typedef, const, and type conversions
In this lesson I will cover the following topics
Using typedef to introduce new type names.
Using const.
Type conversion operators (casts).
As you have seen, C programs make use of many data types. There are the built in types like int, long, and double. There are also arrays of those things and pointers to those things. Yet you can also define structure types of all kinds as well. The typedef statement lets you define a new, descriptive name for an existing type. It does not create a new type, only a new type name. Here is an example
typedef int counter;
This statement introduces counter as a synonym for the type int. After the compiler has seen this typedef statement you can then declare variables to be of type counter.
counter x, y, z; // Three counter variables.
counter *p; // A pointer to a counter.
counter counts[128]; // An array of counters.
counter advance(counter in);
// A function taking a counter and returning a counter.
struct example {
char name[64];
counter quantity;
};
// A structure containing a counter as a member.
In short you can use counter anywhere the compiler expects to find a type name. Since I used typedef to make counter synonymous with int, the above declarations are identical to
int x, y, z;
int *p;
int counts[128];
int advance(int in);
struct example {
char name[64];
int quantity;
};
However, typedef does not do a simple text substitution. Consider this example
typedef char *pchar; // "pchar" now means the type "pointer to char". pchar p1, p2; // Declares two variables of type "pointer to char". char *p1, p2; // Declares p1 to be a pointer and p2 to be a plain character.
The compiler understands the typedef name as the name of a type and handles declarations accordingly. It does not just substitute one string for another the way the preprocessor does. Thus using typedef is much preferred over using #define to get a similar effect.
#define PCHAR char * PCHAR p1, p2; // Only declares p1 to be a pointer.
Let's look at a couple of more examples using the pchar typedef name I defined above.
pchar *p; // p is a pointer to a pointer to character. pchar array[10]; // array is an array of 10 pointers to char.
Now check this out
typedef pchar pArray[10];
This introduces the name pArray as the name of the type "Array of 10 pointers to character". Then I can write
pArray array; // array is an array of 10 pointers to char.
Notice how the size of the array is part of its type and thus remembered by typedef. Notice also how I was able to use one typedef name to define another. And why not? Once a typedef name is defined, the compiler treats it like any other type.
At this point you might be wondering how you can know the format of a typedef statement. There is a very easy rule for it. First write down the declaration of a variable of the desired type. Then put the word "typedef" in front of that declaration. The name you used for the variable becomes the new type name. Here's an example: suppose I wanted to create a type name buffer that represents an array of 128 characters. First I declare such an array, using buffer as the name of the array.
char buffer[128];
Now I just put typedef in front of that.
typedef char buffer[128];
Ta da! The name buffer is now a name for the type "array of 128 characters". I can now do this
buffer line1, line2;
to declare two such arrays.
It turns out that in C you never need to use typedef. There is no context where you can't just write out the type, no matter how complex it might be. So why have typedef? Here are some reasons:
You can create type names that reflect what you are trying to do and then declare variables with those names as a way of documenting your program and making it easier to read. This allows you to cut down on comments! For example
typedef int loop_index; ... loop_index i, j, k;
It's a safe bet that i, j, and k are going to be used as loop indicies.
You can modify the program in some cases by just changing the typedef statement. Suppose, for example, that you decide your loops need long integer indicies rather than plain integer indicies. If you defined the typedef above (and used it consistently) then you can just change it to
typedef long loop_index;
and recompile. In this respect the typedef statement is similar to using preprocessor macros. However, unlike the preprocessor the compiler understands typedef and can handle it more consistently and naturally, as I explained above.
Types can become very complicated in C. Although we have not explored the full depth of C's type system in this course, I can give you an example to illustrate. Here is the declaration of the signal function from the standard library:
void (*signal(int, void (*)(int)))(int);
This declares signal to be of the type "function taking int and (pointer to function taking int and return void) returning (pointer to function taking int and returning void)". If that doesn't make sense to you, don't worry. It baffles many practicing C programmers too. And this is not a made-up example. It is a real function that is part of the standard library.
Declarations like this can be made much more readable by using typedef names.
typedef void signal_handler(int);
The name signal_handler is now a name for the type "function taking int and returning void".
signal_handler *signal(int, signal_handler *);
This declaration of signal is the same as the really nasty one I showed above. However, it makes a lot more sense. Although you never need to use typedef names in C for this purpose (in C++ there are situations where things are so nasty that you have to use typedef to avoid syntax errors) doing so can greatly improve readability.
Actually one of the most popular use of typedef names is with structures. You probably noticed that constantly putting "struct" in front of a structure name is rather tedious.
struct example f(struct example *p)
{
struct example temp;
...
Here I'm trying to write a function named f that takes a pointer to a struct example as a parameter, returns a struct example and uses a local variable of type struct example. This can be cleaned up quite a bit with an application of typedef.
typedef struct example example;
This looks rather strange because it seems to be using the same name twice. Isn't that a conflict? Actually it isn't because structure "tags" and type names are in different "namespaces" in the compiler. The compiler won't confuse struct example with a variable named example or with a type named example. Now my function becomes
example f(example *p)
{
example temp;
...
Much nicer.
C allows you to declare things to be constants using the const keyword. This feature was borrowed from C++ when C was standardized in the late 1980s. At that time C++ was quite young, but it had already enhanced C in various ways. When the ANSI C standard was created, the standards group borrowed some ideas from C++ and made them part of standard C. The const feature was one of them.
Actually const is very necessary in C++ because it interacts with some of C++'s more elaborate features (that were not put into C) in very important ways. In C the const feature is not really necessary, but it is nice to use anyway. Here's how it looks.
const int MAX = 10;
This declares MAX to be a constant integer. Because it is a constant, its value can't be changed. Thus whenever you declare a const, you must initialize it. The following is illegal.
const int MAX; MAX = 10; // Error! Can't modify a const.
Actually the compiler will object to the declaration of the constant without an initializer. Declaring constants without initializing them just doesn't make sense.
Notice how I'm using all uppercase letters for constants. That is fairly traditional; I suggest that you do the same.
You can use const to create nice names for raw numbers in much the same way as you can use object-like macros in the preprocessor. In fact, before const existed in C the preprocessor was the only way to create such names. After const was added to C, many C programmers continued to use the preprocessor since that was what they were used to. However, the compiler knows about const and thus the problems with scope are not an issue.
const int MAX = 10;
...
void f(void)
{
double MAX;
// Not a problem. This declaration hides the global const.
...
For this reason, using const is probably better than using object-like macros. (This is especially true in C++ where const is handled a bit differently and more efficiently).
By the way... in this course I have consistently used the word "variable" to describe data objects that you declare. I do this because many of you may have heard the term "variable" before in earlier courses. In addition, many texts use the word "variable". However, the reality is that "variable" is an inaccurate term. Take a look at this declaration
const int MAX = 10;
Would you say that MAX is a "constant variable?" That doesn't make any sense! The fact is that some "variables" don't vary. The name is rather misleading. A more accurate term is "object". In the example above I would say that MAX is a "constant integer object". Some objects are modifiable and some objects aren't. Yet they are all objects just the same.
Sometimes when you start talking about objects people jump to the conclusion that you are talking about objects in an "object-oriented" language. It is true that the objects manipulated by such a language are objects. But the integers, characters, and doubles manipulated by a C program are objects too. In fact, the C standard uses the term "object" consistently throughout its text—and for good reason. That term is accurate and clear.
Some of you may end up taking Object Oriented Programming, the follow-up course to this one, with me. If so, you will see that in that course I will shift my language a bit and drop the word "variable" from my vocabulary. I will talk only about objects because that is more accurate and because that way I don't have to talk about constant variables.
The const "qualifier" (as it is called) interacts with pointers in a very rich way. I won't say too much about it here. This is a topic that we will cover more in Object Oriented Programming. However, to get a feeling for the issues, take a look at the following declarations.
const char *p; // p points to a constant character. The pointer is NOT const. char * const p = "Hello, World!"; // p is a constant pointer that points at normal characters. const char * const p = "Hello, World!"; // p is a constant pointer that points at constant characters.
The most common and interesting situation is the first one: a non-constant pointer to constant things. Now take a look at this
char *p1; const char *p2; ... p2 = p1; // Fine p1 = p2; // Error
I can't put a pointer to const into a plain pointer because if I were allowed to do that I might be able to later modify constant objects by accessing them via the plain pointer.
*p1 = 'X'; // Fine because p1 points at non-constant characters.
However, I can put a plain pointer into a pointer to const. By doing so I sacrifice the right to modify the non-constant objects via that new pointer, but that's fine. It won't cause an undefined error.
These rules are very significant and important, but you'll have to wait until Object Oriented Programming for all the details!
Normally it is highly desirable to avoid mixing types in expressions. For example
x = 2 * y / (z - 1);
If x, y, and z are all integers this expression is easy to interpret. If, for example, y was a double then the behavior is less clear. What happens in that case is this
The compiler does z - 1 using integer subtraction since z is an integer and so is 1.
The compiler upgrades 2 to 2.0 (that is: converts it to double) and multiplies 2.0 and y using floating point multiplication in double precision.
The compiler upgrades the result of z - 1 to double and computes the final result using floating point division in double precision.
The compiler converts the resulting double back into an integer and stuffs it into x.
The actual order might vary from what I described above. For example the compiler might compute 2.0 * y before z - 1. But my point is that there are several automatic type conversions going on in the expression. If those conversions occur at unexpected time or yield unexpected results you might get a very unexpected answer. To avoid these problems, you should not mix types in expressions unless absolutely necessary. It can be argued that if you have an expression in your program with mixed types, it means that you declared one of the variables inappropriately or that your program is designed poorly. This is not always true, but it often is.
There are two types of conversions that the compiler is likely to do automatically. The first is "widening conversions" where the target type can hold all the values of the source type. These conversions are generally safe and hold few opportunities for surprises. For example, the conversion of int to long is a widening conversion.
The other type is called a "narrowing conversion". It occurs where the target type can't hold all the values of the source type. Thus it might be impossible to put the source value into the target variable. Many compilers produce warning messages when you do a narrowing conversion. If you see such a warning (something like "Possible lose of significance") you should look into it. Perhaps you should redeclare the target object so that its type is wider. Or perhaps in your situation you can argue that at that particular point in the program the value in the source object will fit into the target object despite the different types. In any case, narrowing conversions are worrisome and you should do your best to avoid them. In my example above I am doing a narrowing conversion when I stuff the double result of the expression into the integer x.
The compiler uses rather elaborate rules for defining what conversions it will apply automatically and when. I won't bore you with all the details. You really don't need to know them. The rules are designed to offer the fewest surprises with the most natural usage. However, there are times when you have to force the compiler to do conversions it would not normally do. Even when the compiler does a conversion automatically that you approve of, you can make your intentions explicit in your program as a way of documenting your program. (It will also turn off any warning messages the compiler might be producing). To do these things, you need to use a type conversion operator. In C these operators have traditionally been called "casts".
Here's how it looks:
int x; long y; ... x = y; // Compiler produces warning. Dangerous conversion!
Some compilers will warn you about the assignment because it is not always true that a long will fit into an int. For example, on machines using the Alpha processor longs are 64 bits and ints are 32 bits. Thus there is a real chance that the value being put into x above is corrupted. Assuming that you really intend to do this, you can turn off the warning like this
x = (int)y;
Putting a type name in parentheses creates a cast operator that can convert its operand into the named type. In this case (int) converts y into an integer. The compiler would have done this conversion automatically anyway, but since you are making it explicit you are effectively saying, "I know this is a conversion, but it's okay". You are telling the compiler to not bother warning you.
You can also use type casts to do conversions the compiler would otherwise not allow. For example
char buffer[128]; int address; address = buffer;
Here I'm trying to put a pointer into an integer. This is usually bad. But if I really mean to do this I can force the issue
address = (int)buffer;
It now becomes up to me to be sure integers on this machine can properly hold addresses. If they can, fine. If they can't my program will malfunction. However, I may very well know that addresses and integers are compatible (as they are on many systems). In that case, this approach allows me to do some very interesting operations. Keep in mind, however, that a program that engages in otherwise illegal conversions may not run properly on machines other than the one on which it was developed.
Suppose I want to store an integer into my array of characters. Suppose also that on your machine a character is one byte and an integer is four bytes. That's fine. I want my integer to overlay the first four characters in the array. This doesn't work
buffer[0] = 123456;
This just hacks the integer down to character size (tossing away significant bits) and stuffs the result into a single array element. I might get a warning about this, but in any case it doesn't have the effect I'm after.
*(int *)buffer = 123456;
This is more like it. Here I'm converting buffer (which has type "pointer to character") to the type "pointer to integer". Then I use the indirection operator to access the integer that pointer is pointing at. I store 123456 into that integer. Of course buffer is not really pointing at an integer, but because of my use of a cast I can force the compiler to act as if it was.
Code like this is very unusual and very system dependent. Often programmers use tricks with casts to cover up a bad design or other problems. However, there are certainly times when casts are necessary---particularly when you are trying to interact with low level hardware devices. C is know as a hacker's language and it's ability to do odd things with casts is one of the reasons.
You can introduce a new, descriptive name for an existing type with typedef. To do this, declare a variable of the desired type and put the word "typedef" in front of the declaration. The name you used for the variable becomes the name of the new type.
Introducing new type names helps to document the program, make the program easier to modify, and simplifies complicated declarations.
You can put the word "const" in front of a declaration to declare objects that are constants. Such objects must be initialized when they are declared since you can't give them a value afterwards. Typically constants are given names in all uppercase letters.
The compiler often converts variables of one type into another type in order to make sense out of the expressions you write. Some of these conversions are safe and some are unsafe. You can control which conversions occur and when using cast operators. A type cast operator is the name of a type enclosed in parentheses. It coverts its operand to the type specified. For example
x = (int)y;
Causes y to be converted to an integer and the result put into x. Casts can be used to make the automatic conversions explicit or to do conversions that the compiler would not normally allow. Keep in mind, however, that if you use a cast to force a normally illegal conversion your program may do odd things or may only work on certain machines.