Upload
annis-sherman
View
214
Download
0
Embed Size (px)
Citation preview
CSCI 383
Object-Oriented Programming & Design
Lecture 17
Martin van Bommel
Subtype, Subclass and Substitution
The distinction between subtype and subclass is important because of their relationship to substitution
Recall the argument that asserted a child class has the same behavior as the parent, and thus a variable declared as the parent class should in fact be allowed to hold a value generated from a child class
But does this argument always hold true?
Subtypes
What is wanted here is something like the following substitution principle:
If for each object o1 of type S there is an object o2 of type T such that for all programs P defined in terms of T the behavior of P is unchanged when o1 is substituted for o2
then S is a subtype of T. [Liskov 1988]
What is a type?
What do we mean when we use the term (data) type in describing a programming language?
A set of values (the type int, for example, describes -2147483648 to 2147483647)
A set of operations (we can do arithmetic on ints, not on booleans)
A set of properties (if we divide 8 by 5 we are not surprised when the result is 1, and not 1.6)
What about when we consider classes (or interfaces) as a system for defining types?
The Problem of Defining Types
Consider how we might define a Stack ADT
The Problem of Defining Types
Notice how the interface itself says nothing about the LIFO property, which is the key defining feature of a stack. Is the following a stack?
This class definition satisfies the Stack interface but does not satisfy the properties we expect for a stack, since it violates the LIFO property for all but the most recent item placed into the stack
The Definition of Subtype
From this example we see that the properties that are key to the meaning of the Stack are not specified by the interface definition.
It is not that we were lazy; Java (like most other languages) gives us no way to specify the properties that an interface should satisfy
The Definition of Subtype
So now we can better understand the concept of a subtype
A subtype preserves the meaning (purpose, or intent) of the parent
The problem is that meaning is extremely difficult to define. Think about how to define the LIFO characteristics of the stack
The Substitution Paradox
There is a curious paradox that lies at the heart of most strongly typed object-oriented programming languages
Substitution is permitted, based on subclasses. That is, a variable declared as the parent type is allowed to hold a value derived from a child type
Yet from a semantic point of view, substitution only makes sense if the expression value is a subtype of the target variable
If substitution only makes sense for subtypes and not for all subclasses, why do programming languages based the validity of assignment on subclasses?
The Undecidability of the Subtype Relationship
It is trivial to determine if one class is a subclass of another
It is extremely difficult to define meaning (think of the Stack ADT), and even if you can it is almost always impossible to determine if one class preserves the meaning of another
One of the classic corollaries of the halting problem is that there is no procedure that can determine, in general, if two programs have equivalent behavior
There is simply no way that a compiler can ensure that a subclass created by a programmer is indeed a subtype
Is This a Problem?
What does it take to create a subclass that is not a subtype?
The new class must override at least one method from the parent
It must preserve the type signatures But it must violate some important property of the
parent Is this common? Not likely. But it shows you where to look for problem areas
Subtyping in C++
Subtyping in C++ is provided through inheritance Suppose a class Circle is derived from a class Shape. Then, as mentioned before, a Circle object can be implicitly converted into a Shape object Circle circle;Shape shape = circle;
This is called upcasting in C++ because you’re moving up in the class hierarchy
What happens here is that shape loses all information about circle which isn’t contained within the Shape class (e.g., a radius data member).
This is known as object slicing
Static and Dynamic Behavior
We are able to derive new types from existing types
We can upcast from a derived type to its base type
This is useful because it allows us to share some code
However, it really does not give us the polymorphism we want to extract from similar types
Static and Dynamic Behavior
For example, suppose we have the following inheritance hierarchy
class Shape {
...
void draw(void) const;
...
};
class Circle:public Shape{ class Rectangle:public Shape{
... ...
void draw(void) const; void draw(void) const;
... ...
}; };
Static and Dynamic Behavior
Suppose we declared an array of shapes like thisShape* shapeList[2];shapeList[0] = new Circle(...); // Upcast to Shape*shapeList[1] = new Rectangle(...); // Upcast to Shape*
We might want a function to draw all of our shapes in the shape listvoid draw(int numShapes, Shape* shapeList[]){
for(int i=0; i<numShapes; i++)shapeList[i]->draw();
}
Static and Dynamic Behavior
Even though we created a Circle object and a Rectangle object, drawShapes only sees them as generic Shape objects
There are different ways of solving this problem, and we’ll look at a couple of them
Type Fields
One way to solve this polymorphism problem is to manually store the type within every Shape, Circle, or Rectangle object
We could do this by defining an enumerated type, and marking each object in its constructor with its proper type
However, this is very error prone and tends to lead to a lot of switch statements and casting throughout your code
Static vs. Dynamic Type
So far, we haven’t been able to break the bounds of the declaration type, or the static type, of an object
The declared type of an object or an object pointer has been determining which method will be called
The compiler pays no attention to the true type of the object – which is the type we provided when we allocated it with new.
This is also called the object’s dynamic type
Static and Dynamic
Much of the power of object-oriented languages derives from the ability of objects to change their behavior dynamically at run time
In Programming languages Static almost always means fixed or bound at compile
time, and cannot thereafter be changed Dynamic almost always means not fixed or bound until
run time, and therefore can change during the course of execution
Static and Dynamic Typing
In a statically typed programming language, variables have declared types – fixed at compile time(e.g., C++, Java, Pascal)
In a dynamically typed programming language, a variable is just a name. Types are associated with values, not variables. A variable can hold different types during the course of execution (e.g., Smalltalk, Python)
Arguments For and Against
Static and Dynamically typed languages have existed as long as there have been programming languages. Arguments for and against:
Static typing allows better error detection, more work at compile time and hence faster execution time
Dynamic typing allows greater flexibility, easier to write
(for example, no declaration statements) Both arguments have some validity, and hence
both types of languages will continue to exist in the future
The Polymorphic Variable
The addition of object-oriented ideas in statically typed languages adds a new twist. Recall the argument for substitution: an instance of a child class should be allowed to be assigned to a variable of the parent class
Static Class and Dynamic Class
In a statically typed language we say the class of the declaration is the static class for the variable, while the class of the value it currently holds is the dynamic class
Most statically typed OO languages constrain the dynamic class to be a child class of the static class
Animal pet;
Dog fido;
Cat fluffy;
pet = fido; // legal
pet = fluffy; // legal
fido = pet; // not legal!
Importance of Static Class
In a statically typed object-oriented language, the legality of a message is determined at compile time, based on the static class
A message can produce a compile error, even if no run-time error could possibly arise
class Animal {};
class Dog : Animal {
void bark() { std::cout << "woof"; }
};
Animal *pet = new Dog;
pet->bark(); // generates error, Animals don’t bark
Binding Times
The binding time of a function call is when the call is bound to a specific function implementation
There generally are two binding times: static binding and dynamic binding
Static binding (also known as early binding) is when the function choice is based on the characteristics of the variable (i.e., the type of the variable)
Static binding generally takes place at compile time Function overloading and overriding in C++ are
examples of static binding
Binding Times
Dynamic binding (also known as late binding) is when the function choice is based on the dynamic characteristics of the variable (i.e., the type of the data in the variable)
Dynamic binding is deferred until run time In a previous example, if one sends a draw message to
an instance of Rectangle or Circle, one would want the appropriate implementation of draw to be executed
However, if at compile time one knows only that the receiving object is “an instance of a Shape subclass, to be determined at run time”, static binding cannot be used.
In this case, dynamic binding is used Dynamic binding is a powerful but dangerous tool To eliminate the danger of run time errors in C++,
dynamic binding has been limited to virtual functions
Virtual Functions
The way to retain the behavior of the instantiated type is through the use of virtual functions
To declare a method to be a virtual function, you simply use the virtual keyword when declaring the method in the class definition
When you call a function, it will check the dynamic type of the object before choosing which function to call – this process is reification
Virtual Functions
For example, we could declare our Shape, Circle, and Rectangle classes as follows
class Shape {
...
virtual void draw() const;
...
};
class Circle : public Shape { class Rectangle : public Shape {
... ...
virtual void draw() const; virtual void draw() const;
... ...
}; };
Virtual Functions
Now we get the polymorphic behavior we want
Shape* shapeList[maxShapes];
shapeList[0] = new Circle(...);
shapeList[1] = new Rectangle(...);
shapeList[0]->draw(); // Calls Circle::draw()
shapeList[]->draw(); // Calls Rectangle::draw()
Virtual Functions
It is important that you declare the function to be virtual throughout your class hierarchy, or its behavior can be quite unexpected
Virtual and non-virtual method can call each other Overloaded operator functions can be virtual
functions A constructor cannot be a virtual function
because it needs to know the exact type to create However, a destructor can be declared as virtual,
and generally should be virtual functions may not seem significant at
first, but they enable a ton of code reuse when using class hierarchies