Casting a parent class to a child class

[S. Lorétan]

Hi guys,

Sorry for this stupid question, but I don't know why it isn't working.

Here is my (example) code:

namespace Test {
class A {
public string Label1;
}

class B : A {
public string Label2;
}

class Program {
static void Main(string[] args) {
A test = new A();
B test2 = (B)test;
}
}
}

It work fine with the compiler, but when executing, it ends with an
InvalidCastException error. Why?

Thank you.

 

[steve.kaye]

Hi guys,

Sorry for this stupid question, but I don't know why it isn't working.

Here is my (example) code:

namespace Test {
class A {
public string Label1;
}

class B : A {
public string Label2;
}

class Program {
static void Main(string[] args) {
A test = new A();
B test2 = (B)test;
}
}
}

It work fine with the compiler, but when executing, it ends with an
InvalidCastException error. Why?

This is because, in your example, "test" is not of type B. You did "A
test = new A()" so test is an object whose class is A.

Perhaps you wanted to type "A test = new B()" - that would work.

steve.kaye

 

[Truong Hong Thi]

B could be cast to A. A could not be cast to B.
Imagine if you could cast A to B, what if you call test.Label2, which
is not a member of class A.

 

[Chris Dunaway]

namespace Test {
class A {
public string Label1;
}

class B : A {
public string Label2;
}

class Program {
static void Main(string[] args) {
A test = new A();
B test2 = (B)test;
}
}
}

Let's use different names for your classes and see if that clears up
the question. (I added the Cat class to help clarify)

namespace Test {
class Animal {
public string Label1;
}

class Dog : Animal {
public string Label2;
}

class Cat : Animal {
public string Label3;
}

class Program {
static void Main(string[] args) {
Animal test = new Cat(); //Valid because a Cat IS an Animal
Dog test2 = (Dog)test; //Whoa, can't change a Cat into a Dog!!
}
}
}

The reason it won't cast is becuase while all Dogs are Animals, not all
Animals are Dogs! The test variable in my example is declared as
Animal, but is actually a Cat! A rule of thumb would be that you can
cast a descendent type back to one of it's ancestors, but you can't
cast an ancestor type into a specific descendent.

 

[S. Lorétan]

Ok. But I don't understand why it is designed this way.

For me, there is data loss during the cast from a Cat to an Animal. But
there isn't during the (impossible) cast from an Animal to a Cat, because
Cat inherit all the attributes of Animal *plus* some cat-specific ones.

I feel much more natural with this way. I understand that there can't be
both casting (Child to Superclass and Superclass to Child), but the
Child->Superclass cast as C# implements feels strange to my logic.

Let me explain myself.

Animal test = new Cat(); //Valid because a Cat IS an Animal
Dog test2 = (Dog)test; //Whoa, can't change a Cat into a
Dog!!

I don't think it this way. On the line 1, we cast a Cat to an Animal. So,
the Cat object *lose* its specific attributes. Then, on the line 2, we cast
an Animal (with *only* the Animal's attributes) to a Dog, which is valid,
because the Dog class implements every attributes of the Animal class
because it's a child class.

Here is the way I'm thinking about:

Animal an = new Animal(); // We create an Animal with basics attributes
setting to their default value in the constructor...
Dog dog = (Dog)an; // And we add some Dog-specific attributes to this
object while casting it to a Dog.

In this way, there is no data-loss. We can't cast a Dog to an Animal,
because not Animals are Dogs. But we can cast an Animal to a Dog, because
every Dog are Animals!

I don't know if I am understandable, I'm not very comfortable with the
english language. Sorry. By the way, still your point of view is the OOP
way-to-think, it's probably because it is better than mine. I just can't see
where or why. Can you maybe explain me what's wrong in my reasoning?

Thanks and have a nice day!

(e-mail address removed)...

namespace Test {
class A {
public string Label1;
}

class B : A {
public string Label2;
}

class Program {
static void Main(string[] args) {
A test = new A();
B test2 = (B)test;
}
}
}

Let's use different names for your classes and see if that clears up
the question. (I added the Cat class to help clarify)

namespace Test {
class Animal {
public string Label1;
}

class Dog : Animal {
public string Label2;
}

class Cat : Animal {
public string Label3;
}

class Program {
static void Main(string[] args) {
Animal test = new Cat(); //Valid because a Cat IS an Animal
Dog test2 = (Dog)test; //Whoa, can't change a Cat into a
Dog!!
}
}
}

The reason it won't cast is becuase while all Dogs are Animals, not all
Animals are Dogs! The test variable in my example is declared as
Animal, but is actually a Cat! A rule of thumb would be that you can
cast a descendent type back to one of it's ancestors, but you can't
cast an ancestor type into a specific descendent.

 

[Peter Duniho]

Ok. But I don't understand why it is designed this way.

It seems that may be simply because you're looking at class inheritance the
"wrong" way (that is, differently from how it's viewed in the general
paradigm of object-oriented programming).

For me, there is data loss during the cast from a Cat to an Animal.

This is what I mean. You don't lose data when you cast an instance from
a Cat to an Animal. All you do is change your *view* of the data. All of
the original data is still there, but the reference to the data that is of
the type Animal only sees the parts of the data that are common to all
Animals.

When you cast it back to the original type, your view changes and you get to
see all of the data that was originally there, and was always there even
after the instance got cast to an Animal.

I suspect part of the confusion lies in the fact that C# includes some very
robust type-conversion functionality that shares the casting syntax. It's
important to keep in mind the difference, however. When you "cast" an int
to a string (for example), this isn't really a cast...it's a type conversion
that just looks like a cast. This might lead one to believe that that's
what casting is all about: converting one type to another. But in reality,
true casting doesn't change the data at all. It just changes how you treat
the data.

But there isn't during the (impossible) cast from an Animal to a Cat,
because Cat inherit all the attributes of Animal *plus* some cat-specific
ones.

The Cat *class* inherits all of the attributes and behaviors of the Animal
class. But it has its own attributes and behaviors as well (and may modify
the generic attributes and behaviors of the Animal class). These attributes
and behaviors are decided when the *instance* of the class is created.
Casting cannot change this.

The problem with casting something that is an Animal to something that is a
Cat comes when that something wasn't a Cat in the first place. You can't
treat just any old Animal as a Cat. You can only treat an Animal that *is*
a Cat as a Cat.

So, in the example that was given:

* You can treat a Cat as an Animal, because it is. When you do this,
you ignore the things that make the Cat a Cat, paying attention only to
those things that Cats share with all other Animals (including Dogs).

* If you have a "something" that you're treating as an Animal, but which
you know is actually a Cat, you can cast that something to a Cat. In doing
so, what you're saying is "I've got an Animal, I know the Animal is actually
a Cat, and now I'm going to start paying attention to the things that make
it a Cat".

You appear to be feeling as though when you cast something from an Animal to
a child class (such as a Dog), that more-derived reference should be somehow
populated with the Animal parts, along with new default Dog parts. But
that's not how it works. When you cast something, you're casting an
*instance*, not the class itself. The instance has already been determined
as to what type it actually is. The only thing that casting does is change
what your *view* of that instance is. It can't change the data that's
actually in the instance.

So, this means that in the example given, the *instance* created was a Cat.
You can treat the Cat simply as an Animal for a time. After all, all Cats
are Animals. Like all Animals, they need to eat, they need to sleep, they
have characteristics such as mass and gender. These are things common to
all Animals, and if you look at only those characteristics, you don't need
to know that the instance is a Cat. Knowing it's an Animal is sufficient.

But the instance is still a Cat, and if you start treating it like a Dog,
you get into trouble. A Cat won't chew on a bone, it won't fetch your
newspaper, and it won't bark. So your instance, currently being viewed as
an Animal but actually created as a Cat, cannot do any of these things.
Casting it to a Dog doesn't change the fact that it's actually a Cat.

I feel much more natural with this way. I understand that there can't be
both casting (Child to Superclass and Superclass to Child), but the
Child->Superclass cast as C# implements feels strange to my logic.

It's not just that C# implements it this way. All object-oriented languages
handle it this way. As far as "there can't be both casting" goes...I'm not
sure what you mean by that, but in OOP the casting *is* bidirectional. In
the example given, you can cast something that starts out as a Cat to an
Animal, and then cast it back to a Cat. Then back to an Animal. Then back
to a Cat. Over and over, as many times as you like.

At no point during all that casting does any of the actual data change. All
that changes is how the program views the data. Sometimes the code only
cares that the instance is an Animal, and in those cases it will use the
instance cast as an Animal. Other times, the code is dealing with
Cat-specific things and in those cases it will use the instance cast as a
Cat.

Let me explain myself.


I don't think it this way. On the line 1, we cast a Cat to an Animal. So,
the Cat object *lose* its specific attributes.

But it doesn't lose its specific attributes. It retains everything about
itself that was Cat-like. The only thing that is "lost" is that the code
cannot access the Cat-like attributes while using the Animal-typed
reference.

Then, on the line 2, we cast an Animal (with *only* the Animal's
attributes) to a Dog, which is valid, because the Dog class implements
every attributes of the Animal class because it's a child class.

But the *instance* isn't a Dog. It's a Cat. When you try to cast the
instance, originally created as a Cat, to be viewed as a Dog you are making
an error.

Your impression of what a cast does is fundamentally flawed. Casting
something doesn't change what that is. All it does is change how you
perceive that thing. You can try to perceive a Cat as a Dog all day long,
but it will still be a Cat.

Here is the way I'm thinking about:

In this way, there is no data-loss. We can't cast a Dog to an Animal,
because not Animals are Dogs. But we can cast an Animal to a Dog, because
every Dog are Animals!

But you can't do what you're describing. In order to cast an object
instance to a specific type, it needs to already *be* that specific type.
In your example, you've created an instance of a generic Animal. This
instance will never be able to be a Cat *or* a Dog, because it has none of
the characteristics of either.

I don't know if I am understandable, I'm not very comfortable with the
english language.

IMHO, your English is fine. I do believe that you've explained your view of
casting correctly. I'm hopeful that the above helps you understand why that
view is incorrect, and to see what the correct view is.

And for what it's worth, the OOP way of thinking is well-established, but
that doesn't mean it's necessarily intuitive. For someone trying to learn
how OOP works, it's perfectly understandable that there may be some trouble
getting a good grasp on some of the basic concepts. Just keep in mind that
it's your job as a programmer to come to terms with the way the language
works, rather than to invest too much effort believing that the way the
language works is wrong (unless what you want to do is design a computer
language ).

Even if the language behaves in a way that isn't intuitive to you, at end of
the day you still need to comply with the language's requirements in order
to use the language. That will go more smoothly if you can get yourself
into the same frame of mind that the language designers intended, rather
than holding tightly to your own preconceived notion of how things should
be.

Pete

 

[S. Lorétan]

Very, very good explanation. I think I've now understood what a cast is
really, and the fact that the datas of a class instance aren't changed, but
only the view of these datas. Thank you very much, I feel better now that
one of my autodidact's error seems corrected!

Bests regards, and thank you again.

 

[Peter Duniho]

[...]
Bests regards, and thank you again.

You're very welcome. Glad to have been able to help.