Oop
0 likes37 views
This document provides an overview of an Object Oriented Programming course. It discusses four parts of the course: computer fundamentals, object-oriented concepts, the Java platform, and design patterns. It notes that the course will use Java as the teaching language but will discuss object-oriented principles more broadly. It also lists some recommended textbooks and online resources for further learning.
![1.4 Types
Types and Variables
We write code like:
The high-level language has a series of primitive
(built-in) types that we use to signify what’s in the
memory
The compiler then knows what to do with them
E.g. An “int” is a primitive type in C, C++, Java and many
languages. It’s usually a 32-bit signed integer
A variable is a name used in the code to refer to a
specific instance of a type
x,y,z are variables above
They are all of type int
int x = 512;
int y = 200;
int z = x+y;
Primitive Types in Java
“Primitive” types are the built in ones.
They are building blocks for more complicated types
that we will be looking at soon.
boolean – 1 bit (true, false)
char – 16 bits
byte – 8 bits as a signed integer (-128 to 127)
short – 16 as a signed integer
int – 32 bits as a signed integer
long – 64 bits as a signed integer
float – 32 bits as a floating point number
double – 64 bits as a floating point number
[See practicals]
19
You met the notion of types in FoCS and within ML. They’re impor-
tant because they allow the compiler to keep track of what the data
in its memory actually means and stop us from doing dumb things
like interpreting a floating point number as an integer.
In ML, you know you can specify the types of variables. But you
almost never do: you usually let the compiler infer the type unless
there’s some ambiguity (e.g. between int and double). In an imper-
ative language such as Java you must explicitly assign the types to
every variable.
For any C/C++ programmers out there: yes, Java looks a lot like
the C syntax. But watch out for the obvious gotcha — a char in C is
a byte (an ASCII character), whilst in Java it is two bytes (a Unicode
character). If you have an 8-bit number in Java you may want to
use a byte, but you also need to be aware that a byte is signed..!
You do lots more work with number representation and primitives in
your Java practical course. You do a lot more on floats and doubles
in your Floating Point course.
20](https://image.slidesharecdn.com/oop-200207222233/85/Oop-10-320.jpg)
![Reference Types (Classes)
Lets imagine you’re creating a program that stores
info about someone doing CS
You might create a set of variables (instances of
types) as follows:
Now you need to represent two people so you add:
Now you need to add more – this is getting rather
messy
Better if we could create our own type – call it a
“Person”
Let it contain two values: forename and surname
String forename = “Kay”;
String surname = “Oss”;
String forename2 = “Don”;
String surname2 = “Keigh”;
Reference Types (Classes)
In Java we create a class which acts as a
blueprint for a custom type
A class:
Has attributes (things it is assigned e.g. name, age)
Has methods (things it can do e.g. walk, think)
When we create an instance of a class we:
Assign memory to hold the attributes
Assign the attributes
We call the instance an object
As in object-oriented programming
public class Person {
String forename;
String surname;
}
You’ve met things called ‘reference types’ in ML so hopefully the idea
of a reference isn’t completely new to you. But if you’re wondering
21
why classes are called reference types in Java, we’ll come to that
soon.
Definitions
Class
A class is a grouping of conceptually-related attributes
and methods
Object
An object is a specific instance of a class
Once we have compiled our Java source code, we end up with a set
of .class files. We can then distribute these files without their source
code (.java) counterparts.
In addition to javac you will also find a javap program which allows
you to poke inside a class file. For example, you can disassemble a
class file to see the raw bytecode using javap -c classfile:
Input:
public class HelloWorld {
public static void main(String[] args) {
System.out.println("Hello World");
}
22](https://image.slidesharecdn.com/oop-200207222233/85/Oop-11-320.jpg)
![}
javap output:
Compiled from "HelloWorld.java"
public class HelloWorld extends java.lang.Object{
public HelloWorld();
Code:
0: aload_0
1: invokespecial #1; //Method java/lang/Object."<init>":()V
4: return
public static void main(java.lang.String[]);
Code:
0: getstatic #2; //Field java/lang/System.out:
//Ljava/io/PrintStream;
3: ldc #3; //String Hello World
5: invokevirtual #4; //Method java/io/PrintStream.println:
//(Ljava/lang/String;)V
8: return
}
This probably won’t make a lot of sense to you right now: that’s OK.
Just be aware that we can view the bytecode and that sometimes this
can be a useful way to figure out exactly what the JVM will do with
a bit of code. You aren’t expected to know bytecode.
23
1.5 Pointers and References
Pointers
In some languages we have variables that hold
memory addresses.
These are called pointers
A pointer is just the memory address of the first
memory slot used by the object
The pointer type tells the compiler how many
slots the whole object uses
xptr2
xxptr1
MyType x;
MyType *xptr1 = &x;
MyType *xptr2 = xptr1;
C++
A pointer is just the name given to memory addresses when we handle
them in code. They are very powerful concepts to work with, but
you need to take a lot of care.
Java doesn’t expose pointers to the programmer, so we’ll use some
basic C examples to make some points in the lectures. There are just
a few bits of C you may need to know to follow:
• int x declares a variable x that is a 32-bit signed integer
• int *p declares a pointer p that can point to an int in memory.
It must be manually initialised to point somewhere useful.
• p=&x Gets the memory address for the start of variable x and
stick it in p so p is a pointer to x.
• int a=*p declares a new int variable called a and sets its value to
that pointed to by pointer p. We say that we are dereferencing
24](https://image.slidesharecdn.com/oop-200207222233/85/Oop-12-320.jpg)
![Q5. Draw box diagrams like the one above to illustrate what hap-
pens in each step of the following Java code in memory:
Person p = null;
Person p2 = new Person();
p = p2;
p2 = new Person();
p=null;
Pass By Value and By Reference
A. “70 Bob”
B. “70 Alice”
C.“71 Bob”
D.“71 Alice”
void myfunction(int x, Person p) {
x=x+1;
p.name=“Alice”;
}
void static main(String[] arguments) {
int num=70;
Person person = new Person();
person.name=“Bob”;
myfunction(num, p);
System.out.println(num+” “+person.name)
}
Now things are getting a bit more involved. We have added a function
prototype into the mix: void myfunction(int x, Person p). This means
it returns nothing (codenamed void) but takes an int and a Person
object as input.
When the function is run, the computer takes a copy of the inputs
and creates local variables x and p to hold the copies.
29
Here’s the thing: if you copy a primitive type, the value gets copied
across (so if we were watching memory we would see two integers in
memory, the second with the same value as the first).
But for the reference type, it is the reference that gets copied and
not the object it points to (i.e. we just create a new reference p and
point it at the same object).
When we copy a primitive value we say that it is pass by value. When
we send a reference to the object rather than the object, it is pass
by reference.
In my opinion, this all gets a bit confusing in Java because you have
no choice but to use references (unlike in, say C++). So I prefer to
think of all variables as representing either primitives or references
(rather than primitives and actual objects), and when we pass any
variable around its value gets copied. The ‘value’ of a primitive is its
actual value, whilst the ‘value’ of a reference is a memory address.
You may or may not like this way of thinking: that’s fine. Just
make certain you understand the different treatment of primitives
and reference types.
30](https://image.slidesharecdn.com/oop-200207222233/85/Oop-15-320.jpg)
![The “has-a” Association
College Student1 0...*
Arrow going left to right says “a College has zero or
more students”
Arrow going right to left says “a Student has exactly 1
College”
What it means in real terms is that the College class
will contain a variable that somehow links to a set of
Student objects, and a Student will have a variable
that references a College object.
Note that we are only linking classes: we don't start
drawing arrows to primitive types.
The graphical notation used here is part of UML (Unified Modeling
Language). UML is basically a standardised set of diagrams that
can be used to describe software independently of any programming
language used to create it.
UML contains many different diagrams (touched on in the Software
Design course). Here we just use the UML class diagram such as the
one in the slide.
35
Anatomy of an OOP Program (Java)
public class MyFancyClass {
public int someNumber;
public String someText;
public void someMethod() {
}
public static void main(String[] args) {
MyFancyClass c = new
MyFancyClass();
}
}
Class name
Class state (properties
that an object has such as
colour or size)
Class behaviour (actions
an object can do)
'Magic' start point
for the program
(named main by
convention)
Create an object of
type MyFancyClass in
memory and get a
reference to it
There are a couple of interesting things to note for later discussion.
Firstly, the word public is used liberally. Secondly, the main function
is declared inside the class itself and as static. Finally there is the
notation String[] which represents an array of String objects in Java.
You will see arrays in the practicals.
36](https://image.slidesharecdn.com/oop-200207222233/85/Oop-18-320.jpg)
![Recap
Important OOP concepts you need to
understand:
Modularity (classes, objects)
Data Encapsulation
Inheritance
Abstraction
Polymorphism
2.5 Lifecycle of an Object
Constructors
You will have noticed that the RHS looks rather like a
function call, and that's exactly what it is.
It's a method that gets called when the object is
constructed, and it goes by the name of a constructor
(it's not rocket science).
We use constructors to initialise the state of the class
in a convenient way.
A constructor has the same name as the class
A constructor has no return type specified
MyObject m = new MyObject();
55
The return we expect is obviously a reference to the newly created
object, which the compiler sorts out for us.
Examples
public class Person {
private String mName;
// Constructor
public Person(String name) {
mName=name;
}
public static void main(String[] args) {
Person p = new Person(“Bob”);
}
}
class Person {
private:
std::string mName;
public:
Person(std::string &name) {
mName=name;
}
};
int main(int argc, char ** argv) {
Person p (“Bob”);
}
Java C++
Default Constructor
public class Person {
private String mName;
public static void main(String[] args) {
Person p = new Person();
}
}
If you specify no constructor
at all, the Java fills in an
empty one for you
The default constructor
takes no arguments
56](https://image.slidesharecdn.com/oop-200207222233/85/Oop-28-320.jpg)
![Beware: As soon as you specify any constructor whatsoever, Java
won’t add in a default constructor...
Multiple Constructors
public class Student {
private String mName;
private int mScore;
public Student(String s) {
mName=s;
mScore=0;
}
public Student(String s, int sc) {
mName=s;
mScore=sc;
}
public static void main(String[] args) {
Student s1 = new Student("Bob");
Student s2 = new Student("Bob",55);
}
}
You can specify as many
constructors as you like.
Each constructor must have
a different signature
(argument list)
Constructor Chaining
When you construct an object of a type with
parent classes, we call the constructors of all
of the parents in sequence
Student s = new Student();
Animal
Person
Student
1. Call Animal()
2. Call Person()
3. Call Student()
57
What actually happens is that the first line of a constructor always
starts with super(), which is a call to the parent constructor (which
itself starts with super(), etc.). If it does not, the compiler adds one
for you:
public class Person {
public Person() {
}
}
becomes:
public class Person {
public Person() {
super();
}
}
This can get messy though: what if the parent does not have a
default constructor? In this case, the code won’t compile, and we will
have to manually add a call to super ourselves, using the appropriate
arguments. E.g.
public class Person {
protected String mName;
public Person(String name) {
mName=name;
}
}
58](https://image.slidesharecdn.com/oop-200207222233/85/Oop-29-320.jpg)
![Cleaning Up (Java)
Good:
System cleans up after us
Bad:
It has to keep searching for objects with
no references. This requires effort on
the part of the CPU so it degrades
performance.
We can't easily predict when an object
will be deleted
Cleaning Up (Java)
So we can't tell when a destructor would run – so
Java doesn't have them!!
It does have the notion of a finalizer that gets
run when an object is garbage collected
BUT there's no guarantee an object will ever
get garbage collected in Java...
Garbage Collection != Destruction
61
2.6 Class-Level Data
Class-Level Data and Functionality
Imagine we have a class ShopItem. Every
ShopItem has an individual core price to
which we need to add VAT
Two issues here:
1. If the VAT rate changes, we need to find
every ShopItem object and run
SetVATRate(...) on it. We could end up
with different items having different VAT
rates when they shouldn't...
2. It is inefficient. Every time we create a
new ShopItem object, we allocate another
32 bits of memory just to store exactly the
same number!
public class ShopItem {
private float price;
private float VATRate = 0.175;
public float GetSalesPrice() {
return price*(1.0+VATRate);
}
public void SetVATRate(float rate) {
VATRate=rate;
}
}
What we have is a piece of information that is class-level not object level
Each individual object has the same value at all times
We throw in the static keyword:
public class ShopItem {
private float price;
private static float VATRate;
....
}
Variable created only once
and has the lifetime of the
program, not the object
Class-Level Data and Functionality
We now have one place to update
More efficient memory usage
17.5
17.5
17.5
17.5
17.5
17.5
Can also make methods static too
A static method must be instance independent i.e. it can't rely on member
variables in any way
Sometimes this is obviously needed. E.g
public class Whatever {
public static void main(String[] args) {
...
}
}
Must be able to run this
function without creating an
object of type Whatever
(which we would have to do in
the main()..!)
62](https://image.slidesharecdn.com/oop-200207222233/85/Oop-31-320.jpg)
![Java Cloning
So do you want shallow or deep?
The default implementation of clone() performs a
shallow copy
But Java developers were worried that this might not be
appropriate: they decided they wanted to know for sure
that we'd thought about whether this was appropriate
Java has a Cloneable interface
If you call clone on anything that doesn't extend this
interface, it fails
Marker Interfaces
If you go and look at what's in the Cloneable interface,
you'll find it's empty!! What's going on?
Well, the clone() method is already inherited from Object
so it doesn't need to specify it
This is an example of a Marker Interface
A marker interface is an empty interface that is used to
label classes
This approach is found occasionally in the Java libraries
You might also see these marker interfaces referred to as tag inter-
faces. They are simply a way to label or tag a class. They can be very
73
useful, but equally they can be a pain (you can’t dynamically tag a
class, nor can you prevent a tag being inherited by all subclasses).
Q15. An alternative strategy to clone()-ing an object is to provide a
copy constructor. This is a constructor that takes the enclosing
class as an argument and copies everything manually:
public class MyClass {
private String mName;
private int[] mData;
// Copy constructor
public MyClass(MyClass toCopy) {
this.mName = toCopy.mName;
// TODO
}
...
}
(a) Complete the copy constructor.
(b) Make MyClass clone()-able (you should do a deep copy).
(c) * Why might the Java designers have disliked copy con-
structors? [Hint: What happens if you want to copy an
object that is being referenced using its parent type?].
(d) * Under what circumstances is a copy constructor a good
solution?
Q16. Consider the class below. What difficulty is there in providing
a deep clone() method for it?
public class CloneTest {
private final int[] mData = new int[100];
}
74](https://image.slidesharecdn.com/oop-200207222233/85/Oop-37-320.jpg)
![...
public static void main(String[]) {
// Always gets the same object
Database d = System.database;
}
This is really just global variables by the back door.
Nothing fundamentally prevents us from making multiple Database
objects!
Solution 3: Create an instance of Database at startup, and pass
it as a constructor parameter to every Session we create, storing a
reference in a member variable for later use.
public class System {
public System(Database d) {...}
}
public class Session {
public Session(Database d) {...}
}
...
public static void main(String[]) {
Database d = new Database();
System sys = new System(d);
Session sesh = new Session(d);
115
}
This solution could work, but it doesn’t enforce that only one
Database be instantiated – someone could quite easily create
a new Database object and pass it around.
We start to clutter up our constructors.
It’s not especially intuitive. We can do better.
Solution 4: (Singleton) Let’s adapt Solution 2 as follows. We
will have a single static instance. However we will access it through a
static member function. This function, getInstance() will either cre-
ate a new Database object (if it’s the first call) or return a reference
to the previously instantiated object.
Of course, nothing stops a programmer from ignoring the getInstance()
function and just creating a new Database object. So we use a neat
trick: we make the constructor private or protected. This means code
like new Database() isn’t possible from an arbitrary class.
Guarantees that there will be only one instance.
116](https://image.slidesharecdn.com/oop-200207222233/85/Oop-58-320.jpg)
![4.12.1 Classifying Patterns
Often patterns are classified according to what their intent is or what
they achieve. The original book defined three classes:
Creational Patterns . Patterns concerned with the creation of
objects (e.g. Singleton, Abstract Factory).
Structural Patterns . Patterns concerned with the composition
of classes or objects (e.g. Composite, Decorator, Proxy).
Behavioural Patterns . Patterns concerned with how classes or
objects interact and distribute responsibility (e.g. Observer,
State, Strategy).
4.12.2 Other Patterns
You’ve now met eight Design Patterns. There are plenty more (23
in the original book), but this course will not cover them. What has
been presented here should be sufficient to:
• Demonstrate that object-oriented programming is powerful.
• Provide you with (the beginnings of) a vocabulary to describe
your solutions.
• Make you look critically at your code and your software archi-
tectures.
• Entice you to read further to improve your programming.
Of course, you probably won’t get it right first time (if there even
s a ‘right’). You’ll probably end up refactoring your code as new
situations arise. However, if a Design Pattern is appropriate, you
should probably use it.
133
4.12.3 Performance
Note that all of the examples here have concentrated on structuring
code to be more readable and maintainable, and to incorporate con-
straints structurally where possible. At no point have we discussed
whether the solutions perform better. Many of the solutions exploit
runtime polymorphic behaviour, for example, and that carries with
it certain overheads.
This is another reason why you can’t apply Design Patterns blindly.
[This is a good thing since, if it wasn’t true, programming wouldn’t
be interesting, and you wouldn’t get jobs!].
134](https://image.slidesharecdn.com/oop-200207222233/85/Oop-67-320.jpg)
Oop
- 1. Object Oriented Programming Dr Robert Harle IA NST CS and CST Lent 2009/10 OO Programming This is a new course this year that absorbs what was “Programming Methods” and provides a more formal look at Object Oriented programming with an emphasis on Java Four Parts Computer Fundamentals Object-Oriented Concepts The Java Platform Design Patterns and OOP design examples If we teach Java in isolation, there’s a good chance that students don’t manage to mentally separate the object-oriented concepts from 1 Java’s implementation of them. Understanding the underlying prin- ciples of OOP allows you to transition quickly to a new OOP lan- guage. Because Java is the chosen teaching language here, the vast majority of what I do will be in Java, but with the occasional other language thrown in to make a point. Actually learning to program is best done practically. That’s why you have your practicals. I don’t want to concentrate on the minu- tiae of programming, but rather the larger principles. It might be that some of this stuff will make more sense when you’ve done your practicals; I hope that this material will help set you up to complete your practicals quickly and efficiently. But we’ll see. Java Ticks This course is meant to complement your practicals in Java Some material appears only here Some material appears only in the practicals Some material appears in both: deliberately! A total of 7 workbooks to work through Everyone should attend every week CST: Collect 7 ticks NST: Collect at least 5 ticks 2
- 2. Books and Resources OOP Concepts Look for books for those learning to first program in an OOP language (Java, C++, Python) Java: How to Program by Deitel & Deitel (also C++) Thinking in Java by Eckels Java in a Nutshell (O' Reilly) if you already know another OOP language Lots of good resources on the web Design Patterns Design Patterns by Gamma et al. Lots of good resources on the web Books and Resources Also check the course web page Updated notes (with annotations where possible) Code from the lectures Sample tripos questions http://www.cl.cam.ac.uk/teaching/0910/OOProg/ There really is no shortage of books and websites describing the basics of object oriented programming. The concepts themselves 3 are quite abstract, but most texts will use a specific language to demonstrate them. The books I’ve given favour Java (because that’s the primary language you learn this term). You shouldn’t see that as a dis-recommendation for other books. In terms of websites, SUN produce a series of tutorials for Java, which cover OOP: http://java.sun.com/docs/books/tutorial/ but you’ll find lots of other good resources if you search. And don’t forget your practical workbooks, which will aim not to assume any- thing from these lectures. 4
- 3. Chapter 1 Computer Fundamentals What can Computers Do? The computability problem Given infinite computing 'power' what can we do? How do we do it? What can't we do? Option 1: Forget any notion of a physical machine and do it all in maths Leads to an abstract mathematical programming approach that uses functions Gets us Functional Programming (e.g. ML) Option 2: Build a computer and extrapolate what it can do from how it works Not so abstract. Now the programming language links closely to the hardware This leads naturally to imperative programming (and on to object-oriented) λ 5 What can Computers Do? The computability problem Both very different (and valid) approaches to understanding computer and computers Turns out that they are equivalent Useful for the functional programmers since if it didn't, you couldn't put functional programs on real machines... WWII spurred an interest in machinery that could compute. During the war, this interest was stoked by a need to break codes, but also to compute relatively mundane quantities such as the trajectory of an artillery shell. After the war, interest continued in the abstract notion of ‘computability’. Brilliant minds (Alan Turing, etc) began to wonder what was and wasn’t ‘computable’. They defined this in an abstract way: given an infinite computing power (whatever that is), could they solve anything? Are there things that can’t be solved by machines? Roughly two approaches to answering these questions appeared: Maths, maths, maths. Ignore the mechanics of any real machine and imagine a hypothetical machine that had infinite compu- tation power, then write some sort of ‘programming language’ for it. This language turned out to be a form of mathematics known as ‘Lambda calculus’. It was based around the notion of functions in a mathematical sense. 6
- 4. Build it and they will come. Understand how to build and use a real computing machine. This resulted in the von Neumann architecture and the notion of a Turing machine (basically what we would call a computer). It turned out that both approaches were useful for answering the fundamental questions. In fact, they can be proven to be the same thing now! Imperative This term you transition from functional (ML) to imperative (Java) Most people find imperative more natural, but each has its own strengths and weaknesses Because imperative is a bit closer to the hardware, it does help to have a good understanding of the basics of computers. All the CST Students have this All the NST students don't... yet. All this is very nice, but why do you care? Well, there’s a legacy from these approaches, in the form of two programming paradigms: Functional Languages. These are very mathematically oriented and have the notion of functions as the building blocks of com- putation. 7 Imperative or Procedural Languages. These have variables (state) and procedures as the main building blocks1 . Such languages are well known—e.g. C—and include what we call object- oriented languages such as C++ and Java. Note that I have pluralised “Language” in the above sentences. Terms like “Object-oriented” are really a set of ideas and concepts that var- ious languages implement to varying degrees. Last term you toured around Computer Science (in FoCS) and used a particular language to do it (ML). Although predominantly a func- tional programming language, ML has acquired a few imperative ‘features’ so you shouldn’t consider it as a pure functional language. (For example, the ML reference types you looked at are not func- tional!). This term you will shift attention to an object-oriented language in the form of Java. What we will be doing in this course is looking at the paradigm of object-oriented programming itself so you can better understand the underlying ideas and separate the Java from the paradigm. 1.1 Hardware Fundamentals As I said before, the imperative style of programming maps quite easily to the underlying computer hardware. A good understand- ing of how computers work can greatly improve your programming capabilities with an imperative language. Thing is, only around a half of you have been taught the fundamentals of a computer (the 1A procedure is a function-like chunk of code that takes inputs and gives outputs. However, unlike a formal function it can have side effects i.e. can make non-local changes to something other than its inputs and declared outputs. 8
- 5. CST half). Those people will undoubtedly be rather bored by what follows in the next few pages, but it’s necessary so that everyone is at the same point when we start to delve deeper into object-oriented programming. Computers do lots of very simple things very fast. Over the years we have found optimisation after optimisation to make the simple processes that little bit quicker, but really the fundamentals involve some memory to store information and a CPU to perform simple actions on small chunks of it. We use lots of different types of memory, but conceptually only two are of interest here. System memory is a very large pool of memory (the 2GB or so you get when you buy a machine). Then there are some really fast, but small chunks of memory called registers. These are built into the CPU itself. The CPU acts only on the chunks in the registers so the computer is constantly loading chunks of data from system memory into registers and operating on the register values. The example that follows loads in two numbers and adds them. Those doing the Operating Systems course will realise this is just the fetch-execute cycle. Basically, there is a special register called the program counter (marked P) that tells the computer where to look to get its next instruction. Here I’ve made up some operations: LAM. LOAD the value in memory slot A into register M AMNO. ADD the values in registers M and N and put the result in register O. SMA. STORE the value in register M into memory slot A 9 Dumb Model of a Computer L6X L7Y AXYZ SZ8 Registers Memory 63 12 0 1 2 3 4 5 6 7 8 1P X Y Z ALU CPU Dumb Model of a Computer L6X L7Y AXYZ SZ8 Registers Memory 63 12 0 1 2 3 4 5 6 7 8 63 2P X Y Z ALU CPU 10
- 6. Dumb Model of a Computer L6X L7Y AXYZ SZ8 Registers Memory 63 12 0 1 2 3 4 5 6 7 8 12 63 3P X Y Z ALU CPU Dumb Model of a Computer L6X L7Y AXYZ SZ8 Registers Memory 63 12 0 1 2 3 4 5 6 7 8 75 12 63 4P X Y Z ALU CPU All a computer does is execute instruction after instruction. Never forget this when programming! 11 Memory (RAM) You probably noticed we view memory as a series of slots Each slot has a set size (1 byte or 8 bits) Each slot has a unique address Each address is a set length of n bits Mostly n=32 or n=64 in today’s world Because of this there is obviously a maximum number of addresses available for any given system, which means a maximum amount of installable memory Memory 0 1 2 3 4 5 6 7 8 We slice system memory up into 8-bit (1 byte) chunks and give each one an address (i.e. a slot number). Each of the registers is a small, fixed size. On today’s machines, they are usually just 32 or 64 bits. Since we have to be able to squeeze a memory address into a single register, the size of the registers dictates how much of the system memory we can use. Q1. How much system memory can we use if we have 8, 32 or 64 bit registers? 12
- 7. Big Numbers So what happens if we can’t fit the data into 8 bits e.g. the number 512? We end up distributing the data across (consecutive) slots Now, if we want to act on the number as a whole, we have to process each slot individually and then combine the result Perfectly possible, but who wants to do that every time you need an operation? Memory 0 1 2 3 4 5 6 7 8 If I have two lots of 128-bits of data, how do I add them together on a 32 bit machine? I can’t squeeze 128 bits into 32 bits. I have to chop the 128 bits up into 32 bit chunks, add corresponding chunks together and then put those new chunks back together again to get a 128-bit answer. Can you imagine having to do this every time you needed to do an addition? Nasty. 13 High Level Languages Instead we write in high-level languages that are human readable (well, compsci-readable anyway) High-Level Source Code (e.g. C++, Java) Binary executable (i.e. CPU-speak) Compiler You already know the solution from FoCS and ML. We build up layers of abstractions and write tools to drill down from one layer to the next. We program in high level code and leave it to the compiler to figure out that when we say c=a+b; we actually mean “chop up a and b, load them into registers, add them, look out for overflow, then put all the pieces back together in memory somewhere called c”. 14
- 8. 1.2 Functional and Imperative Revisited We said imperative programming developed from consideration of the hardware. Given the hardware we’ve just designed, the logical thing to do is to represent data using state. i.e. explicit chunks of memory used to store data values. A program can then be viewed as running a series of commands that alter (‘mutate’) that state (hopefully in a known way) to give the final result. So c=a+b is referring to three different chunks of memory that repre- sent numbers. Unlike in ML, it is perfectly valid to say a=a+b in an imperative language, which computes the sum and then overwrites the state stored in a. Functional programs, on the other hand, can be viewed as indepen- dent of the machine. Strictly speaking, this means they can’t have state in the same way since there’s simply nowhere to put it (no notion of memory). Instead of starting with some data that gets mutated to give a result, we only have an input and an output for each function. Of course, ML runs on real machines and therefore makes use of real hardware components such as memory. But that’s because someone has mapped the ideal mathematical processes in functional code to machine operations (this must be possible because I said both the mathematical view and the system view are equivalent). The key point is that, although functional code runs on real hardware, it is independent of it since it is really a mathematical construct. Most people find the imperative approach easier to understand and, in part, that is why it has gained such a stronghold within modern programming languages. 15 1.3 Machine Architectures Machine Architectures Actually, there’s no reason for e.g ARM and Intel to use the same instructions for anything – and they don’t! The result? Any set of instructions for one processor only works on another processor if they happen to use the same instruction set... i.e. The binary executable produced by the compiler is CPU- specific We say that each set of processors that support a given set of instructions is a different architecture E.g. x86, MIPS, SPARC, etc. But what if you want to run on different architectures?? Compilation So what do we have? We need to write code specifically for each family of processors... Aarrggh! The ‘solution’: Source Code (e.g. C+ +) Binary executable for PC (x86) Binary executable for ARM C++ Compiler for ARM C++ Compiler for x86 16
- 9. Remember those weird commands I made up when we did the fetch- execute cycle? Well, not all CPUs are equivalent: newer ones have cool new instructions to do things better. All that means is that a set of instructions for CPU X won’t mean anything to CPU Y unless the manufacturers have agreed they will. Q2. Why do you think CPU manufacturers haven’t all agreed on a single set of instructions? There is one very popular set of machine instructions known as x86. Basically, these were what IBM came up with for the original PC. Now the term ‘PC’ has come to mean a machine that supports at least the set of x86 instructions. Manufacturers like Apple tradition- ally chose not to do so, meaning that Mac applications couldn’t run on PC hardware and vice-versa. Q3. Apple recently started using Intel processors that support x86 instructions. This means Apple machines can now run Mi- crosoft Windows. However, off-the-shelf PC software (which is compiled for x86) does not run on a Mac that is using the Apple operating system compiled for the Intel processor. Why not? 17 Enter Java Sun Microcomputers came up with a different solution They conceived of a Virtual Machine – a sort of idealised computer. You compile Java source code into a set of instructions for this Virtual Machine (“bytecode”) Your real computer runs a program (the “Virtual machine” or VM) that can efficiently translate from bytecode to local machine code. Java is also a Platform So, for example, creating a window is the same on any platform The VM makes sure that a Java window looks the same on a Windows machine as a Linux machine. Sun sells this as “Write Once, Run Anywhere” Java compiles high-level source code not into CPU-specific instruc- tions but into a generic CPU instruction set called bytecode. Ini- tially there were no machines capable of directly using bytecode, but recently they have started to appear in niche areas. For the most part, however, you use a virtual machine. This is a piece of soft- ware that has been compiled specifically for your architecture (like most off-the-shelf software) and which acts an an interpreter, con- verting bytecode instructions into meaningful instructions for your local CPU. All the CPU-specific code goes into the virtual machine, which you should just think of as a piece of intermediary software that translates commands so the local hardware can understand. SUN publishes the specification of a Java Virtual Machine (JVM) and anyone can write one, so there are a plenty available if you want to explore. Start here: http://java.sun.com/docs/books/jvms/ 18
- 10. 1.4 Types Types and Variables We write code like: The high-level language has a series of primitive (built-in) types that we use to signify what’s in the memory The compiler then knows what to do with them E.g. An “int” is a primitive type in C, C++, Java and many languages. It’s usually a 32-bit signed integer A variable is a name used in the code to refer to a specific instance of a type x,y,z are variables above They are all of type int int x = 512; int y = 200; int z = x+y; Primitive Types in Java “Primitive” types are the built in ones. They are building blocks for more complicated types that we will be looking at soon. boolean – 1 bit (true, false) char – 16 bits byte – 8 bits as a signed integer (-128 to 127) short – 16 as a signed integer int – 32 bits as a signed integer long – 64 bits as a signed integer float – 32 bits as a floating point number double – 64 bits as a floating point number [See practicals] 19 You met the notion of types in FoCS and within ML. They’re impor- tant because they allow the compiler to keep track of what the data in its memory actually means and stop us from doing dumb things like interpreting a floating point number as an integer. In ML, you know you can specify the types of variables. But you almost never do: you usually let the compiler infer the type unless there’s some ambiguity (e.g. between int and double). In an imper- ative language such as Java you must explicitly assign the types to every variable. For any C/C++ programmers out there: yes, Java looks a lot like the C syntax. But watch out for the obvious gotcha — a char in C is a byte (an ASCII character), whilst in Java it is two bytes (a Unicode character). If you have an 8-bit number in Java you may want to use a byte, but you also need to be aware that a byte is signed..! You do lots more work with number representation and primitives in your Java practical course. You do a lot more on floats and doubles in your Floating Point course. 20
- 11. Reference Types (Classes) Lets imagine you’re creating a program that stores info about someone doing CS You might create a set of variables (instances of types) as follows: Now you need to represent two people so you add: Now you need to add more – this is getting rather messy Better if we could create our own type – call it a “Person” Let it contain two values: forename and surname String forename = “Kay”; String surname = “Oss”; String forename2 = “Don”; String surname2 = “Keigh”; Reference Types (Classes) In Java we create a class which acts as a blueprint for a custom type A class: Has attributes (things it is assigned e.g. name, age) Has methods (things it can do e.g. walk, think) When we create an instance of a class we: Assign memory to hold the attributes Assign the attributes We call the instance an object As in object-oriented programming public class Person { String forename; String surname; } You’ve met things called ‘reference types’ in ML so hopefully the idea of a reference isn’t completely new to you. But if you’re wondering 21 why classes are called reference types in Java, we’ll come to that soon. Definitions Class A class is a grouping of conceptually-related attributes and methods Object An object is a specific instance of a class Once we have compiled our Java source code, we end up with a set of .class files. We can then distribute these files without their source code (.java) counterparts. In addition to javac you will also find a javap program which allows you to poke inside a class file. For example, you can disassemble a class file to see the raw bytecode using javap -c classfile: Input: public class HelloWorld { public static void main(String[] args) { System.out.println("Hello World"); } 22
- 12. } javap output: Compiled from "HelloWorld.java" public class HelloWorld extends java.lang.Object{ public HelloWorld(); Code: 0: aload_0 1: invokespecial #1; //Method java/lang/Object."<init>":()V 4: return public static void main(java.lang.String[]); Code: 0: getstatic #2; //Field java/lang/System.out: //Ljava/io/PrintStream; 3: ldc #3; //String Hello World 5: invokevirtual #4; //Method java/io/PrintStream.println: //(Ljava/lang/String;)V 8: return } This probably won’t make a lot of sense to you right now: that’s OK. Just be aware that we can view the bytecode and that sometimes this can be a useful way to figure out exactly what the JVM will do with a bit of code. You aren’t expected to know bytecode. 23 1.5 Pointers and References Pointers In some languages we have variables that hold memory addresses. These are called pointers A pointer is just the memory address of the first memory slot used by the object The pointer type tells the compiler how many slots the whole object uses xptr2 xxptr1 MyType x; MyType *xptr1 = &x; MyType *xptr2 = xptr1; C++ A pointer is just the name given to memory addresses when we handle them in code. They are very powerful concepts to work with, but you need to take a lot of care. Java doesn’t expose pointers to the programmer, so we’ll use some basic C examples to make some points in the lectures. There are just a few bits of C you may need to know to follow: • int x declares a variable x that is a 32-bit signed integer • int *p declares a pointer p that can point to an int in memory. It must be manually initialised to point somewhere useful. • p=&x Gets the memory address for the start of variable x and stick it in p so p is a pointer to x. • int a=*p declares a new int variable called a and sets its value to that pointed to by pointer p. We say that we are dereferencing 24
- 13. p. If it helps, we can map these operations to ML something like2 : C ML Comments int x let val xm=0 in... xm is immutable int *p let val pm=ref 0 in... pm is a reference to a vari- able initially set to zero p=&x val pm = ref xm int a=*p val am = !pm The key point to take away is that a pointer is just a number that maps to a particular byte in memory. You can set that number to anything you like, which means there’s no guarantee that it points to anything sensible..! And the computer just obeys — if you say that a pointer points to an int, then it doesn’t argue. Working with pointers helps us to avoid needless copies of data and usually speeds up our algorithms (see next term’s Algorithms I course). But a significant proportion of program crashes are caused by trying to read or write memory that hasn’t been allocated or initialised... 2Now, if all that reference type stuff in ML confused you last term, don’t worry about making these mappings just now. Just be aware that what will come naturally in imperative programming has been retrofitted to ML, where it seems a little less obvious! For now, just concentrate on understanding pointers and references from scratch. 25 Java’s Solution? You can’t get at the memory directly. So no pointers, no jumping around in memory, no problem. Great for teaching. It does, however, have references These are like pointers except they are guaranteed to point to either an object in memory or “null”. So if the reference isn’t null, it is valid In fact, all objects are accessed through references in Java Variables are either primitive types or references! Remember that classes are called Reference Types in Java. This is because they are only handled by reference. Code like: MyClass m = new MyClass(); creates an instance of MyClass somewhere in main memory (that’s what goes on on the right of the equals). And then it creates a reference to that instance and stores the result in a variable m. A reference is really just a special kind of pointer. Remember that a pointer is just a number (representing a memory slot). We can set a pointer to any number we like and so it is possible that dereferencing it gets us garbage. A reference, however, always has a valid address or possibly a special value to signify it is invalid. If you dereference a reference without the special value in it, the memory you read will contain something sensible. 26
- 14. In Java, the ‘special value’ is signified using the keyword null: MyClass m = null; // Not initialised ... m = new MyClass(); If you try to do something with a reference set to null, Java will bail out on you. To test whether a reference is null:3 if (m==null) { ... } Q4. Pointers are problematic because they might not point to any- thing useful. A null reference doesn’t point to anything useful. So what is the advantage of using references over pointers? 3Note that the Java/C syntax uses = for assignment and == for comparison. We’ll discuss this in more detail later. 27 References P1 (50) P2 (50) Person object (name=“Bob”) r2 r1 P1 (50) P2 (73) Person object (name=“Steve”) r2 r1 int p1 = 50; int p2 = p1; Person r1 = new Person(); r1.name=“Bob”; Person r2 = p; p2 = 73; r2.name=“Steve”; So we have primitive types and reference types. A variable for a primitive type can be thought of as being an instance of a type, so each you assign something to it, the same bit of memory gets changed as with P1 and P2 in the above slide. A variable for a reference type holds the memory address. If you assign something to it, what it used to be assigned to still exists in memory (but your variable doesn’t link to it any more). Multiple variables can reference the same object in memory (like r1 and r2 above). 28
- 15. Q5. Draw box diagrams like the one above to illustrate what hap- pens in each step of the following Java code in memory: Person p = null; Person p2 = new Person(); p = p2; p2 = new Person(); p=null; Pass By Value and By Reference A. “70 Bob” B. “70 Alice” C.“71 Bob” D.“71 Alice” void myfunction(int x, Person p) { x=x+1; p.name=“Alice”; } void static main(String[] arguments) { int num=70; Person person = new Person(); person.name=“Bob”; myfunction(num, p); System.out.println(num+” “+person.name) } Now things are getting a bit more involved. We have added a function prototype into the mix: void myfunction(int x, Person p). This means it returns nothing (codenamed void) but takes an int and a Person object as input. When the function is run, the computer takes a copy of the inputs and creates local variables x and p to hold the copies. 29 Here’s the thing: if you copy a primitive type, the value gets copied across (so if we were watching memory we would see two integers in memory, the second with the same value as the first). But for the reference type, it is the reference that gets copied and not the object it points to (i.e. we just create a new reference p and point it at the same object). When we copy a primitive value we say that it is pass by value. When we send a reference to the object rather than the object, it is pass by reference. In my opinion, this all gets a bit confusing in Java because you have no choice but to use references (unlike in, say C++). So I prefer to think of all variables as representing either primitives or references (rather than primitives and actual objects), and when we pass any variable around its value gets copied. The ‘value’ of a primitive is its actual value, whilst the ‘value’ of a reference is a memory address. You may or may not like this way of thinking: that’s fine. Just make certain you understand the different treatment of primitives and reference types. 30
- 16. 31 Chapter 2 Object-Oriented Concepts Modularity A class is a custom type We could just shove all our data into a class The real power of OOP is when a class corresponds to a concept E.g. a class might represent a car, or a person Note that there might be sub-concepts here A car has wheels: a wheel is a concept that we might want to embody in a separate class itself The basic idea is to figure out which concepts are useful, build and test a class for each one and then put them all together to make a program The really nice thing is that, if we've done a good job, we can easily re-use the classes we have specified again in other projects that have the same concepts. “Modularity” 32
- 17. Modularity is extremely important in OOP. It’s the usual CS trick: break big problems down into chunks and solve each chunk. In this case, we have large programs, meaning scope for lots of coding bugs. We split the program into modules. The goal is: • Modules are conceptually easier to handle • Different people can work on different modules simultaneously • Modules may be re-used in other software projects • Modules can be individually tested as we go (unit testing) The module ‘unit’ in OOP is a class. State and Behaviour An object/class has: State Properties that describe that specific instance E.g. colour, maximum speed, value Behaviour/Functionality Things that it can do These often mutate the state E.g. accelerate, brake, turn 33 Identifying Classes We want our class to be a grouping of conceptually- related state and behaviour One popular way to group is using English grammar Noun → Object Verb → Method “The footballer kicked the ball” Representing a Class Graphically (UML) MyFancyClass - age : int + SetAge(age: int) : void Behaviour State “+” means public access “-” means private access 34
- 18. The “has-a” Association College Student1 0...* Arrow going left to right says “a College has zero or more students” Arrow going right to left says “a Student has exactly 1 College” What it means in real terms is that the College class will contain a variable that somehow links to a set of Student objects, and a Student will have a variable that references a College object. Note that we are only linking classes: we don't start drawing arrows to primitive types. The graphical notation used here is part of UML (Unified Modeling Language). UML is basically a standardised set of diagrams that can be used to describe software independently of any programming language used to create it. UML contains many different diagrams (touched on in the Software Design course). Here we just use the UML class diagram such as the one in the slide. 35 Anatomy of an OOP Program (Java) public class MyFancyClass { public int someNumber; public String someText; public void someMethod() { } public static void main(String[] args) { MyFancyClass c = new MyFancyClass(); } } Class name Class state (properties that an object has such as colour or size) Class behaviour (actions an object can do) 'Magic' start point for the program (named main by convention) Create an object of type MyFancyClass in memory and get a reference to it There are a couple of interesting things to note for later discussion. Firstly, the word public is used liberally. Secondly, the main function is declared inside the class itself and as static. Finally there is the notation String[] which represents an array of String objects in Java. You will see arrays in the practicals. 36
- 19. class MyFancyClass { public: int someNumber; public String someText; void someMethod() { } }; void main(int argc, char **argv) { MyFancyClass c; } Anatomy of an OOP Program (C++) Class name Class state Class behaviour 'Magic' start point for the program Create an object of type MyFancyClass This is here just so you can compare. The Java syntax is based on C/C++ so it’s no surprise that there are a lot of similarities. This certainly eases the transition from Java to C++ (or vice-versa), but there are a lot of pitfalls to bear in mind (mostly related to memory management). 37 Let's Build up Java Ourselves We'll start with a simple language that looks like Java and evolve it towards real Java Use the same primitives and Java and the similar syntax. E.g. class MyFancyClass { int someNumber; String someText; void someMethod() { } } void main() { MyFancyClass c = new MyFancyClass(); } A red box on the code box means it is not valid Java. 38
- 20. 2.1 Encapsulation Encapsulation Here we create 3 Student objects when our program runs Problem is obvious: nothing stops us (or anyone using our Student class) from putting in garbage as the age Let's add an access modifier that means nothing outside the class can change the age class Student { int age; } void main() { Student s = new Student(); s.age = 21; Student s2 = new Student(); s2.age=-1; Student s3 = new Student(); s3.age=10055; } Encapsulation Now nothing outside the class can access the age variable directly Have to add a new method to the class that allows age to be set (but only if it is a sensible value) Also needed a GetAge() method so external objects can find out the age. class Student { private int age; boolean SetAge(int a) { if (a>=0 && a<130) { age=a; return true; } return false; } int GetAge() {return age;} } void main() { Student s = new Student(); s.SetAge(21); } 39 Encapsulation We hid the state implementation to the outside world (no one can tell we store the age as an int without seeing the code), but provided mutator methods to... errr, mutate the state This is data encapsulation We define interfaces to our objects without committing long term to a particular implementation Advantages We can change the internal implementation whenever we like so long as we don't change the interface other than to add to it (E.g. we could decide to store the age as a float and add GetAgeFloat()) Encourages us to write clean interfaces for things to interact with our objects Another name for encapsulation is information hiding. The basic idea is that a class should expose a clean interface that allows inter- action, but nothing about its internal state. So the general rule is that all state should start out as private and only have that access relaxed if there is a very, very good reason. Encapsulation helps to minimise coupling between classes. High cou- pling between two class, A and B, implies that a change in A is likely to ripple through to B. In a large software project, you really don’t want a change in one class to mean you have to go and fix up the other 200 classes! So we strive for low coupling. It’s also related to cohesion. A highly cohesive class contains only a set of strongly related functions rather than being a hotch-potch of functionality. We strive for high cohesion. 40
- 21. 2.2 Inheritance Inheritance class Student { public int age; public String name; public int grade; } class Lecturer { public int age; public String name; public int salary; } There is a lot of duplication here Conceptually there is a hierarchy that we're not really representing Both Lecturers and Students are people (no, really). We can view each as a kind of specialisation of a general person They have all the properties of a person But they also have some extra stuff specific to them (I should not have used public variables here, but I did it to keep things simple) Inheritance class Person { public int age; Public String name; } class Student extends Person { public int grade; } class Lecturer extends Person { public int salary; } We create a base class (Person) and add a new notion: classes can inherit properties from it Both state and functionality We say: Person is the superclass of Lecturer and Student Lecturer and Student subclass Person 41 Unfortunately, we tend to use an array of different names for things in an inheritance tree. For A extends B, you might get any of: • A is the superclass of B • A is the parent of B • A is the base class of B • B is a subclass of A • B is the child of A • B derives from A • B inherits from A • B subclasses A Representing Inheritance Graphically Student Lecturer Person Also known as an “is-a” relation As in “Student is-a Person” For reasons I don’t understand, many students seem to confuse “is-a” and “has-a” arrows in their UML class diagrams: please make sure you don’t! 42
- 22. Casting/Conversions As we descend our inheritance tree we specialise by adding more detail ( a salary variable here, a dance() method there) So, in some sense, a Student object has all the information we need to make a Person (and some extra). It turns out to be quite useful to group things by their common ancestry in the inheritance tree We can do that semantically by expressions like: Student s = new Student(); Person p = (Person)s; Person p = new Person(); Student s = (Student)p; This is a widening conversion (we move up the tree, increasing generality: always OK) This would be a narrowing conversion (we try to move down the tree, but it's not allowed here because the real object doesn't have all the info to be a Student)x When we write (Person)s we often say we are casting the Student object to a Person object. It might look a bit odd, but just remember that going down an inheritance tree adds ‘stuff’. Every object has an intrinsic type when we create it and it can’t be cast to anything below that type (because it doesn’t contain the ‘extra’ stuff). 43 2.2.1 Inheritance and State Variables + Inheritance class Person { public String mName; protected int mAge; private double mHeight; } class Student extends Person { public void do_something() { mName=”Bob”; mAge=70; mHeight=1.70; } } Student inherits this as a public variable and so can access it Student inherits this as a protected variable and so can access it Student inherits this as a private variable and so cannot access it Variables + Inheritance: Shadowing class A { public int x; } class B extends A { public int x; } class C extends B { public int x; public void action() { // Ways to set the x in C x = 10; this.x = 10; // Ways to set the x in B super.x = 10; ((B)this).x = 10; // Ways to set the x in A ((A)this.x = 10; } } In memory, you will find three allocated integers for every object of type C. We say that variables in parent classes with the same name as those in child classes are shadowed. Note that the variables are being shadowed: i.e. nothing is being replaced. This is contrast to the behaviour with methods... 44
- 23. There are two new keywords that have appeared here: super and this. The this keyword can be used in any class method1 and provides us with a reference to the current object. In fact, the this keyword is what you need to access anything within a class, but because we’d end up writing this all over the place, it is taken as implicit. So, for example: public class A { private int x; public void go() { this.x=20; } } becomes: public class A { private int x; public void go() { x=20; } } The super keyword gives us access to the direct parent. You’ll meet both in your Java practicals. 1By this I mean it cannot be used outside of a class, such as within a static method: see later for an explanation of these. 45 2.2.2 Inheritance and Functionality Methods + Inheritance: Overriding We might want to require that every Person can dance. But the way a Lecturer dances is not likely to be the same as the way a Student dances... class Person { public void dance() { jiggle_a_bit(); } } class Student extends Person { public void dance() { body_pop(); } } class Lecturer extends Person { } Person defines a 'default' implementation of dance() Lecturer just inherits the default implementation and jiggles Student overrides the default (Subtype) Polymorphism Assuming Person has a default dance() method, what should happen here?? Student s = new Student(); Person p = (Person)s; p.dance(); Option 1 Compiler says “p is of type Person” So p.dance() should do the default dance() action in Person Option 2 Compiler says “The object in memory is really a Student” So p.dance() should run the Student dance() method Polymorphic behaviour 46
- 24. Option 1 would match our expectations from shadowing: the idea that the object contains two dance() methods, and we can choose between them. Option 2 (called polymorphic) effectively has the parent method replaced by the child method. It turns out that option 2 is immensely useful and so is the way Java went. However, not all OO languages choose between the options. C++, for example, allows you to choose: #include <iostream> class A { public: void printNormal() { std::cout << "A" << std::endl; } virtual void printVirtual() { std::cout << "A" << std::endl; } }; class B : public A { public: void printNormal() { std::cout << "B" << std::endl; } virtual void printVirtual() { std::cout << "B" << std::endl; } 47 }; int main(int argc, char ** argv) { B bobj; // Declare an object of type B. // Set up pointers in C++ B *bptr = &bobj; A *aptr = (A*)bptr; // Cast bptr->printNormal(); aptr->printNormal(); bptr->printVirtual(); aptr->printVirtual(); } Which results in: B A B B i.e. you need to add a special tag (virtual) to the methods that you want to behave polymorphically in C++. 48
- 25. The Canonical Example A drawing program that can draw circles, squares, ovals and stars It would presumably keep a list of all the drawing objects Option 1 Keep a list of Circle objects, a list of Square objects,... Iterate over each list drawing each object in turn What has to change if we want to add a new shape? Circle + draw() Square Oval Star + draw() + draw() + draw() The Canonical Example Option 2 Keep a single list of Shape references Figure out what each object really is, narrow the reference and then draw() What if we want to add a new shape? Shape Circle + draw() Square Oval Star + draw() + draw() + draw() For every Shape s in myShapeList If (s is really a Circle) Circle c = (Circle)s; c.draw(); Else if (s is really a Square) Square sq = (Square)s; sq.draw(); Else if... 49 The Canonical Example Option 3 (Polymorphic) Keep a single list of Shape references Let the compiler figure out what to do with each Shape reference What if we want to add a new shape? Shape - x_position: int - y_position: int + draw() Circle + draw() Square Oval Star + draw() + draw() + draw() For every Shape s in myShapeList s.draw(); Implementations Java All methods are polymorphic. Full stop. Python All methods are polymorphic. C++ Only functions marked virtual are polymorphic Polymorphism is an extremely important concept that you need to make sure you understand... What happens when you run a polymorphic function is that the sys- tem must dynamically decide which version to run. As you might 50
- 26. expect, this isn’t instantaneous and so there is an overhead associ- ated with using polymorphic functions. This can be an advantage of languages like C++ (which give you the choice), but more and more languages seem to favour simplicity over power, and this choice is one of the first things to go! 2.3 Abstract Classes Abstract Methods There are times when we have a definite concept but we expect every specialism of it to have a different implementation (like the draw() method in the Shape example). We want to enforce that idea without providing a default method E.g. We want to enforce that all objects that are Persons support a dance() method But we don't now think that there's a default dance() We specify an abstract dance method in the Person class i.e. we don't fill in any implementation (code) at all in Person. class Person { public void dance(); } class Student extends Person { public void dance() { body_pop(); } } class Lecturer extends Person { public void dance() { jiggle_a_bit(); } } An abstract method can be thought of as a contractual obligation: any non-abstract class that inherits from this class will have that method implemented. 51 Abstract Classes Before we could write Person p = new Person() But now p.dance() is undefined Therefore we have implicitly made the class abstract ie. It cannot be directly instantiated to an object Languages require some way to tell them that the class is meant to be abstract and it wasn't a mistake: Note that an abstract class can contain state variables that get inherited as normal Note also that, in Java, we can declare a class as abstract despite not specifying an abstract method in it!! public abstract class Person { public abstract void dance(); } class Person { public: virtual void dance()=0; } Java C++ Depending on who you’re talking to, you’ll find different terminol- ogy for the initial declaration of the abstract function (e.g. the public abstract void dance() bit). Common terms include method pro- totype and method stub. 52
- 27. Representing Abstract Classes Student Lecturer Person + dance() + dance()+ dance() Italics indicate the class or method is abstract 2.4 Interfaces Multiple Inheritance Student Lecturer StudentLecturer What if we have a Lecturer who studies for another degree? If we do as shown, we have a bit of a problem StudentLecturer inherits two different dance() methods So which one should it use if we instruct a StudentLecturer to dance()? The Java designers felt that this kind of problem mostly occurs when you have designed your class hierarchy badly Their solution? You can only extend (inherit) from one class in Java (which may itself inherit from another...) This is a Java oddity (C++ allows multiple class inheritance) 53 Interfaces (Java only) Java has the notion of an interface which is like a class except: There is no state whatsoever All methods are abstract For an interface, there can then be no clashes of methods or variables to worry about, so we can allow multiple inheritance <<interface>> Drivable + turn() + brake() Car <<interface>> Identifiable + getIdentifier() Bicycle + turn() + brake() + turn() + brake() + turn() + brake() + getIdentifier() Interface Drivable { public void turn(); public void brake(); } Interface Identifiable { public void getIdentifier(); } class Bicycle implements Drivable { public void turn() {...} public void brake() {… } } class Car implements Drivable, Identifiable { public void turn() {...} public void brake() {… } Public void getIdentifier() {...} } abstract assumed for interfaces Q6. Explain the differences between a class, an abstract class and an interface. Q7. * Imagine you have two classes: Employee (which represents being an employee) and Ninja (which represents being a Ninja). An Employee has both state and behaviour; a Ninja has only behaviour. You need to represent an employee who is also a ninja. By creating only one interface and only one class(NinjaEmployee), show how you can do this without having to copy method im- plementation code from either of the original classes. 54
- 28. Recap Important OOP concepts you need to understand: Modularity (classes, objects) Data Encapsulation Inheritance Abstraction Polymorphism 2.5 Lifecycle of an Object Constructors You will have noticed that the RHS looks rather like a function call, and that's exactly what it is. It's a method that gets called when the object is constructed, and it goes by the name of a constructor (it's not rocket science). We use constructors to initialise the state of the class in a convenient way. A constructor has the same name as the class A constructor has no return type specified MyObject m = new MyObject(); 55 The return we expect is obviously a reference to the newly created object, which the compiler sorts out for us. Examples public class Person { private String mName; // Constructor public Person(String name) { mName=name; } public static void main(String[] args) { Person p = new Person(“Bob”); } } class Person { private: std::string mName; public: Person(std::string &name) { mName=name; } }; int main(int argc, char ** argv) { Person p (“Bob”); } Java C++ Default Constructor public class Person { private String mName; public static void main(String[] args) { Person p = new Person(); } } If you specify no constructor at all, the Java fills in an empty one for you The default constructor takes no arguments 56
- 29. Beware: As soon as you specify any constructor whatsoever, Java won’t add in a default constructor... Multiple Constructors public class Student { private String mName; private int mScore; public Student(String s) { mName=s; mScore=0; } public Student(String s, int sc) { mName=s; mScore=sc; } public static void main(String[] args) { Student s1 = new Student("Bob"); Student s2 = new Student("Bob",55); } } You can specify as many constructors as you like. Each constructor must have a different signature (argument list) Constructor Chaining When you construct an object of a type with parent classes, we call the constructors of all of the parents in sequence Student s = new Student(); Animal Person Student 1. Call Animal() 2. Call Person() 3. Call Student() 57 What actually happens is that the first line of a constructor always starts with super(), which is a call to the parent constructor (which itself starts with super(), etc.). If it does not, the compiler adds one for you: public class Person { public Person() { } } becomes: public class Person { public Person() { super(); } } This can get messy though: what if the parent does not have a default constructor? In this case, the code won’t compile, and we will have to manually add a call to super ourselves, using the appropriate arguments. E.g. public class Person { protected String mName; public Person(String name) { mName=name; } } 58
- 30. public class Student extends Person { private int mScore; public Student (String name, int score) { super(name); mScore = score; } } Q8. Write a Java program that demonstrates constructor chaining in action. Destructors Most OO languages have a notion of a destructor too Gets run when the object is destroyed Allows us to release any resources (open files, etc) or memory that we might have created especially for the object class FileReader { public: // Constructor FileReader() { f = fopen(“myfile”,”r”); } // Destructor ~FileReader() { fclose(f); } private : FILE *file; } int main(int argc, char ** argv) { // Construct a FileReader Object FileReader *f = new FileReader(); // Use object here ... // Destruct the object delete f; } C++ 59 Cleaning Up A typical program creates lots of objects, not all of which need to stick around all the time Approach 1: Allow the programmer to specify when objects should be deleted from memory Lots of control, but what if they forget to delete an object? Approach 2: Delete the objects automatically (Garbage collection) But how do you know when an object is finished with if the programmer doesn't explicitly tell you it is? Cleaning Up (Java) Java reference counts. i.e. it keeps track of how many references point to a given object. If there are none, the programmer can't access that object ever again so it can be deleted Person object #ref = 2 r2 r1 r1 = null; r2 = null; Person object #ref = 0 r2 r1 Deletable 60
- 31. Cleaning Up (Java) Good: System cleans up after us Bad: It has to keep searching for objects with no references. This requires effort on the part of the CPU so it degrades performance. We can't easily predict when an object will be deleted Cleaning Up (Java) So we can't tell when a destructor would run – so Java doesn't have them!! It does have the notion of a finalizer that gets run when an object is garbage collected BUT there's no guarantee an object will ever get garbage collected in Java... Garbage Collection != Destruction 61 2.6 Class-Level Data Class-Level Data and Functionality Imagine we have a class ShopItem. Every ShopItem has an individual core price to which we need to add VAT Two issues here: 1. If the VAT rate changes, we need to find every ShopItem object and run SetVATRate(...) on it. We could end up with different items having different VAT rates when they shouldn't... 2. It is inefficient. Every time we create a new ShopItem object, we allocate another 32 bits of memory just to store exactly the same number! public class ShopItem { private float price; private float VATRate = 0.175; public float GetSalesPrice() { return price*(1.0+VATRate); } public void SetVATRate(float rate) { VATRate=rate; } } What we have is a piece of information that is class-level not object level Each individual object has the same value at all times We throw in the static keyword: public class ShopItem { private float price; private static float VATRate; .... } Variable created only once and has the lifetime of the program, not the object Class-Level Data and Functionality We now have one place to update More efficient memory usage 17.5 17.5 17.5 17.5 17.5 17.5 Can also make methods static too A static method must be instance independent i.e. it can't rely on member variables in any way Sometimes this is obviously needed. E.g public class Whatever { public static void main(String[] args) { ... } } Must be able to run this function without creating an object of type Whatever (which we would have to do in the main()..!) 62
- 32. Why use other static functions? A static function is like a function in ML – it can depend only on its arguments Easier to debug (not dependent on any state) Self documenting Allows us to group related methods in a Class, but not require us to create an object to run them The compiler can produce more efficient code since no specific object is involved public class Math { public float sqrt(float x) {…} public double sin(float x) {…} public double cos(float x) {…} } … Math mathobject = new Math(); mathobject.sqrt(9.0); ... public class Math { public static float sqrt(float x) {…} public static float sin(float x) {…} public static float cos(float x) {…} } … Math.sqrt(9.0); ... vs 2.7 Examples Vector2D Example We will create a class that represents a 2D vector Vector2D - mX: float - mY : float + Vector2D(x:float, y:float) + GetX() : float + GetY() : float + Add(Vector2D v) : void 63 One of the examples we will develop in class is a representation of a two dimensional vector (x, y). This is obviously a very simple class, but it brings us to an interesting question of mutability. An im- mutable class cannot have its state changed after it has been created (you’re familiar with this from ML, where everything is immutable). A mutable class can be altered somehow (usually as a side effect of calling a method). To make a class immutable: • Make sure all state is private. • Consider making state final (this just tells the compiler that the value never changes once constructed). • Make sure no method tries to change any internal state. Some advantages of immutability: • Simpler to contruct, test and use • Automatically thread safe (don’t worry if this means nothing to you yet). • Allows lazy instatiation of objects. In fact, to quote Effective Java by Joshua Bloch: “Classes should be immutable unless there’s a very good reason to make them mutable... If a class cannot be made immutable, limit its mutability as much as possible.” 64
- 33. Q9. Modify the Vector2D class to handle subtraction and scalar product. Q10. Is the Vector2D class developed in lectures mutable or im- mutable? Q11. Vector2D provides a method to add two vectors. Contrast the following approaches to providing an add method assuming Vector2D is (i) mutable, and (ii) immutable. (a) public void add(Vector2D v) (b) public Vector2D add(Vector2D v) (c) public Vector2D add(Vector2D v1, Vector2D v2) (d) public static Vector2D add(Vector2D v1, Vector2D v2) Q12. A student wishes to create a class for a 3D vec- tor and chooses to derive from the Vector2D class (i.e. public void Vector3D extends Vector2D). The argument is that a 3D vector is a “2D vector with some stuff added”. Explain the conceptual misunderstanding here. Q13. Explain how to write a Java program that is ‘functional’ in nature i.e. no modifiable variables, methods are mathematical functions without side effects. Remember the final keyword. 65 66
- 34. Chapter 3 Other Important Concepts 3.1 Exceptions Error Handling You do a lot on this in your practicals, so we'll just touch on it here The traditional way of handling errors is to return a value that indicates success/failure/error Problems: Could ignore the return value Have to keep checking what the 'codes' are for success, etc. The result can't be returned in the usual way public int divide(double a, double b) { if (b==0) return -1; // error double result = a/b; return 0; // success } … if ( divide(x,y)<0) System.out.println(“Failure!!”); 67 Q14. How would you rewrite divide so that it still returns an error code but also gives us the result of the division (in Java)? Exceptions An exception is an object that can be thrown up by a method when an error occurs and caught by the calling code public double divide(double a, double b) throws DivideByZeroException { if (b==0) throw DivideByZeroException(); else return a/b } … try { double z = divide(x,y); } catch(DivideByZeroException d) { // Handle error here } 68
- 35. Exceptions Advantages: Class name is descriptive (no need to look up codes) Doesn't interrupt the natural flow of the code by requiring constant tests The exception object itself can contain state that gives lots of detail on the error that caused the exception Can't be ignored, only handled There is a lot more we could say about exceptions, but you have the basic tools to understand them and they will be covered in your practical Java course. Just be aware that exceptions are very pow- erful and very popular in most modern programming languages. If you’re struggling to understand them, take a look at: http://java.sun.com/docs/books/tutorial/essential/exceptions/ 69 3.2 Copying or Cloning Java Objects Cloning Sometimes we really do want to copy an object Java calls this cloning We need special support for it Person object (name = “Bob”) r Person object (name = “Bob”) r Person object (name = “Bob”) r_copy Cloning Every class in Java ultimately inherits from the Object class The Object class contains a clone() method So just call this to clone an object, right? Wrong! Surprisingly, the problem is defining what copy actually means 70
- 36. Cloning Sometimes we really do want to copy an object Java calls this cloning We need special support for it Person object (name = “Bob”) r Person object (name = “Bob”) r Person object (name = “Bob”) r_copy Cloning Every class in Java ultimately inherits from the Object class The Object class contains a clone() method So just call this to clone an object, right? Wrong! Surprisingly, the problem is defining what copy actually means 71 Cloning public class MyClass { private float price = 77; } MyClass object (price=77) Clone MyClass object (price=77) MyClass object (price=77) Shallow and Deep Copies public class MyClass { private MyOtherClass moc; } MyClass object Shallow MyOtherClass object MyClass object MyOtherClass object MyClass object MyOtherClass object MyClass object MyClass object MyOtherClass object D eep 72
- 37. Java Cloning So do you want shallow or deep? The default implementation of clone() performs a shallow copy But Java developers were worried that this might not be appropriate: they decided they wanted to know for sure that we'd thought about whether this was appropriate Java has a Cloneable interface If you call clone on anything that doesn't extend this interface, it fails Marker Interfaces If you go and look at what's in the Cloneable interface, you'll find it's empty!! What's going on? Well, the clone() method is already inherited from Object so it doesn't need to specify it This is an example of a Marker Interface A marker interface is an empty interface that is used to label classes This approach is found occasionally in the Java libraries You might also see these marker interfaces referred to as tag inter- faces. They are simply a way to label or tag a class. They can be very 73 useful, but equally they can be a pain (you can’t dynamically tag a class, nor can you prevent a tag being inherited by all subclasses). Q15. An alternative strategy to clone()-ing an object is to provide a copy constructor. This is a constructor that takes the enclosing class as an argument and copies everything manually: public class MyClass { private String mName; private int[] mData; // Copy constructor public MyClass(MyClass toCopy) { this.mName = toCopy.mName; // TODO } ... } (a) Complete the copy constructor. (b) Make MyClass clone()-able (you should do a deep copy). (c) * Why might the Java designers have disliked copy con- structors? [Hint: What happens if you want to copy an object that is being referenced using its parent type?]. (d) * Under what circumstances is a copy constructor a good solution? Q16. Consider the class below. What difficulty is there in providing a deep clone() method for it? public class CloneTest { private final int[] mData = new int[100]; } 74
- 38. 3.3 Distributing Java Classes Distributing Classes So you've written some great classes that might be useful to others. You release the code. What if you've named your class the same as someone else? E.g. There are probably 100s of “Vector” classes out there..! Most languages define some way that you can keep your descriptive class name without getting it confused with others. Java uses packages. A class belongs to a package A nameless 'default' package unless you specify otherwise You're supposed to choose a package name that is unique. Sun decided you should choose your domain name You do have your own domain name, right? ;) Distributing Classes package uk.cam.ac.rkh23; import uk.cam.ac.abc21.*; Class Whatever { … } Class Whatever is part of this package Import all the Classes from some other package You get to do lots more about this in your practicals 75 Access Modifiers Revisited Most Languages: public – everyone can access directly protected – only subclasses can access directly private – nothing can access directly Java adds: package – anything in the same package can access directly 3.4 Java Class Libraries Java Class Library Java the platform contains around 4,000 classes/interfaces Data Structures Networking, Files Graphical User Interfaces Security and Encryption Image Processing Multimedia authoring/playback And more... All neatly(ish) arranged into packages (see API docs) 76
- 39. Remember Java is a platform, not just a programming language. It ships with a huge class library: that is to say that Java itself contains a big set of built-in classes for doing all sorts of useful things like: • Complex data structures and algorithms • I/O (input/output: reading and writing files, etc) • Networking • Graphical interfaces Of course, most programming languages have built-in classes, but Java has a big advantage. Because Java code runs on a virtual ma- chine, the underlying platform is abstracted away. For C++, for example, the compiler ships with a fair few data structures, but things like I/O and graphical interfaces are completely different for each platform (windows, OSX, Linux, whatever). This means you usually end up using lots of third-party libraries to get such ‘extras’ — not so in Java. There is, then, good reason to take a look at the Java class library to see how it is structured. In fact, we’ll see a lot of design patterns used in the library... 77 3.4.1 Collections and Generics Java's Collections Framework <<interface>> Collection <<interface>> Collection <<interface>> Collection <<interface>> Iterable Important chunk of the class library A collection is some sort of grouping of things (objects) Usually when we have some grouping we want to go through it (“iterate over it”) The Collections framework has two main interfaces: Iterable and Collections. They define a set of operations that all classes in the Collections framework support add(Object o), clear(), isEmpty(), etc. The Java Collections framework is a set of interfaces and classes that handles groupings of objects and allows us to implement various algorithms invisibly to the user (you’ll learn about the algorithms themselves next term). 78
- 40. Major Collections Interfaces <<interface>> Set Like a mathematical set in DM 1 A collection of elements with no duplicates Various concrete classes like TreeSet (which keeps the set elements sorted) <<interface>> List An ordered collection of elements that may contain duplicates ArrayList, Vector, LinkedList, etc. <<interface>> Queue An ordered collection of elements that may contain duplicates and supports removal of elements from the head of the queue PriorityQueue, LinkedLIst, etc. A B C A B C B A B C B Major Collections Interfaces <<interface>> Map Like relations in DM 1 Maps key objects to value objects Keys must be unique Values can be duplicated and (sometimes) null. K1 A B B K3 K2 There are other interfaces in the Collections class, and you may want to poke around in the API documentation. In day-to-day program- 79 ming, however, these are likely to be the interfaces you use. Obviously, you can’t use the interfaces directly. So Java includes a few implementations that implement sensible things. Again, you will find them in the API docs, but as an example for Set: TreeSet. A Set that keeps the elements in sorted order so that when you iterate over them, they come out in order. HashSet. A Set that uses a technique called hashing (don’t worry — you’re not meant to know about this yet) that happens to make certain operations (add, remove, etc) very efficient. However, the order the elements iterate over is neither obvious nor constant. Now, don’t worry about what’s going on behind the scenes (that comes in the Algorithms course), just recognise that there are a series of implementations in the class library that you can use, and that each has different properties. Generics The original Collections framework just dealt with collections of Objects Everything in Java “is-a” Object so that way our collections framework will apply to any class we like without any special modification. It gets messy when we get something from our collection though: it is returned as an Object and we have to do a narrowing conversion to make use of it: // Make a TreeSet object TreeSet ts = new TreeSet(); // Add integers to it ts.add(new Integer(3)); // Loop through iterator it = ts.iterator(); while(it.hasNext()) { Object o = it.next(); Integer i = (Integer)o; } 80
- 41. Generics It gets worse when you realise that the add() method doesn't stop us from throwing in random objects: // Make a TreeSet object TreeSet ts = new TreeSet(); // Add integers to it ts.add(new Integer(3)); ts.add(new Person(“Bob”)); // Loop through iterator it = ts.iterator(); while(it.hasNext()) { Object o = it.next(); Integer i = (Integer)o; } Going to fail for the second element! (But it will compile: the error will be at runtime) Generics To help solve this sort of problem, Java introduced Generics in JDK 1.5 Basically, this allows us to tell the compiler what is supposed to go in the Collection So it can generate an error at compile-time, not run- time // Make a TreeSet of Integers TreeSet<Integer> ts = new TreeSet<Integer>(); // Add integers to it ts.add(new Integer(3)); ts.add(new Person(“Bob”)); // Loop through iterator<Integer> it = ts.iterator(); while(it.hasNext()) { Integer i = it.next(); } Won't even compile No need to cast :-) 81 Notation in Java API Set<E> List<E> Queue<E> Map<K,V> Here the letter between the brackets just signifies some class, so you might do: TreeSet<Person> ts = new TreeSet<Person>() 82
- 42. Generics and SubTyping // Object casting Person p = new Person(); Animal o = (Animal) p; // List casting List<Person> plist = new LinkedList<Person>(); List<Animal> alist = (List<Animal>)plist;<<interface>> Collection Person <<interface>> Collection Animal So a list of Persons is a list of Animals, yes? Q17. Java generics can handle this issue using wildcards. Explain how these work. Yes, some research will be required... 83 3.4.2 Comparing Java Objects Comparing Primitives > Greater Than >= Greater than or equal to == Equal to != Not equal to < Less than <= Less than or equal to Clearly compare the value of a primitive But what does (object1==object2) mean?? Same object? Same state (“value”) but different object? The problem is that we deal with references to objects, not objects. So when we compare two things, do we compare the references of the objects they point to? As it turns out, both can be useful so we want to support both. 84
- 43. Option 1: a==b, a!=b These compare the references Person p1 = new Person(“Bob”); Person p2 = new Person(“Bob”); (p1==p2); (p1!=p2); p1==p1; False (references differ) True (references differ) True (references the same) String s = “Hello”; if (s==”Hello”) System.out.println(“Hello”); else System.out.println(“Nope”); Option 2: The equals() Method Object defines an equals() method. By default, this method just does the same as ==. Returns boolean, so can only test equality Override it if you want it to do something different Most (all?) of the core Java classes have properly implemented equals() methods Person p1 = new Person(“Bob”); Person p2 = new Person(“Bob”); (p1==p2); String s1 = “Bob”; String s2 = “Bob”; (s1==s2); False (we haven't overridden the equals() method so it just compares references True (String has equals() overridden) I find this mildly irritating: every class you use will support equals() but you’ll have to check whether or not it has been overridden to do 85 something other than ==. Personally, I only use equals() on objects from core Java classes. Option 3: Comparable<T> Interface int compareTo(T obj); Part of the Collections Framework Returns an integer, r: r<0 This object is less than obj r==0 This object is equal to obj r>0 This object is greater than obj Option 3: Comparable<T> Interface public class Point implements Comparable<Point> { private final int mX; private final int mY; public Point (int, int y) { mX=x; mY=y; } // sort by y, then x public int compareTo(Point p) { if ( mY>p.mY) return 1; else if (mY<p.mY) return -1; else { if (mX>p.mX) return 1; else if (mX<p.mX) return -1; else return 0. } } } // This will be sorted automatically by y, then x Set<Point> list = new TreeSet<Point>(); 86
- 44. Note that the class itself contains the information on how it is to be sorted: we say that it has a natural ordering. Q18. Write a class that represents a 3D point (x,y,z). Give it a natural order such that values are sorted in ascending order by z, then y, then x. Option 4: Comparator<T> interface int compareTo(T obj1, T obj2) Also part of the Collections framework and allows us to specify a particular comparator for a particular job E.g. a Person might have a compareTo() method that sorts by surname. We might wish to create a class AgeComparator that sorts Person objects by age. We could then feed that to a Collections object. At first glance, it may seem that Comparator doesn’t add much over Comparable. However it’s very useful to be able to specify Comparators and apply them dynamically to Collections. If you look in the API, you will find that Collections has a static method sort(List l, Comparator, c). So, imagine we have a class Student that stores the forename, sur- name and exam percentage as a String, String, and a float respec- tively. The natural ordering of the class sorts by surname. We 87 might then supply two Comparator classes: ForenameComparator and ExamScoreComparator that do as you would expect. Then we could write: List list = new SortedList(); // Populate list // List will be sorted naturally ... // Sort list by forename Collections.sort(list, new ForenameComparator()); // Sort list by exam score Collections.sort(list, new ExamScoreComparator()); Q19. (a) Remembering the idea of data encapsulation, write the code for the Student class. (b) Write the code for the two Comparators. (c) Write a program that demonstrates that your code works. (d) Without generics, the Comparator signature has to be compareTo(Object o1, Object o2). Show how you would make this work for your Student class. Assume that it only makes sense to compare two Student objects. 88
- 45. Q20. The user of the class Car below wishes to maintain a collection of Car objects such that they can be iterated over in some specific order. public class Car { private String manufacturer; private int age; } (a) Show how to keep the collection sorted alphabetically by the manufacturer without writing a Comparator. (b) Show how to keep the collection sorted by {manufacturer, age}. i.e. sort first by manufacturer, and sub-sort by age. 89 3.4.3 Java’s I/O Framework Java's I/O framework Support for system input and output (from/to sources such as network, files, etc). <<interface>> Collection Reader Abstract class for reading data from some source <<interface>> Collection InputStreamReader <<interface>> Collection FileReader Concrete Instance that works on an InputStream object Specialisation that allows us to specify a filename, then creates and InputStream for it Speeding it up In general file I/O is sloowwww One trick we can use is that whenever we're asked to read some data in (say one byte) we actually read lots more in (say a kilobyte) and buffer it somewhere on the assumption that it will be wanted eventually and it will just be there in memory, waiting for us. :-) Java supports this in the form of a BufferedReader FileReader f = new FileReader(); BufferedReader br = new BufferedReader(f); Reader InputStreamReader BufferedReader BufferedReader Whenever we call read() on a BufferedReader it looks in its buffer to see whether it has the data already If not it passes the request onto the Reader object We'll come back to this... 90
- 46. The reason file I/O is typically so slow is that hard drives take a long time to deliver information. They contain big, spinning disks and the read head has to move to the correct place to read the data from, then wait until the disc has spun around enough to read all the data it wanted (think old 12 inch record players). Contrast this with memory (in the sense of RAM), where you can just jump wherever you like without consequence and with minimal delay. The BufferedReader simply tries to second guess what will happen next. If you asked for the first 50 bytes of data from a file, chances are you’ll be asking for the next 50 bytes (or whatever) before long, so it loads that data into a buffer (i.e. into RAM) so that if you do turn out to want it, there will be little or no delay. If you don’t use it: oh well, we tried. The key thing is to look at the tree structure: a BufferedReader is-a Reader but also has-a Reader. The idea is that a BufferedReader has all the capabilities of the Reader object that it contains, but also adds some extra functionality. For example, a Reader allows you to read text in byte-by-byte using read(). If you have a string of text, you have to read it in character by character until you get to the terminating character that marks the end of the string. A BufferedReader reads ahead and can read the entire string in one go: it adds a readLine() function to do so. But it still supports the read() functionality if you want to do it the hard way. The really nice thing is that we don’t have to write a BufferedReader for a Reader that we create from scratch. I could create a SerialPortReader that derives from Reader and I could immediately make a BufferedReader for it without having to write any more code. This sort of solution crops up again and again in OOP, and this is one of the “Design Patterns” we’ll talk about later in the course. So 91 you may want to come back to this at the end of the course if you don’t fully ‘get’ it now. 92
- 47. Chapter 4 Design Patterns 4.1 Introduction Coding anything more complicated than a toy program usually bene- fits from forethought. After you’ve coded a few medium-sized pieces of object-oriented software, you’ll start to notice the same general problems coming up over and over. And you’ll start to automati- cally use the same solution each time. We need to make sure that set of default solutions is a good one! In his 1991 PhD thesis, Erich Gamma compared this to the field of architecture, where recurrent problems are tackled by using known good solutions. The follow-on book (Design Patterns: Elements of Reusable Object-Oriented Software, 1994) identified a se- ries of commonly encountered problems in object-oriented software design and 23 solutions that were deemed elegant or good in some way. Each solution is known as a Design Pattern: 93 A Design Pattern is a general reusable solution to a com- monly occurring problem in software design. The modern list of design patterns is ever-expanding and there is no shortage of literature on them. In this course we will be looking at a few key patterns and how they are used. 94
- 48. 4.1.1 So Design Patterns are like coding recipes? No. Creating software by stitching together a series of Design Pat- terns is like painting by numbers — it’s easy and it probably works, but it doesn’t produce a Picasso! Design Patterns are about intel- ligent solutions to a series of generalised problems that you may be able to identify in your software. You might that find they don’t apply to your problem, or that they need adaptation. You simply can’t afford to disengage your brain (sorry!) 4.1.2 Why Bother Studying Them? Design patterns are useful for a number of things, not least: 1. They encourage us to identify the fundamental aims of given pieces of code 2. They save us time and give us confidence that our solution is sensible 3. They demonstrate the power of object-oriented programming 4. They demonstrate that na¨ıve solutions are bad 5. They give us a common vocabulary to describe our code The last one is important: when you work in a team, you quickly realise the value of being able to succinctly describe what your code is trying to do. If you can replace twenty lines of comments1 with a single word, the code becomes more readable and maintainable. Fur- thermore, you can insert the word into the class name itself, making the class self-describing. 1You are commenting your code liberally, aren’t you? 95 4.2 Design Patterns By Example We’re going to develop a simple example to look at a series of design patterns. Our example is a new online venture selling books. We will be interested in the underlying (“back-end”) code — this isn’t going to be a web design course! We start with a very simple setup of classes. For brevity we won’t be annotating the classes with all their members and functions. You’ll need to use common sense to figure out what each element supports. Session. This class holds everything about a current browser ses- sion (originating IP, user, shopping basket, etc). Database. This class wraps around our database, hiding away the query syntax (i.e. the SQL statements or similar). Book. This class holds all the information about a particular book. 96
- 49. 4.3 Supporting Multiple Products Problem: Selling books is not enough. We need to sell CDs and DVDs too. And maybe electronics. Oh, and sports equipment. And... Solution 1: Create a new class for every type of item. It works. We end up duplicating a lot of code (all the products have prices, sizes, stock levels, etc). This is difficult to maintain (imagine changing how the VAT is computed...). Solution 2: Derive from an abstract base class that holds all the common code. “Obvious” object oriented solution 97 If we are smart we would use polymorphism2 to avoid con- stantly checking what type a given Product object is in order to get product-specific behaviour. 4.3.1 Generalisation This isn’t really an ‘official’ pattern, because it’s a rather funda- mental thing to do in object-oriented programming. However, it’s important to understand the power inheritance gives us under these circumstances. 2There are two types of polymorphism. Ad-hoc polymorphism (a.k.a. runtime or dynamic polymorphism) is concerned with object inheritance. It is familiar to you from Java, when the computer automatically figures out which version of an inherited method to run. Parametric polymorphism (a.k.a. static polymor- phism) is where the compiler figures out which version of a type to use before the program runs. You are familiar with this in ML, but you also find it in C++ (templates) and Java (look up generics). 98
- 50. 4.4 The Decorator Pattern Problem: You need to support gift wrapping of products. Solution 1: Add variables to the Product class that describe whether or not the product is to be wrapped and how. It works. In fact, it’s a good solution if all we need is a binary flag for wrapped/not wrapped. As soon as we add different wrapping options and prices for different product types, we quickly clutter up Product. Clutter makes it harder to maintain. Clutter wastes storage space. Solution 2: Add WrappedBook (etc.) as subclasses of Product as shown. Implementing this solution is a shortcut to the Job centre. We are efficient in storage terms (we only allocate space for wrapping information if it is a wrapped entity). We instantly double the number of classes in our code. 99 If we change Book we have to remember to mirror the changes in WrappedBook. If we add a new type we must create a wrapped version. This is bad because we can forget to do so. We can’t convert from a Book to a WrappedBook without copying lots of data. Solution 3: Create a general WrappedProduct class that is both a subclass of Product and references an instance of one of its siblings. Any state or functionality required of a WrappedProduct is ‘passed on’ to its internal sibling, unless it relates to wrapping. We can add new product types and they will be automatically wrappable. We can dynamically convert an established product object into a wrapped product and back again without copying overheads. 100
- 51. We can wrap a wrapped product! We could, in principle, end up with lots of chains of little ob- jects in the system 4.4.1 Generalisation This is the Decorator pattern and it allows us to add functionality to a class dynamically without changing the base class or having to derive a new subclass. Real world example: humans can be ‘deco- rated’ with contact lenses to improve their vision. 101 Note that we can use the pattern to add state (variables) or func- tionality (methods), or both if we want. In the diagram above, I have explicitly allowed for both options by deriving StateDecorator and FunctionDecorator. This is usually unnecessary — in our book seller example we only want to decorate one thing so we might as well just put the code into Decorator. 102
- 52. 4.5 State Pattern Problem: We need to handle a lot of gift options that the customer may switch between at will (different wrapping papers, bows, gift tags, gift boxes, gift bags, ...). Solution 1: Take our WrappedProduct class and add a lot of if/then statements to the function that does the wrapping — let’s call it initiate wrapping(). void initiate_wrapping() { if (wrap.equals(BOX)) { ... } else if (wrap.equals(BOW)) { ... } else if (wrap.equals(BAG)) { ... } else ... } It works. The code is far less readable. Adding a new wrapping option is ugly. Solution 2: We basically have type-dependent behaviour, which is code for “use a class hierarchy”. 103 This is easy to extend. The code is neater and more maintainable. What happens if we need to change the type of the wrapping (from, say, a box to a bag)? We have to construct a new GiftBag and copy across all the information from a GiftBox. Then we have to make sure every reference to the old object now points to the new one. This is hard! Solution 3: Let’s keep our idea of representing states with a class hierarchy, but use a new abstract class as the parent: 104
- 53. Now, every WrappedProduct has-a GiftType. We have retained the advantages of solution 2 but now we can easily change the wrap- ping type in-situ since we know that only the WrappedObject object references the GiftType object. 4.5.1 Generalisation This is the State pattern and it is used to permit an object to change its behaviour at run-time. A real-world example is how your behaviour may change according to your mood. e.g. if you’re angry, you’re more likely to behave aggressively. Q21. Suppose you have an abstract class TeachingStaff with two con- crete subclasses: Lecturer and Professor. Problems arise when a lecturer gets promoted because we cannot convert a Lecturer object to a Professor object. Using the State pattern, show how you would redesign the classes to permit promotion. 105 4.6 Strategy Pattern Problem: Part of the ordering process requires the customer to enter a postcode which is then used to determine the address to post the items to. At the moment the computation of address from postcode is very slow. One of your employees proposes a different way of computing the address that should be more efficient. How can you trial the new algorithm? Solution 1: Let there be a class AddressFinder with a method getAddress(String pcode). We could add lots of if/then/else state- ments to the getAddress() function. String getAddress(String pcode) { if (algorithm==0) { // Use old approach ... } else if (algorithm==1) { // use new approach ... } } The getAddress() function will be huge, making it difficult to read and maintain. Because we must edit AddressFinder to add a new algorithm, we have violated the open/closed principle3 . 3This states that a class should be open to extension but closed to modifi- cation. So we allow classes to be easily extended to incorporate new behavior without modifying existing code. This makes our designs resilient to change but flexible enough to take on new functionality to meet changing requirements. 106
- 54. Solution 2: Make AddressFinder abstract with a single abstract function getAddress(String pcode). Derive a new class for each of our algorithms. We encapsulate each algorithm in a class. Code is clean and readable. More classes kicking around 4.6.1 Generalisation This is the Strategy pattern. It is used when we want to support dif- ferent algorithms that achieve the same goal. Usually the algorithm is fixed when we run the program, and doesn’t change. A real life 107 example would be two consultancy companies given the same brief. They will hopefully produce the same result, but do so in different ways. i.e. they will adopt different strategies. From the (external) customer’s point of view, the result is the same and he is unaware of how it was achieved. One company may achieve the result faster than the other and so would be considered ‘better’. Note that this is essentially the same UML as the State pattern! The intent of each of the two patterns is quite different however: • State is about encapsulating behaviour that is linked to spe- cific internal state within a class. • Different states produce different outputs (externally the class behaves differently). • State assumes that the state will continually change at run- time. • The usage of the State pattern is normally invisible to external classes. i.e. there is no setState(State s) function. • Strategy is about encapsulating behaviour in a class. This behaviour does not depend on internal variables. 108
- 55. • Different concrete Strategys may produce exactly the same out- put, but do so in a different way. For example, we might have a new algorithm to compute the standard deviation of some vari- ables. Both the old algorithm and the new one will produce the same output (hopefully), but one may be faster than the other. The Strategy pattern lets us compare them cleanly. • Strategy in the strict definition usually assumes the class is selected at compile time and not changed during runtime. • The usage of the Strategy pattern is normally visible to exter- nal classes. i.e. there will be a setStrategy(Strategy s) function or it will be set in the constructor. However, the similarities do cause much debate and you will find people who do not differentiate between the two patterns as strongly as I tend to. 109 4.7 Composite Pattern Problem: We want to support entire groups of products. e.g. The Lord of the Rings gift set might contain all the DVDs (plus a free cyanide capsule). Solution 1: Give every Product a group ID (just an int). If someone wants to buy the entire group, we search through all the Products to find those with the same group ID. Does the basic job. What if a product belongs to no groups (which will be the majority case)? Then we are wasting memory and cluttering up the code. What if a product belongs to multiple groups? How many groups should we allow for? Solution 2: Introduce a new class that encapsulates the notion of groups of products: 110
- 56. If you’re still awake, you may be thinking this is a bit like the Deco- rator pattern, except that the new class supports associations with multiple Products (note the * by the arrowhead). Plus the intent is different – we are not adding new functionality but rather supporting the same functionality for groups of Products. Very powerful pattern. Could make it difficult to get a list of all the individual objects in the group, should we want to. 4.7.1 Generalisation This is the Composite pattern and it is used to allow objects and collections of objects to be treated uniformly. Almost any hierarchy uses the Composite pattern. e.g. The CEO asks for a progress 111 report from a manager, who collects progress reports from all those she manages and reports back. Notice the terminology in the general case: we speak of Leafs because we can use the Composite pattern to build a tree structure. Each Composite object will represent a node in the tree, with children that are either Composites or Leafs. This pattern crops up a lot, and we will see it in other contexts later in this course. 112
- 57. Q22. A drawing program has an abstract Shape class. Each Shape object supports a draw() method that draws the relevant shape on the screen (as per the example in lectures). There are a series of concrete subclasses of Shape, including Circle and Rectangle. The drawing program keeps a list of all shapes in a ListShape object. (a) Should draw() be an abstract method? (b) Write Java code for the function in the main application that draws all the shapes on each screen refresh. (c) Show how to use the Composite pattern to allow sets of shapes to be grouped together and treated as a single entity. (d) Which design pattern would you use if you wanted to extend the program to draw frames around some of the shapes? Show how this would work. 113 4.8 Singleton Pattern Problem: Somewhere in our system we will need a database and the ability to talk to it. Let us assume there is a Database class that abstracts the difficult stuff away. We end up with lots of simultane- ous user Sessions, each wanting to access the database. Each one creates its own Database object and connects to the database over the network. The problem is that we end up with a lot of Database objects (wasting memory) and a lot of open network connections (bogging down the database). What we want to do here is to ensure that there is only one Database object ever instantiated and every Session object uses it. Then the Database object can decide how many open connections to have and can queue requests to reduce instantaneous loading on our database (until we buy a half decent one). Solution 1: Use a global variable of type Database that every- thing can access from everywhere. Global variables are less desirable than David Hasselhoff’s great- est hits. Can’t do it in Java anyway... Solution 2: Use a public static variable which everything uses (this is as close to global as we can get in Java). public class System { public static Database database; } 114
- 58. ... public static void main(String[]) { // Always gets the same object Database d = System.database; } This is really just global variables by the back door. Nothing fundamentally prevents us from making multiple Database objects! Solution 3: Create an instance of Database at startup, and pass it as a constructor parameter to every Session we create, storing a reference in a member variable for later use. public class System { public System(Database d) {...} } public class Session { public Session(Database d) {...} } ... public static void main(String[]) { Database d = new Database(); System sys = new System(d); Session sesh = new Session(d); 115 } This solution could work, but it doesn’t enforce that only one Database be instantiated – someone could quite easily create a new Database object and pass it around. We start to clutter up our constructors. It’s not especially intuitive. We can do better. Solution 4: (Singleton) Let’s adapt Solution 2 as follows. We will have a single static instance. However we will access it through a static member function. This function, getInstance() will either cre- ate a new Database object (if it’s the first call) or return a reference to the previously instantiated object. Of course, nothing stops a programmer from ignoring the getInstance() function and just creating a new Database object. So we use a neat trick: we make the constructor private or protected. This means code like new Database() isn’t possible from an arbitrary class. Guarantees that there will be only one instance. 116
- 59. Code to get a Database object is neat and tidy and intuitive to use. e.g. (Database d=Database.getInstance();) Avoids clutter in any of our classes. Must take care in Java. Either use a dedicated package or a private constructor (see below). Must remember to disable clone()-ing! 4.8.1 Generalisation This is the Singleton pattern. It is used to provide a global point of access to a class that should be instantiated only once. There is a caveat with Java. If you choose to make the constructor protected (this would be useful if you wanted a singleton base class for multiple applications of the singleton pattern, and is actually the ‘official’ solution) you have to be careful. Protected members are accessible to the class, any subclasses, and all classes in the same package. Therefore, any class in the same package 117 as your base class will be able to instantiate Singleton objects at will, using the new keyword! Additionally, we don’t want a crafty user to subclass our singleton and implement Cloneable on their version. The examples sheet asks you to address this issue. Q23. One technique to break a Singleton object is to extend it and implement Cloneable. This allows Singleton objects can be cloned, breaking the fundamental goal of a Singleton! Write a Java Singleton class that is not final but still prevents sub- classes from being cloned. 118
- 60. 4.9 Proxy Pattern(s) The Proxy pattern is a very useful set of three patterns: Virtual Proxy, Remote Proxy, and Protection Proxy. All three are based on the same general idea: we can have a place- holder class that has the same interface as another class, but actually acts as a pass through for some reason. 4.9.1 Virtual Proxy Problem: Our Product subclasses will contain a lot of information, much of which won’t be needed since 90% of the products won’t be selected for more detail, just listed as search results. 119 Solution : Here we apply the Proxy pattern by only loading part of the full class into the proxy class (e.g. name and price). If someone requests more details, we go and retrieve them from the database. 4.9.2 Remote Proxy Problem: Our server is getting overloaded. Solution : We want to run a farm of servers and distribute the load across them. Here a particular object resides on server A, say, whilst servers B and C have proxy objects. Whenever the proxy objects get called, they know to contact server A to do the work. i.e. they act as a pass-through. Note that once server B has bothered going to get something via the proxy, it might as well keep the result locally in case it’s used again (saving us another network trip to A). This is caching and we’ll return to it shortly. 4.9.3 Protection Proxy Problem: We want to keep everything as secure as possible. Solution : Create a User class that encapsulates all the informa- tion about a person. Use the Proxy pattern to fill a proxy class with public information. Whenever private information is requested of the proxy, it will only return a result if the user has been authenticated. In this way we avoid having private details in memory unless they have been authorised. 120
- 61. 4.10 Observer Pattern Problem: We use the Remote Proxy pattern to distribute our load. For efficiency, proxy objects are set to cache information that they retrieve from other servers. However, the originals could easily change (perhaps a price is updated or the exchange rate moves). We will end up with different results on different servers, dependent on how old the cache is!! Solution 1: Once a proxy has some data, it keeps polling the authoritative source to see whether there has been a change (c.f. polled I/O). How frequently should we poll? Too quickly and we might as well not have cached at all. Too slow and changes will be slow to propagate. Solution 2: Modify the real object so that the proxy can ‘register’ with it (i.e. tell it of its existence and the data it is interested in). The proxy then provides a callback function that the real object can call when there are any changes. 121 4.10.1 Generalisation This is the Observer pattern, also referred to as Publish-Subscribe when multiple machines are involved. It is useful when changes need to be propagated between objects and we don’t want the objects to be tightly coupled. A real life example is a magazine subscription — you register to receive updates (magazine issues) and don’t have to keep checking whether a new issue has come out yet. You unsub- scribe as soon as you realise that 4GBP for 10 pages of content and 60 pages of advertising isn’t good value. 122
- 62. 123 Q24. Assume there is a machine somewhere on the internet that can supply the latest stock price for a given stock. The software it runs is written in Java and implements the interface: public interface StockReporter { public double getStockPrice(String stockid); } You are given a Java class MyStockReporter that implements this interface for you. When you use a MyStockReporter object, your request is automatically passed onto the real machine. (a) Identify the design pattern in use here (b) Why is this design inefficient? (c) Draw a UML class diagram to explain how the Observer pattern could improve efficiency. Give the changes to the interface that would be required. 124
- 63. 4.11 Abstract Factory Assume that the front-end part of our system (i.e. the web interface) is represented internally by a set of classes that represent various entities on a web page: Let’s assume that there is a render() method that generates some HTML which can then be sent on to web browsers. Problem: Web technology moves fast. We want to use the latest browsers and plugins to get the best effects, but still have older browsers work. e.g. we might have a Flash site, a SilverLight site, a DHTML site, a low-bandwidth HTML site, etc. How do we handle this? Solution 1: Store a variable ID in the InterfaceElement class, or use the State pattern on each of the subclasses. 125 Works. The State pattern is designed for a single object that regularly changes state. Here we have a family of objects in the same state (Flash, HTML, etc.) that we choose between at compile time. Doesn’t stop us from mixing FlashButton with HTMLButton, etc. Solution 2: Create specialisations of InterfaceElement: 126
- 64. Lots of code duplication. Nothing keeps the different TextBoxes in sync as far as the interface goes. A lot of work to add a new interface component type. Doesn’t stop us from mixing FlashButton with HTMLButton, etc. Solution 3: Create specialisations of each InterfaceElement sub- class: Standardised interface to each element type. Still possible to inadvertently mix element types. Solution 4: Apply the Abstract Factory pattern. Here we associate every WebPage with its own ‘factory’ — an object that is 127 there just to make other objects. The factory is specialised to one output type. i.e. a FlashFactory outputs a FlashButton when cre- ate button() is called, whilst a HTMLFactory will return an HTMLButton() from the same method. Standardised interface to each element type. A given WebPage can only generate elements from a single family. Page is completely decoupled from the family so adding a new family of elements is simple. Adding a new element (e.g. SearchBox) is difficult. 128
- 65. Still have to create a lot of classes. 4.11.1 Generalisation This is the Abstract Factory pattern. It is used when a system must be configured with a specific family of products that must be used together. 129 Note that usually there is no need to make more than one factory for a given family, so we can use the Singleton pattern to save memory and time. 130
- 66. Q25. Explain using diagrams how to the Abstract Factory pattern would help in writing an application that must support differ- ent languages (english, french, german, etc). 131 4.12 Summary From the original Design Patterns book: Decorator Attach additional responsibilities to an object dynam- ically. Decorators provide flexible alternatives to subclassing for extending functionality. State Allow and object to alter its behaviour when its internal state changes. Strategy Define a family of algorithms, encapsulate each on, and make them interchangeable. Strategy lets the algorithm vary independently from clients that use it. Composite Compose objects into tree structures to represent part- whole hierarchies. Composite lets clients treat individual ob- jects and compositions of objecta uniformly. Singleton Ensure a class only has one instance, and provide a global point of access to it. Proxy Provide a surrogate or placeholder for another object to con- trol access to it. Observer Define a one-to-many dependency between objects so that when one object changes state, all its dependents are notified and updated accordingly. Abstract Factory Provide an interface for creating families of re- lated or dependent objects without specifying their concrete classes. 132
- 67. 4.12.1 Classifying Patterns Often patterns are classified according to what their intent is or what they achieve. The original book defined three classes: Creational Patterns . Patterns concerned with the creation of objects (e.g. Singleton, Abstract Factory). Structural Patterns . Patterns concerned with the composition of classes or objects (e.g. Composite, Decorator, Proxy). Behavioural Patterns . Patterns concerned with how classes or objects interact and distribute responsibility (e.g. Observer, State, Strategy). 4.12.2 Other Patterns You’ve now met eight Design Patterns. There are plenty more (23 in the original book), but this course will not cover them. What has been presented here should be sufficient to: • Demonstrate that object-oriented programming is powerful. • Provide you with (the beginnings of) a vocabulary to describe your solutions. • Make you look critically at your code and your software archi- tectures. • Entice you to read further to improve your programming. Of course, you probably won’t get it right first time (if there even s a ‘right’). You’ll probably end up refactoring your code as new situations arise. However, if a Design Pattern is appropriate, you should probably use it. 133 4.12.3 Performance Note that all of the examples here have concentrated on structuring code to be more readable and maintainable, and to incorporate con- straints structurally where possible. At no point have we discussed whether the solutions perform better. Many of the solutions exploit runtime polymorphic behaviour, for example, and that carries with it certain overheads. This is another reason why you can’t apply Design Patterns blindly. [This is a good thing since, if it wasn’t true, programming wouldn’t be interesting, and you wouldn’t get jobs!]. 134