a “black box”, which can
accept requests from actors and return results to the actor. We are not concerned (yet)
with how the black box fulfils that request.

We recommend the use of a UML Sequence Diagram. A sequence diagram is useful
in a variety of different situations, especially at the design stage. However, the



diagram can be used in analysis to help us with this “black box” analysis of the
system. Here’s how the diagram works:

We place the actor on the
1 left of the diagram, and on
the right of the diagram, we
represent the entire system
as a single box.

Next, we add vertical
2 “timelines”. The lines
represent the passage of
time in the downward
direction

Interactions between the
3 user and the system are
represented as a line with an
arrow between the system
and the actor.

A description of the
interaction is written
alongside the arrow.



Continue adding the
4 interactions down the
timeline.

The long boxes down the
timeline indicate when the
system of the actor is
“active”. This notation is
more important when we
draw sequence diagrams in
design – for now, it doesn’t
really matter(these boxes
were added by our CASE
tool).

Once the System Sequence Diagram is complete, it is a fairly simple and mechanical
task to write the description of the main flow for the use case. There is no need to
laboriously draw these diagrams for every single alternate and exception flow,
although it would be worthwhile for very complicated or interesting alternatives.

Summary
In this chapter, we moved into the construction phase. We focussed on a handful of
Use Cases in the iteration, and we explored the detail we need to develop for the full
Use Case.

We learnt the basics of a new UML diagram, the System Sequence Diagram, and
saw that this diagram can be helpful when producing the detailed use case.

Now we have the details behind the use cases, the next stage is to produce a detailed
design. We’ve looked at the what – we now look at the how.



Chapter 12
The Construction Phase : Design

Design - Introduction
By now, we have a full grasp of the problem we are trying to solve (for this iteration).
We have developed the Use Cases for the first iteration to a deep level of detail, and
we are now ready to design the solution to the problem.

Use Cases are satisfied by objects interacting. So in this stage, we need to decide on
what objects we need, what the objects are responsible for doing, and when the
objects need to interact.

The UML provides two diagrams to allow us to express the interaction of objects,
namely the Sequence Diagram and the Collaboration Diagram. These two diagrams
are very closely related (some tools can generate one diagram from the other one!)
Collectively, the Sequence and Collaboration Diagrams are called the Interaction
Diagrams.

When we decide on the objects we need, we must document the classes of objects we
have, and how the classes are related. The UML Class Diagram allows us to capture
this information. In fact, much of the work of producing the Class Diagram is already
done – we’ll use the Conceptual Model we produced earlier as a starting point.

Finally, a useful model to build at the design stage is the State Model. More details
on this later.

So, in design, we are going to produce three types of model – the Interaction
Diagram, the Class Diagram and the State Diagram.


Short Use Cases
Conceptual Model
Prototypes


Class Diagrams
State Diagrams


Figure 44 - The deliverables from the design stage



Collaboration of Objects in Real Life
So, our Use Cases are going to be satisfied by the collaboration of different objects.
This is actually what happens in real life. Consider a library. The library is run by a
librarian who runs the front desk. The librarian is responsible for handling queries
from customers, and is responsible for managing the library index. The librarian is
also in charge of several library assistants. The assistants are responsible for managing
the library shelves (the librarian couldn’t do this – or she wouldn’t be able run the
front desk efficiently).

The objects in this library system are:

Customer
Librarian
Library Assistant
Library Index
Bookshelf

Let’s consider the obvious Use Case – Borrow Book

How would this Use Case be satisfied? Let’s assume that
the customer does not know where the book is located and
needs help. This could be the chain of events:

1. The Customer goes to the
librarian and asks for
“Applied UML” by
Ariadne Training.

2. The Librarian looks in the
index for the name of the
book. She finds the index
card for the book, which
says that the book is
located at shelf 4F.

3. The Librarian asks the
assistant to retrieve the
book from shelf 4F.

4. The Librarian gets the requested book and returns it to the librarian.

5. The librarian checks out the book and hands it to the customer.

Figure 45 - Chain of events for Borrow Book

Even though this example is very simple, it was still not particularly easy to consider
the responsibilities of each object. This is one of the key activities of Object Oriented
Design – getting the responsibilities of each object correct. For example, if I had
decided to let the librarian physically retrieve the book, I’d have designed a very



inefficient system. Later on in this chapter, we’ll present some guidelines for
allocating responsibilities to objects.

Collaboration Diagrams
In this section, we will look at the syntax of the UML Collaboration diagram. We will
look at how to use the diagram in the next section.

A collaboration diagram allows us to show the interactions between objects over time.
Here is an example of a completed collaboration diagram:

Figure 46 - Collaboration Diagram

Collaboration Syntax : The Basics
A class on the collaboration diagram is notated as follows:

An instance of a class (in other words, an object) is notated as follows:

Sometimes, we will find it useful to name an instance of a class. In the following
example, I want an object from the Account class, and I want to call it “first”:



If we want one object to communicate with another object, we notate this by
connecting the two objects together, with a line. In the following example, I want a
“bet” object to communicate with an “account” object:

Once we have notated that one object can communicate with another, we can send a
named message from one object to the other. Here, the “bet” object sends a message
to the “account” object, telling it to debit itself:

If we want to pass parameters with the message, then we can include the parameter in
the brackets, as follows. The datatype of the parameter (in this case, a class called
“Money”) can be shown, optionally.

A message can return a value (analogous to a function call at the programming stage).
The following syntax is recommended in the UML standard if you are aiming for a
language neutral design. However, if you have a target language in mind, you can
tailor this syntax to match your preferred language.

return := message (parameter : parameterType) : returnType

In the following example, the bet object needs to know the balance of a particular
account. The message “getBalance” is sent, and the account object returns an integer:



Collaboration Diagrams : Looping
If we need to introduce a loop into a collaboration diagram, we use the following
syntax. In this example, an object from the “Race List” class needs to assemble itself.
To do this, it asks every member of the “Race” class to return its name.

The asterisk denotes that the message is to be repeated. Rather than specifying an
individual object name, we have used the name “All” to denote that we are going to
iterate across all of the objects. Finally, we have used the UML notation for a
collection of objects with the “stacking” of the object boxes.

Collaboration Diagrams : Creating new objects
Sometimes, one object will want to create a new instance of another object. The
method for doing this varies between languages, so the UML standardises creation
through the following syntax:

The syntax is rather strange, really – you are sending a message called “Create” to an
object that doesn’t yet exist!12

Message Numbering
Notice that all of the messages we have included so far have a mysterious “1”
alongside them? This indicates the order in which the message is executed in order to
satisfy the Use Case. As we add more messages (see the full example on page 69), we
increment the message number sequentially.

12
In reality, you are sending a message to the class, and in most languages you are also calling the
constructor.



Collaboration Diagrams : Worked Example
Let’s pull all of that theory together and see how the notation works in practise. Let’s
build the “place bet” Use Case using a collaboration diagram.

This example is far from perfect, and leaves a lot of questions unanswered (we’ll list
the unresolved issues at the end of this chapter). However, as a first cut, the example
should illustrate how collaboration diagrams are built. We’ll revisit these design
issues in later chapters.

See page 61 for the full Use Case Description for “Place Bet”.

To build this diagram, we need some objects. Where do we get the objects from?
Well, we will certainly have to invent some new objects as we go along, but many of
the candidate objects should come directly from our old friend, the conceptual model
we built at the elaboration phase. Here is the conceptual model for the betting system:

Figure 47 - Betting System Conceptual Model

Where there are associations, such as “is placed on”, we will probably use these
associations to pass messages on the collaboration diagram. We could possibly decide
that we need to (for example) pass a message between “account” and “race”. This is
perfectly valid, but as the association was not discovered at the conceptual stage, we
might have possibly broke some of the customer’s requirements. If this happens,
check with the customer!

With the Use Case description and the conceptual model in mind, let’s build the
collaboration for “Place Bet”.

1. First of all, we start with the initiating actor, the customer. The actor symbol is
not strictly part of the UML collaboration diagram, but it is extremely useful
to include it on the diagram anyway.



2. Now, according the Use Case description, when the customer selects the
“place bet” option, a selection of races are presented. So we’ll need an object
that contains a full list of the races for today, so we create an object called
“Race List”13 This is an object which was not represented on the conceptual
model. This is called a design class.

3. So, the actor sends a message to the new “Race List” object called
“getRaceList”. Now, the next job is for the race list to assemble itself. It does
this by looping around all of the Race objects, and asking them what their
name is.

The race object has been taken from the conceptual model.

4. Next, we assume the race list is now returned back to the customer. The ball is
now in their court, and according to the Use Case description (page 61), the
user now selects a race from the list.

13
This will be a container, or array, or something similar, depending on the target language



We may now assume that a race has been selected. We now need to get a list
of runners for the selected race, and to find that out I have decided to make the
race object responsible for keeping a list of its runners. We’ll see in future
chapters why and how this kind of decision is made.

So, message number 3 is sent to the selected race, and the message is asking
for a list of runners for that race.

5. How does the race object know what runners are part of that race? Once again,
we’ll do this using a loop, and we’ll get the race object to assemble a list of
runners.

How this is actually achieved in the real system is not trivial. Clearly, we are
going to store these runners on a database of some kind, so part of the physical
design process is going to be to construct a mechanism for retrieving the horse
records from the database. For now, however, it is enough to say that the
selected Race object is responsible for assembling a list of the runners for that
race.

6. The list of runners is now returned back to the customer. The ball is again in
their court, and according to the Use Case description, they must now select a
runner, and then select their stake money.



Now that we know the runner and the stake, we can send a message to the
selected runner, and tell it that a bet has been placed upon it.

By building this collaboration diagram, we have NOT mapped out exactly what the
code will look like. The following issues have been left unresolved:

1. We have not mentioned anything on the diagram about how the user inputs
data into the system, and how the data (such as the list of runners) is output on
the screen. Somehow all of this happens as if by magic “inside” the actor.

We’ll see later that this is good design. We want to make our design as
flexible as possible, and by including detail about the User Interface at this
stage, we are tying ourselves down to one specific solution.

2. How does the “Race” object find out what runners are part of that race?
Clearly, there is some kind of database (or even network) operation going on
here. Again, we do not want to tie our design down at this stage, so we defer
these details until later.

3. Why have we made the “runner” object responsible for tracking which bets
have been placed on it? Why didn’t we create another class, perhaps called
“bet handler” or “betting system”? This issue will be explored in the following
chapter.

What we have done is decide on the responsibilities of each class. What we are doing,
in effect, is building upon the conceptual model we produced at the elaboration stage.

Some Guidelines For Collaboration Diagrams
As we progress through this course, we’ll expand on how to produce good diagrams.
For now, keep the following guidelines in mind:



1. Keep the diagram simple!!! It seems that in our industry, unless a diagram
covers hundreds of pages of A4 and looks very complicated, the diagram is
trivial! The best rule to apply to the collaboration diagram (and the other UML
diagrams too) is to keep them as simple as possible. If the collaboration for a
use case gets complicated, break it up. Perhaps produce a separate diagram for
each user/system interaction.

2. Don’t try to capture every scenario. Every use case comprises a number of
different scenarios (the main flow, several alternatives and several exceptions).
Usually, the alternatives are fairly trivial and not really worth the bother of
including.

A common mistake is to cram every scenario on one diagram, making the
diagram complex and difficult to code.

3. Avoid creating classes whose name contains “controller”, “handler”,
manager” or “driver”. Or at least, be suspicious if you do come up with an
object with such a name. Why? Well, these classes tend to suggest that your
design is not object oriented. For example, in the “Place Bet” use case, I could
have created a class called “BetHandler” that deals with all of the betting
functionality. But this would be an action oriented rather than an object
oriented solution. We already have the “Bet” object from the collaboration
diagram, so why not use it, and give it the responsibility of handling bets?

4. Avoid God classes. Similarly, if you end up with a single object that does a
lot of work and does not collaborate much with other objects, you have
probably built an action oriented solution. Good OO solutions consist of small
objects who don’t do too much work themselves, but work with other objects
to achieve their goal. We’ll go into much more detail about this later.

Chapter Summary
In this chapter, we began to construct a software solution to our Use Cases. The
collaboration diagram enabled us to allocate responsibilities to the classes we derived
during elaboration.

We touched on a few issues we should be aware of when allocating responsibilities,
even though we need to learn more about this topic later. An example for “Place Bet”
was studied.

In the next section, we’ll see how we can expand the Conceptual Model and progress
it towards a true Class Diagram.



Chapter 13
Design Class Diagrams

Recall that at the elaboration stage, we built a conceptual model. The conceptual
model contained details about the customer’s problem, and concentrated on the
customer’s concepts, and the properties of those concepts. We did not allocate
behaviour to any of those concepts.

Now that we have begun to create collaboration diagrams, we can progress the
conceptual model, and build it into a true Design Class Diagram. In other words, a
diagram which we can base our final program code upon.

Producing the design class diagram is a fairly mechanical process. In this chapter,
we’ll examine an example Use Case, and how the conceptual model is modified
because of it.

Crediting and Debiting Accounts
At the end of the “place bet” Use Case, the “bet” object sends a message to the
customer’s “Account” object, to tell it that it must be decremented. The following
conceptual model was the basis of this design:

Figure 48 - Conceptual Model for this example

From the conceptual model, the following (portion of a) collaboration was developed:

Figure 49 - Part of the Collaboration for Place Bet



Step 1 : Add Operations
From the collaboration diagram, we can see that the “Account” class needs to provide
the behaviour “debit”. So we add this operation to the lower half of the class symbol.

Figure 50 - Classes with Operation Added

Note that most people don’t bother to add the create operations, as this will clutter the
diagram (most classes need one anyway).

Step 2 : Add Navigability
The direction of the messages which are being passed through an association are also
added. In this case, the message is being sent from the bet class to the account class,
so we orient the association from the caller to the receiver:

Figure 51 - Account Class with Navigability Added

Sometimes, a situation will arise where messages need to be passed both ways across
an association. What to do in this case? The UML notation for this situation is to
simply leave the arrow head off the association – a bi-direction association.

Many modellers believe that bi-directional associations are erroneous and need to be
removed from the model somehow. In actual fact, there is nothing fundamentally
wrong with a bi-directional relationship, but it does suggest a bad design. We’ll
explore this problem in a later chapter.

Step 3 : Enhance Attributes
We can also decide upon the datatype of the attributes at this stage. Here, we have
decided to store the balance of an account as a float. Of course, this is dependent upon
the choice of language.

Figure 52 - Datatypes added



Step 4 : Determine Visibility
A fundamental concept of Object Orientation is encapsulation – the idea that the data
held by an object is kept private from the outside world (ie from other objects).

We can signal which attributes and operations are public or private on a UML Class
diagram by preceding the attribute/operation name by a plus sign (for public) and a
minus sign (for private).

All attributes will be private, unless there is an extremely good reason (and there
rarely is). Usually, operations will be public, unless they are helper functions, only to
be used by operations contained within the class.

Figure 53 - The Class Diagram, complete!

Now that the class diagram is complete, we now have enough information to produce
the code. We’ll examine the transition to code in a later chapter.

Aggregation
An important aspect of object oriented design is the concept of Aggregation – the
idea that one object can be built from (aggregated from) other objects.

For example, in a typical computer system, the computer is an aggregation of a CPU,
a graphics card, a soundcard and so on.

We can denote aggregation in UML using the aggregation symbol – a diamond at the
end of an association link.



Figure 54 - Computer built from other objects

If you spot aggregation on your conceptual model, it may be clearer to explicitly
notate the fact, using the aggregation symbol.

Composition
A very similar concept to aggregation is composition. Composition is stronger than
aggregation, in the sense that the relationship implies that the whole cannot exist
without the parts.

In the aggregation example above, for example, if we removed the soundcard, the
Computer would still be a computer. However, a book isn’t a book without its pages,
so we say that a book is composed of the pages.

The notation for this is similar to aggregation, except this time the diamond is filled,
as follows:

Figure 55 - A book is composed of 1 or more pages

Finding Aggregation and Composition
Finding these relationships on your class diagram is useful, but it is far from crucial to
the success of your design. Some UML practitioners go further, and claim that these
relationships are redundant, and should be removed (aggregation and composition can
be modelled as an association with a name like “is composed from”).

I believe, however, that as aggregation is one of the key concepts of Object
Orientation, it is definitely worth explicitly notating its presence.

Summary
In this chapter, we looked at how to progress the class model, based on our work on
collaborations. The transition from conceptual to design class model is actually fairly
trivial and mechanical, and shouldn’t cause too many headaches.



Chapter 14
Responsibility Assignment Patterns

In this section, we are going to take time out from the development process, and look
closely at the skills involved in building good Object Oriented Designs.

Some of the advice given in this chapter may appear to be quite obvious and trivial. In
fact, it is the violation of these simple guidelines that causes most of the problems in
object oriented design.

The GRASP Patterns
To improve the way in which we produce our collaboration diagrams, we’ll study the
so-called “GRASP” patterns, as described by Larman in reference [2].

What is a pattern?
A pattern is a well used, extremely general, solution to a common occurring problem.
The pattern movement began as an internet-based discussion community, but was
popularised through the classic textbook “Design Patterns” (reference 6), written by
the so called “Gang of Four”.

To aid communication, each design pattern has an easy to remember name (such as
Factory, Flywheel, Observer), and there are at least a handful of design patterns that
every self respecting designer should be familiar with.

We’ll be looking at some of the “Gang of Four’s” patterns later, but first we’ll study
the GRASP patterns.

GRASP stands for “General Responsibility Assignment Software Patterns”, and they
help us to ensure we allocate behaviour to classes in the most elegant way possible.

The patterns are called Expert, Creator, High Cohesion, Low Coupling and
Controller. Let’s look at them in turn:

Grasp 1 : Expert
This is, on the face of it, a very simple pattern. It is also the one which is most
commonly broken! So, this pattern should be at the forefront of your mind whenever
you are building collaboration diagrams or creating design class diagrams.

Essentially, the Expert pattern says “given a behaviour, which class should the
behaviour be allocated to?”.

Wise allocation of behaviour leads to systems which are:



• Easier to understand
• More easily extendible
• Reusable
• More Robust
Let’s look at a simple example. We have three classes, one representing a Purchase
Order, one for an Order Line, and finally, one for a SKU (see the Course Case Study
if these terms are not clear to you).

Here is a fragment from the conceptual model:

Figure 56 - Fragment from the conceptual model

Now, imagine that we are building collaboration for one of the use cases. This use
case demands that the total value of a selected purchase order is presented to the user.
Which class should be allocated the behaviour called “calculate_total()”?

The expert pattern tells us that the only class who should be allowed to deal with the
total cost of purchase orders is the purchase order class itself – because that class
should be an expert about all things to do with purchase orders.

So, we allocate the “calcuate_total()” method to the Purchase Order class.

Now, to calculate the total of a purchase order, the purchase order needs to find out
the value of all of the order lines.

A poor design for this would be to let the purchase order see the contents of every
order line (via accessor functions), and then sum up the total. This is breaking the
expert pattern, because the only class who should be allowed to calculate the total of
an order line is the order line class itself.

So we allocate a further behaviour to the order line class, called subtotal(). This
method returns the total cost of the single order line. To achieve this behaviour, the
order line class needs to find out the cost of a single SKU through another method
(this time, an accessor) called price() in the SKU class.



Figure 57 - Behaviours allocated by observing the expert pattern

This leads to the following collaboration diagram:

Figure 58 - three classes of objects collaborating to provide the total cost of a
purchase order

Grasp 2 : Creator
The Creator pattern is a specific application of the Expert pattern. It asks the question
“who should be responsible for creating instances of a particular class?”

The answer is that Class A should be responsible for creating objects from Class B if:
• A Contains B Objects
• A closely uses B Objects
• A has the initialising data that will be passed to B Objects



For example, lets return to the purchase order example. Let’s say that a new purchase
order has been created. Which class should be responsible for creating the
corresponding purchase order lines?

Figure 59 - Which class should create purchase orders?

The solution is that, as a purchase contains purchase order lines, then the purchase
order class (and only that class) should be responsible for creating order lines.

Here is the collaboration diagram for this situation:

Figure 60 - The Purchase Order creating Order Lines

Grasp 3 : High Cohesion
It is extremely important to ensure that the responsibilities of each class are focussed.
In a good object oriented design, each class should not do too much work. A sign of a
good OO design is one where every class has only a small number of methods.

Consider the following example. In the design for a Lift Management system, the
following class has been designed:



Figure 61 - A class from the Lift Management System

Is this a good design? Well, the class obviously does a lot of work – raising alarms,
starting up/shutting down, moving the lift and updating the display indicator. This is a
bad design because the class is not cohesive.

This class would be difficult to maintain, as it isn’t obviously clear what the class is
supposed to be doing.

The rule of thumb to follow when building classes is that each class should capture
only one key abstraction – in other words, the class should represent one “thing”
from the real world.

Our Lift controller is trying to model at least three separate key abstractions – an
Alarm, the lift Doors and the Fault Log. So a better design is to break the Lift
Controller into separate classes.

Figure 62 - The Lift Controller class modelled as four separate, more cohesive,
classes



Grasp 4 : Low Coupling
Coupling is a measure of how dependent one class is on other classes. High Coupling
leads to code that is difficult to change or maintain – a single to change to just one
class could lead to changes “rippling” throughout the system.

The Collaboration Diagram provides an excellent means for spotting coupling, and as
a result, high coupling can be also be spotted through the Class Diagram.

The following is an example extract from a Class Diagram that is showing clear signs
of high coupling:

Figure 63 - High Coupling in a Class Diagram

Are all of the associations in Figure 63 really necessary? The designer of this system
should ask some serious questions about the design. For example:

• Why is the Fault class associated to the Outlets class directly, when an indirect
association exists through the Tank Controller Class? This link may have been put
in place for performance reasons, which is fine, but more likely the link has
appeared due to sloppiness in the Collaboration Modelling

• Why does Tank Controller have so many associations? This class is probably
incohesive and doing too much work.



Following the conceptual model is an excellent way of reducing coupling. Only send
a message from one class to another class if an association was identified at the
conceptual modelling stage. This way, you are restricting yourself to introducing
coupling only if the customer agreed that the concepts are coupled in real life.

If, at the Collaboration stage, you realise that you wish to send a message from one
class to another and they are NOT associated on the conceptual model, then ask a very
serious question about whether or not the coupling exists in the real world. Talking to
the customer may help here - perhaps the association was overlooked when you built
the conceptual model.

So the rule of thumb here is : Keep coupling to a minimum - the conceptual model
is an excellent source of advice on what that minimum should be. It is fine to
raise the level of coupling, as long as you have thought very carefully about the
consequences!

Worked Example

Let us look at a Purhcase Ordering System. In the conceptual model, it was identified
that Customers own Purchase Orders (because they raise them):

For the Create Purchase Order Use Case, which class should be responsible for
creating new purchase orders? The Creator Pattern suggests that the Customer Class
should be responsible:

Figure 64 - Class Diagram and Collaboration for Create Purchase Order

So, we have now coupled Customer and Purchase Order together. That is fine,
because they are coupled together in real life too.

Next, once the Purchase Order has been created, the Use Case needs to add lines to
the Order. Who should be responsible for adding lines?

One approach is to let the Customer class do the work (after all, it does have the
required initialising data - how many lines, what products, what quantities, etc).



Figure 65 - First attempt. The customer objects creates the lines because it holds
the initialising data

This approach has raised coupling artificially however. We now have all three class
dependent upon each other, whereas if we had made the Purchase Order responsible
for creating the lines, we would have the situation where Customers have no
knowledge of Order Lines:

Figure 66 - Adding Lines, with reduced coupling

So now, if the implementation of the Order Line class changed for any reason, the
only class affected would be the Purchase Order Class. All the coupling that exists on
this design was coupling that was identified at the conceptual stage. Customers own
Purchase Orders; Purchase Orders own Lines. So it makes sense that this coupling
should exist!



The Law of Demeter

This Law, also known as Don't Talk to Strangers is an effective method of combating
coupling. The Law states that any method of an object should only call methods
belonging to:
• Itself
• Any parameters that were passed in to the method
• Any objects it created
• Any directly held component objects
Make your objects shy and coupling will be reduced!

Final Words on Coupling

Some more issues to consider:
• Never make an attribute of a class public - a public attribute instantly opens up the
class to abuse (the exception is constants held by the class)
• Only provide get/set methods when strictly necessary
• Provide a minimum public interface (ie only make a method public if it has be
accessed by the outside world)
• Don’t let data flow around the system - ie minimise data passed as parameters
• Don’t consider coupling in isolation - remember High Cohesion and Expert! A
completely uncoupled system will have bloated classes doing too much work.
This often manifests itself as a system with a few “active objects” that don’t
communicate.

Grasp 5 : Controller
The final GRASP pattern we will look at in this section is the controller pattern. Let’s
return to the bookmaking system, and return to the Place Bet Use Case. When we
built the collaborations for this Use Case (see page 72), we realised that we hadn’t
really considered how the input would be entered by the user, and how the results are
displayed to the user.

So, for example, we need to display the details of a race to the user. Which class
should be responsible for satisfying this requirement?

Application of the Expert pattern suggests that as the details pertain to a Race, then
the Race class should be the expert in displaying the relevant details.

Figure 67 - Should Race be responsible for displaying its details?



This sounds ok at first, but actually, we have violated the expert pattern! The Race
class should indeed be an expert at everything to do with races, but it must not be an
expert on other matters – like Graphical User Interfaces!

In general, adding information about GUI’s (and databases, or any other physical
object) into our classes is poor design – imagine if we have five hundred classes in
our system, and many of them read and write to the screen, perhaps to a text based
console. What would happen if the requirement changed, and we wanted to replace
the text-based screen for a windows-based GUI? We would have to trawl through all
of our classes, and laboriously work out what needs to be altered.

It is much better to keep all of the classes from the conceptual model (I’ll call them
“Business Classes”) pure, and remove all reference to GUI’s, Databases and the like.
But how our Race class can display the race details?

Solution – Controller Pattern

One possible solution is the use of a Controller Pattern. We can introduce a new class,
and make it sit between the actor and the business classes.

The name of this controller class is usually called UseCaseNameHandler. So in our
case, we need a handler called “PlaceBetHandler”.

The handler reads the commands from the user, and then decides which classes the
messages should be directed to. The handler is the only class that will be allowed to
read and write to the screen.

Figure 68 - Use Case Controller added to the design

In the event of us needing to replace the user interface, the only classes we have to
modify are the controller classes.

Summary
In this chapter, we explored the “GRASP” Patterns. Careful application of the five
GRASP patterns leads to more understandable, modifiable and robust Object Oriented
Designs.



Chapter 15
Inheritance

One of the most well-known and most powerful concepts behind Object Orientation is
Inheritance. In this chapter, we’ll have a basic look at inheritance, when to apply it
(and when not to apply it), and in particular we’ll look at the UML notation for
expressing inheritance.

Inheritance – the basics
Often, several classes in a design will share similar characteristics. In Object
Orientation, we can factor out these common characteristics into a single class. We
can then “inherit” from this single class, and build new classes from it. When we have
inherited from a class, we are free to add new methods and attributes as the need
arises.

Here’s an example. Let’s say we are modelling the attributes and behaviour of Dogs
and People. This is what our classes might look like:

Figure 69 - Modelling Dogs and People

Although the two classes are different, the two classes also share a lot in common.
Every Person has a name; so too does every Dog14. Similarly with Age. Some of the
behaviours are common, such as Eat, Sleep and Die. Talk, however, is unique to the
Person class.

If we decided to add a new class to our design, such as “Parrot”, it would be tedious to
add all of the common attributes and methods to the Parrot class again. Instead, we
can factor out all of the common behaviour and properties into a new class.

14
Let’s assume we are modelling pet dogs!



If we take out the attribute “Age”, and the behaviours “Eat”, “Sleep” and “Die”, we
have attributes and behaviour which should be common amongst all animals.
Therefore, we can build a new class called “Animal”.

Figure 70 - A more general Animal class

So we have now built a more general class than our specific Dog, Person and Parrot
class. This process is known as generalisation.

Now, when we need to create the Dog class, instead of starting from scratch, we
inherit from the Animal class and merely add on the attributes and methods that we
need for our specific class.

The following diagram illustrates this in the UML. Note that in the new Dog class we
don’t include the old methods and attributes – they are implicit.

The class we started from is called the base class (sometimes called the Superclass).
The class we have created from it is called the derived class (sometime the Subclass).

Figure 71 - Creating a Dog class from the Animal Class

We can continue creating new derived classes from the same base class. To create our
Person class, we inherit again, as follows:



Figure 72 - Person derived from Animal

We can continue inheriting from classes to form a class hierarchy.

Figure 73 - A Class Hierarchy

Inheritance is White Box Reuse
A common mistake in object oriented designs is to over use inheritance. This leads to
maintenance problems. Effectively, a derived class is tightly coupled to the base class
– changes to the base class will result in changes to the derived class.

Also, when we use a derived class, we need to find out exactly what the base classes
can do. This might mean trawling through a large hierarchy structure.



This problem is known as the proliferation of classes.

One common cause of proliferation is when inheritance is used when it shouldn’t be.
Follow the following rule-of-thumb:

Inheritance should only be used as a generalisation mechanism.

In other words, only use inheritance when the derived classes are a specialised type of
the base class. There are two rules to help here:
• The is-a-kind-of rule
• The 100% rule

The 100% Rule
All of the base class definition should apply to all of the derived classes. If this rule
doesn’t apply, then when you inherit, you are not creating specialised versions of the
base class. Here’s an example:

Figure 74 - Poor inheritance

In Figure 74, the fly() method should not be part of the Animal class. Not all animals
can fly, so the derived class, Person, has an extraneous method associated with it.

Ignoring the 100% rule is an easy way to create maintenance problems.

Substitutability
In the previous example, why could we not simply remove the “fly” operation in the
person class? That would solve the problem.

Methods cannot be removed in an inheritance hierarchy. This rule is enforced to
ensure that the Substitutability Principle is upheld. We'll look at this in a little more
detail shortly.



The Is-A-Kind-Of Rule
The Is-A-Kind-of rule is a simple way to test if your inheritance hierarchy is valid.
The phrase “derived class is a base class” should make sense. For example “a
dog is a kind of animal” makes sense.

Often, classes are derived from base classes when this rule does not apply, and again,
maintenance problems are likely. Here’s a worked example:

Example - Reusing queues through inheritance
Assume that we have built (in code) a working Queue class. The Queue class allows
us to add items to the back of the queue and remove items from the front of the queue.

Figure 75 - The Queue Class

After a while, we decide that we need to build a new type of queue – a special kind of
queue called a “Dequeue”. This kind of queue allows the same operations as a queue
but with the additional behaviour of allowing items to be added to the front of the
queue and for items to be removed from the back. A kind of “two-way” queue.

To reuse the work we have already completed with the Queue, we can inherit from the
Queue class.

Figure 76 - Building a Dequeue through inheritance

Does this pass the Is-A-Kind-of and 100% tests?



• 100% : Do all of the methods in Queue apply to Dequeue? The answer is yes,
because all of these methods are required.
• Is-A-Kind-Of : Does this sentence make sense? “A Dequeue is-a-kind-of Queue”.
Yes, it does, because a dequeue is a special kind of queue.
So this inheritance was valid.

Now, let’s go further. Let’s say that we need to create a Stack. For a stack, we need to
support the methods add_to_front() and remove_from_front().

Rather than writing the stack from scratch, we could simply inherit from the dequeue,
as the dequeue provides both of these methods.

Figure 77 - Creating a Stack...no work needed!

We can feel very smug that we have re-used the Dequeue and created a stack with no
further work required. Any code using the stack can add items to the front, and
remove them from the front, and therefore we have a working stack.

We shouldn’t feel too smug though. This is a very poor design indeed. Don’t forget
that the stack has also inherited the methods add_to_back() and remove_from_back()
– these are two methods that are meaningless in a stack! So the stack fails the 100%
test. In addition, the stack fails the Is-A-Kind-Of test, because a stack isn’t really a
kind of Dequeue at all!

So we’ve created a maintenance problem – namely that any code using the stack can
erroneously add items to the back of the stack! How do we get around this?? There
really isn’t any way, other than to bodge the stack class. Remember that we cannot
remove methods from a class when we inherit.15

15
The substitutability principle



The solution is to use aggregation rather than inheritance. We create a new class
called stack, and include a Dequeue as one of its private attributes.

We can now provide the two public methods, add_to_front() and
remove_from_front() as part of the stack class. The implementation of these methods
are simple calls to the same methods contained in the Dequeue.

Figure 78 - The Stack class reusing the Dequeue efficiently

Now, users of the Stack class can only call the two public methods – the methods
contained within the “hidden” Dequeue are private. This ensures that the Stack class
has a highly cohesive interface, and will be much easier to maintain and understand.

Problems With Inheritance
Although Inheritance looks like a powerful mechanism to achieve reuse, Inheritance
should be approached with care. Overusing inheritance can lead to very complex and
difficult to understand hierarchies. This problem is known as the proliferation of
classes. The problem is made even more acute when inheritance is used incorrectly
(as described above). So ensure that inheritance is used sparingly, and make sure the
100% and is-a-kind-of rules apply.

In addition, Inheritance is White Box Reuse. Encapsulation between the base class
and the derived class is quite weak - generally, a change to the base class could impact
any derived classes, and certainly, any user of a class also needs to know about how
the chain of parent classes above the class works.

The user of a class that is buried at the foot of an 13-class deep inheritance chain is
going to have a real headache when working with that class - how many classes can
you juggle in your head at once??



Visibility of Attributes

Figure 79 - Simple Inheritance

Consider the inheritance tree in the figure above. It is important to realise that the
private members of the base class, Animal are not visible to the derived class, Person.
So the methods talk(), work() and play() cannot access the age attribute.

This makes sense in a way, because you can argue that the methods that are only
relevant to the Person class should not be able to fiddle with the attributes of the
Animal class.

However, this restriction is sometimes too tight, and you need to allow a derived class
to be able to see the attributes in the base class. Of course, we could make the
attributes public, but that would break encapsulation and open up the attributes to the
entire world. So OO provides a middle ground, called protected visibility.

A protected member is still private to the outside world, but will remain visible to any
derived classes. Most OO languages support protected visibility.

The UML notation for a protected member is shown below:



Figure 80 - The Age attribute is now protected, and is therefore visible to the
Person class. It is still private as far as other classes are concerned.

Polymorphism
Derived classes can redefine the implementation of a method. For example, consider a
class called “Transport”. One method contained in transport must be move(), because
all transport must be able to move:

Figure 81 - The Transport Class

If we wanted to create a Boat and a Car class, we would certainly want to inherit from
the Transport class, as all Boats can move, and all Cars can move:



Figure 82 - Boat and Car derived from Transport. The Is-A-Kind-Of and 100%
rules are satisfied.

However, cars and boats move in different ways. So we will probably want to
implement the two methods in different ways. This is perfectly valid in Object
Orientation, and is called Polymorphism.

Abstract Classes
Often in a design, we need to leave a method unimplemented, and defer its
implementation further “down” the inheritance tree.

For example, back to the situation above. We added a method called “move()” to
Transport. This is good design, because all transport needs to move. However, we
cannot really implement this method, because Transport is describing a wide range of
classes, each with different ways of moving.

What we can do is make the Transport class abstract. This means that some, or
maybe all, of the methods are unimplemented.

When we derive the car class from Transport, we can then go ahead and implement
the method, and similarly in boat.

The UML Syntax for an abstract class and an abstract method is to use italics, as
follows:

Figure 83 - We don't implement the move() method in Transport

This is one area of the UML that I don't think has been too well thought out. Italics
are often difficult to spot on a diagram (and difficult to produce if you are writing the
diagram on paper). The solution is to use a UML Stereotype on the abstract class as
follows:



Figure 84 - Clarifying an Abstract class using a stereotype

The Power of Polymorphism
Polymorphism comes into its own when we apply the principle of substitutability.
This principle says that any method we write that expects to work on a particular
class, can happily work on any derived class too.

An example in code illustrates this point - I have written this snippet in Java-
Pseudocode-ish(!), but the principle should be clear:

public void accelerate (Transport theTransport, int
acceleration)
{
-- some code here
theTransport.move();
-- some more code
}

--
--
--
accelerate (myVauxhall);
accelerate (myHullFerry);
Figure 85 - Java code illustrating substituability

In this example, I have written a method called accelerate. It works using a parameter
of type Transport, and presumably speeds up the transport using the move() method.

Now, I can safely call this method, and pass it a Car object (because it is a subclass of
Transport), and I can safely pass it a Boat object too. The function I have written
simply doesn't care what the actual type of the object is, as long as it is derived from
Transport.

This is extremely flexible. Not only have I written a general purpose method that can
work on a whole range of different classes, I have also written a method that could in



future be used on a class that isn't even designed yet. Later, someone might create a
new class called Aeroplane, derived from Transport, and the accelerate() method
would still work, and happily accept the new Aeroplane class, without modification
or recompilation!

This is the reason why we cannot remove a method when we derive a new class. If we
were allowed to do so, the aeroplane class could conceivably remove the move()
method, and all the benefits listed above would be destroyed!

Summary
The Notation for Inheritance in UML is simple.

Classes can be arranged into an “inheritance hierarchy”

A sub-class must inherit all of the parent class’ public behaviour

Protected methods and attributes are also inherited

Polymorphism is an incredibly powerful tool to achieve code reuse



Chapter 16
System Architecture - Large and Complex
Systems

So far in this book, we have considered relatively small systems. Generally,
everything we have said so far would be easy to apply on a project with, say, 3 or 4
developers, with a handful of iterations lasting a couple of months each.

In this chapter, we'll have a look at some of the issues surrounding larger, more
complex developments. Is the UML within an Iterative, Incremental Framework
scaleable? And what else can the UML offer to help contain the complexity of such
developments?

The UML Package Diagram
All UML Artefacts can be arranged into UML Packages. A package is basically a
logical container into which related elements can be placed - exactly like a folder or
directory in an operating system.

Figure 86 - Notation for a UML Package

In the above example, I have created a package called GUI. I will probably be
placing UML artefacts relating to Graphical User Interfaces inside the package.

We can display groups of packages, and the relationships between them, on the UML
package diagram. The following is a simple example:



Figure 87 - Three UML packages,

In the example above, I have notated the classic three tier model of software
development. Items inside the Presentation package are dependent upon the items
inside the Business package.

Note that the diagram does not show what is actually inside the package. Therefore,
the Package Diagram provides a very high level view of the system. However,
many case tools allow the user to double-click on the package icon to open up the
package and explore the contents.

A package can contain other packages, and therefore packages can be arranged into
hierarchies, again, exactly as with directory structures in operating systems.

Elements Inside a Package
Any UML artefact can be placed inside a package. However, the most common use of
a package is to group related classes together. Sometimes, the model is used to group
related Use Cases together.

Within a UML package, the names of the elements must be unique. So, for example,
the name of every class within the package must be unique. However, one major
benefit of packages is that it doesn't matter if there is a name class between two
elements from different packages. This provides the immediate advantage that if we
have two teams working in parallel, Team A does not need to worry about the
contents of Team B's package (as far as naming goes). Nameclashes will not occur!

Why Packaging?
So why do we bother with packaging? Well, by careful use of packages, we can:
• Group large systems into easier to manage subsystems
• Allow parallel iterative development

Also, if we design each package well and provide clear interfaces between the
packages (more on this shortly), we stand a chance of achieving code reuse. Reusing
classes has turned out to be somewhat difficult (in some sense, a class is quite small
and a bit fiddly to reuse), whereas a well designed package can be turned into a solid,
reusable software component. For example, a graphics package could be used in many
different projects.



Some Packaging Heuristics
Let us assume for this section that we are using the package diagram to partition
classes into easy to understand and maintain packages.

Figure 88 - A Well Designed Package Structure?

Several of the Heuristics from the GRASP chapter apply equally well to packaging.
Three in particular stand out:

Expert
Which package should a class belong to? It should be obvious where each class
belongs - if it isn't obvious then the package diagram is probably lacking.

High Cohesion
A package shouldn't do too much (or it will be difficult to understand, and certainly
difficult to reuse).

Loose Coupling
Dependencies between packages should be kept to an absolute minimum. The
diagram above is a made up example, but it looks fairly horrendous! Why is there so
much cross package communication?

Handling Cross Package Communication
Assume we have two packages, each containing a number of classes.



Figure 89 - Two Subsystems, modelled as UML Packages

From the dependency arrow, we can see that classes in the Subsystem 1 package
make calls to classes in the Subsystem 2 package.

If we were to drill down and look inside the two packages, we might see something
like the following (the attributes and operations have been removed for clarity):

Figure 90 - Classes across two subsystems. The dashed line represents the
subsystem boundary

Basically, we have a situation where any class from the Subsystem 1 package can
call any class from the Subsystem 2 package. Is this a good design?

Clearly, this idea leaves something to be desired. What if we needed to remove the
Subsystem 1 package and replace it with a new subsystem (let's say that we are



removing a terminal based user interface and replacing it with an all singing, all
dancing graphical interface).

There would be a lot of work involved to understand the impact of the change. We
would have to ensure that every class in the old subsystem is replaced with a
corresponding class in the new subsystem. Very messy, and very inelegant. Luckily,
there exists a design pattern called a Facade to help us with this problem.

The Facade Pattern
A better solution is to employ an extra class to act as a go between between the two
subsystems. This type of class is called a Facade, and will provide, via its public
interface, a collection of all of the public methods that the subsystem can support.

Figure 91 - The Facade Solution

Now, calls are not made across the subsystem boundary, but all calls are directed
through the Facade. If one subsystem were to be replaced, then the only change
required would be to update the Facade.

The Java language has excellent support for this concept. As well as the usual Private,
Public and Protected class visibilities, Java provides a fourth level of protection called
Package Protection. If a class is designated as package rather than public, only
classes from the same package may access it. This is a very strong level of
encapsulation - by making all classes except the Facades package, each team building
the subsystem really can work independently of each other.



Architecture-Centric Development
The Rational Unified Process strongly emphasises the concept of Architecture-centric
development. Essentially, this means that the system is planned as a collection of
Subsystems from a very early stage in the project development.

By creating a group of small, easy to manage subsystems, small development teams
(maybe of just 3 or 4 people) can be allocated to each subsystem, and, as much as
possible, can work in parallel, independent of each other.

Clearly, this is far easier said than done. To underline the importance of this
architecture activity, a full time architecture team is appointed (this could be a single
person). This team is charged with the management of the architectural model - they
would own and maintain everything related to the high level big picture of the
system.

In other words, this team would own the package diagram. In addition, the
architecture team would also own and control the interfaces (the Facades) between the
subsystems. Clearly, as the project progress, changes will need to be made to the
Facades, but those changes must be performed by the central architecture team, and
not the developers working on individual subsystems.

As the architecture team maintain a constant high level view of the system, they are
best placed to understand the impact changes to the interfaces between subsystems
might have.

Example
For a major command and control system, the architecture team make a first cut of the
system architecture by identifying the major areas of functionality to be offered by the
system. They produce the following package diagram:

Figure 92 - First Cut Subsystem Plan using a UML package diagram



Note that the architecture is not set in stone - the architecture team will evolve and
expand the architecture as the project progresses, to contain the complexity of each
subsystem.

The team would continue setting up subsystems until the size of each subsystem is not
too complex, and is easy to manage.

Use Cases may well then be built for each subsystem. Each subsystem is treated as a
system in its own right, exactly as we did in the early stages of the book:

Figure 93 - Parallel Use Case Modelling

Handling Large Use Cases
Another problem with such large scale development is that these first cut use cases
identified at the Elaboration phase may well be far too big to develop in a single
iteration. The solution is not to make the iterations longer (this would cause
complexity to rise again). Rather, the solution is break the Use Case into a series of
easier to manage versions.

For example, the Fire Torpedoes Use Case pictured above is identified, after the
elaboration phase, to be a particularly large and difficult Use Case. The Use Case is
therefore divided into separate versions, as follows:
• Version 1 - allows the opening of bow caps
• Version 2 - allows interlocks to be set
• Version 3 - allows the discharge of weapons

The aim is to ensure that each version is easy to understand, and achievable in a single
iteration. So the Fire Torpedoes Use Case would take three iterations to complete.



The Construction Phase
The construction phase carries on as described in earlier chapters, but with each
subsystem being developed, iteratively, by separate teams, working in parallel and as
independently as possible.

At the end of each iteration, a phase of integration testing will take place, where the
interfaces across subsystems are tested.

Summary
This chapter looked at some of the issues surrounding large scale system
development. It is clear that although the UML is designed to be scaleable,
transferring the Iterative Incremental Framework to large projects is far from a simple
exercise.

The best approach at the moment seems to be the Architecture Centric approach
proposed by Rational Corp:
• Define subsystems from an early stage
• Keep complexity as manageable as possible
• Iterate in parallel but don’t hack interfaces
• Appoint a central architecture team
The package model provided by the UML provides a way of containing the large
complexity, and this model should be owned by the architecture team.

More reading : The Rational Website at www.rational.com provides several
interesting whitepapers on scalability issues such as multi-site working and systems
requiring multiple variants. In addition, reference [1] is an excellent introduction to
the Rational Unified Process and how the architecture centric approach can help
contain complexity.



Chapter 17
Modelling States

After taking a break to consider Inheritance and System Architecture, we are now
going to return to the design stage of the construction phase and consider state
modelling.

State Diagrams allow us to model the possible states that an object can be in. The
model allows us to capture the significant events that can act on the object, and also
the effect of the events.

These models have many applications, but perhaps the strongest application is to
ensure that odd, illegal events cannot happen in our system.

The example given in the introductory chapter of the book (page 31) talks about a
situation that seems to happen an awful lot, if local newspapers are anything to go by -
a gas bill is sent to a customer who died five years ago!

Carefully written state diagrams should prevent these kind of erroneous events
occurring.

Example Statechart

Figure 94 - Example Statechart; Telephone



We will look at the syntax of this diagram in detail shortly, but the basics of the
diagram should be obvious. The sequence of events that can occur to the telephone are
shown, and the states that the telephone can be in are also shown.

For example, from being idle, the telephone can either go to being Off the Hook (if
the receiver is lifted), or the telephone can go to Ringing (if a call is received).

State Diagram Syntax

Figure 95 - Syntax of the State Diagram - an E-Mail example

The diagram above shows most of the state diagram syntax. The object will have a
start state (the filled circle), describing the state of the object at the point of creation.
Most objects have an end state (the bullseye), describing the event that happens to
destory the object.

Some events cause a state transition that causes the object to remain in the same state.
In the example above, the e-mail can receive an edit event only if the status of the
object is unsent. But the event does not cause a state change. This is a useful syntax
to illustrate that the edit event can not happen in any of the other states.



Substates

Figure 96 - Messy State Model

Sometimes, we require a model that describes states within states. The above
statechart is perfectly valid (describing a traffic light object's states), but it is hardly
elegant. Essentially, it can be switched off at any time, and it is this set of events that
is causing the mess.

There is a superstate present in this model. The traffic light can be either On or
Off. When it is in the On state, it can be in a series of substates of Red, Amber
or Green. The UML provides for this by allowing nesting of states:

Figure 97 - Simpler state model using substates

Note that in the diagram above, the small arrow pointing in to the red state indicates
that this is the default state - on commencement of the on state, the light will be set
to Red.



Entry/Exit Events
Sometimes it is useful to capture any actions that need to take place when a state
transition takes place. The following notation allows for this:

Figure 98 - Here, we need to issue a betting slip when the state change occurs

Figure 99 - Here, the ring tone starts on entry to the state - the ring tone stops on
exit

Send Events
The above notation is useful when you need to comment that a particular action needs
to take place. Slightly more formally, we can tie this approach to the idea of objects
and collaboration. If a state transition implies that a message has to be sent to another
object, then the following notation is used (alongside the entry or exit box):

^object.method (parameters)

Figure 100 - formal notation indicating that a message must be sent on the state
transition

Guards
Sometimes we need to insist that a state transition is possible only if a particular
condition is true. This is achieved by placing a condition in square brackets as
follows:



Figure 101 - Here, the transition to the Placed state can only occur if the
balance of the account is in credit

History States
Finally, returning to substates briefly, it is possible to notate that if the superstate is
interrupted in some way, when the superstate is resumed, the state will be
remembered.

Take the following example. A criminal investigation starts in the pending state.
Once it switches to the Being Actioned state, it can be in a number of substates.

However, at random intervals, the case can be audited. During an audit, the
investigation is briefly suspended. Once the audit is complete, the investigation must
resume at the state from where it was before the audit.

The simple history notation (a H in a circle) allows us to do this, as follows:

Figure 102 - History State

Other Uses for State Diagrams
Although the most obvious use for these diagrams is to track the state of an object, in
fact, statecharts can be used for any state-based element of the system. Use Cases are
a clear candidate (for example, a use might only be able to proceed if the user has
logged on).

Even the state of the entire system can be modelled using the statechart - this is
clearly a valuable model for the central architecture team in a large development.



Summary
In this chapter, we looked at State Transition Diagrams.

We saw:

• The syntax of the diagram

• How to use Substates

• Entry and Exit Actions

• Send Events and Guards

• History States

Statecharts are quite simple to produce, but often require deep thought processes

Most commonly produced for Classes, but can be used for anything : Use Cases,
entire Systems, etc



Chapter 18
Transition to Code

This brief section describes some of the issues surrounding the move from the model
to code. For the examples, we'll use Java, but the Java is very simple and can be easily
applied to any modern Object Oriented language.

Synchronising Artifacts
One of the key problems of design and coding is keeping the model in line with the
code.

Some projects will want to totally separate design from code. Here, the designs are
built to be as complete as possible, and coding is considered a purely mechanical
transformation process.

For some projects, the design models will be kept fairly loose, with some design
decisions deferred until the coding stage.

Either way, the code is likely to drift away from the model to a lesser or greater
extent. How do we cope with this?

One approach is to add an extra stage to each iteration - Synchronising artifacts. Here,
the models are altered to reflect the design decisions that were made during coding in
the previous iteration.

Figure 103 - Extra stage in the waterfall - synchronisation



Clearly, this is a far from simple solution, as often, major changes will have been
made. However, it is workable as long as the iterations are short and the complexity of
each one is manageable. Well, that's what we've been aiming for all along!!

Some CASE tools allow reverse engineering - that is, the generation of a model
from code. This could be a help with synchronising - at the end of iteration 1,
regenerate the model from the code, and then work from this new model for iteration
2 (and repeat the process). Having said that, the technology of reverse engineering is
far from advanced, so this may not suit all projects!

Mapping Designs to Code
Your code's class definitions will be derived from the Design Class Diagram. The
method definitions will come largely from the Collaboration Diagrams, but extra help
will come from the Use Case descriptions (for the extra detail, particularly on
exception/alternate flows) and the State Charts (again, for trapping error conditions).

Here's an example class, and what the code might look like:

Figure 104 - The Order Line class, with a couple of example members

The resulting code would end up looking something like this (following a mechanical
conversion process):
public class OrderLine
{
public OrderLine(int qty, SKU product)
{
// constructor
}
public double subtotal()
{
// method definition
}

private int quantity;
}
Figure 105 - Sample Order Line Code

Note that in the code above, I have added a constructor. We omitted the create()
methods from the Class Diagram (as it seems to be a convention these days), so this
needed to be added.



Figure 106 - The aggregation of Order Lines and SKU's

An order line contains a reference to a single SKU, so we also need to add this to the
class code:
public class OrderLine
{
public OrderLine(int qty, SKU product);
public float subtotal();

private int quantity;
private SKU SKUOrdered;
}
Figure 107 - Adding the reference attribute (method blocks omitted for clarity)

What if a class needs to hold a list of references to another class? A good example is
the relationship between Purchase Orders and Purchase Order Lines. A Purchase
Order owns a list of lines, as in the following UML:



Figure 108 - A Purchase Order holds a list of Order Lines

The actual implementation of this depends upon the specific requirement (for
example, should the list be ordered, is performance an issue, etc), but assuming we
need a simple array, the following code will suffice:
public class PurchaseOrder
{
public float total();

private date datePlaced;
private int customerID;
private Vector OrderLineList;
}
Figure 109 - Adding a list of references

Initialising the list would be the job of the constructor. For non Java and C++ coders,
a Vector is simply an array that can be dynamically resized. Depending on the
requirement, a bog standard array would have worked too.

Defining the Methods
The collaboration diagram is a large input into the method definitions.

The following worked example describes the get total method for the Purchase
Order. This method returns the total cost of all of the lines in the order:



Figure 110 - Get total collaboration

Step 1
Clearly, we have a method called getTotal() in the purhcase order class:
public double getTotal()
{

}

Figure 111 - method definition in the Purchase Order Class

Step 2
The collaboration says that the purchase order class now polls through each line:
public double getTotal()
{
double total;
for (int x=0; xorderLineList.size();x++)
{
// extract the OrderLine from the list
theLine = (OrderLine)orderLineList.get(x);

total += theLine).getSubtotal();
}
return total;
}
Figure 112 - code for getting the total, by polling all purchase order lines for the
order.



Step 3
We have called a method called getSubtotal() in the OrderLine class. So this needs
to be implemeted:
public double getSubtotal()
{
return quantity * SKUOrdered.getPrice();
}
Figure 113 - implementation of getSubtotal()

Step 4
We have called a method called getPrice() in the SKU Class. This needs
implementing and would be a simple method that returns the private data member.

Mapping Packages into Code
We stressed that building packages is an essential aspect of system architecture, but
how do we map them into code?

In Java
If you are coding in Java, packages are supported directly. In fact, every single class
in Java belongs to a package. The first line of a class declaration should tell Java in
which package to place the class (if this is omitted, the class is placed in a default
package).

So if the SKU class was in a package called Stock, then the following class header
would be valid:
package com.mycompany.stock;

class SKU
{ ...
Figure 114 - Placing classes in packages

Best of all, Java adds an extra level of visibility on top of the standard private, public
and protected. Java includes package protection. A class can be declared as being
visible only to the classes in the same package - and so can the methods inside a class.
This provides excellent support for encapsulation within packages. By making all
classes visible only to the packages they are contained in (except the facades),
subsystems can truly be developed independently.

Sadly, the syntax for package protection in Java is rather poor. The notation is to
simply declare a class with no public, protected or private preceding the class
definition - exactly as in Figure 114.



In C++
There is no direct support for packages in C++, but recently the concept of a
namespace was added to the language. This allows classes to be placed in separate
logical partitions, to avoid name clashes between namespaces (so I could create two
namespaces, say Stock and Orders, and have a class called SKU in both of them).

This provides some of the support of packages, but unfortunately it doesn't offer any
protection via visibilities. A class in one namespace can access all of the public
classes in another namespace.

The UML Component Model

This model shows a map of the physical, hard, software components (as opposed to
the logical view expressed by the package diagram).

Although the model will often be based on the logical package diagram, it can contain
physical run time elements that weren't necessary at the design stage. For example,
the following diagram shows an example logical model, followed by the eventual
software physical model:

Figure 115 - the logical compared to the physical view

The Component Model is very simple. It works in the same way as the package
diagram, showing elements and the dependencies between them. However, this time,
the symbol is different, and each component can be any physical software entity (an
executable file, a dynamic link library, an object file, a source file, or whatever).



Note that the Component Model is based heavily on the package diagram, but has
added a .dll to handle the Terminal Input/Output, and has added a test harness
executable.

Ada Components
Some extra component icons are available through Rational Rose that seem to be
heavily influenced by the Ada language (presumably through the input of Grady
Booch). These icons work in exactly the same way as the components above, but
notate more specific software components:

Figure 116 - Extra Components

For readers from a non Ada background, a Package (not to be confused with a UML
package) is a collection of related procedures, functions and data (roughly the same as
a class), a Subprogram is a procedure or function, and a Task is a subprogram that can
run concurrently with other tasks.

These symbols may be of use to you even if you are not working in Ada - in particular
the Task symbol is useful to denote that the software element is going to run in
parallel with other tasks.

Summary
This chapter has described, in rough terms, the general process of converting the
models into real code. We looked briefly at the issue of keeping the model
synchronised with the code, and a couple of ideas on how to get around the problem.

We saw the component model. The model is not heavily used at present, but it is
helpful in mapping the physical, real life software code and the dependencies between
them.



The UML Applied Course CD shows how the Case Study followed on the course can
be transformed into Java code - please feel free to explore it for more details.



Bibliography

[1] : Krutchten, Philippe. 2000 The Rational Unified Process An Introduction Second
Edition Addison-Wesley
A brief introduction to the Rational Unified Process, and its relationship with the UML

[2] : Larman, Craig. 1998 Applying UML and Patterns An Introduction to Object
Oriented Analysis and Design Prentice Hall
An excellent introduction to the UML, applied to real software development. Used as the basis for this
course.

[3] : Schmuller, Joseph. 1999 Teach Yourself UML in 24 Hours Sams
A surprisingly comprehensive introduction to UML, including details of the metamodel. The first half
concentrates on UML syntax, and the second half applies the UML (using a RUP-style process called
GRAPPLE)

[4] : Collins, Tony. 1998 Crash : Learning from the World’s Worst Computer
Disasters SimonSchuster
An entertaining collection of case studies exploring why so many software development projects fail

[5] : Kruchten, Phillipe 2000 From Waterfall to Iterative Lifecycle - a tough
transition for project managers Rational Software Whitepaper – www.rational.com
An excellent, and short, description of the problems project managers will face on an iterative project

[6] : Gamma, E., Helm, R., Johnson, R., Vlissides, J. 1995 Design Patterns : Elements
of Reusable Object Oriented Software Addison-Wesley
The classic “Gang of Four” catalogue of several design patterns

[7] : Riel, Arthur 1996 Object Oriented Design Heuristics Addison-Wesley
Rules of Thumb for Object Oriented Designers

[8] : UML Distilled
Martin Fowler's pragmatic approach to applying UML on real software developments

[9] : Kulak, D., Guiney, E. 2000 Use Cases : Requirements in Context Addison-
Wesley
An in depth treatment of requirement engineering, driven by Use Cases


UML Applied

More Related Content

Viewers also liked (20)

Similar to UML Applied (20)

More from Rolf Pinto (13)

Recently uploaded (20)

UML Applied