MindView Inc.
[ Viewing Hints ] [ Exercise Solutions ] [ Volume 2 ] [ Free Newsletter ]
[ Seminars ] [ Seminars on CD ROM ] [ Consulting ]

Thinking in C++, 2nd ed. Volume 1

©2000 by Bruce Eckel

[ Previous Chapter ] [ Table of Contents ] [ Index ] [ Next Chapter ]

8: Constants

The concept of constant (expressed by the const keyword) was created to allow the programmer to
draw a line between what changes and what doesn’t. This provides safety and control in a C++
programming project.

Since its origin, const has taken on a number of different purposes. In the meantime it trickled back into the C language where its meaning was changed. All this can seem a bit confusing at first, and in this chapter you’ll learn when, why, and how to use the const keyword. At the end there’s a discussion of volatile, which is a near cousin to const (because they both concern change) and has identical syntax.

The first motivation for const seems to have been to eliminate the use of preprocessor #defines for value substitution. It has since been put to use for pointers, function arguments, return types, class objects and member functions. All of these have slightly different but conceptually compatible meanings and will be looked at in separate sections in this chapter.

Value substitution

When programming in C, the preprocessor is liberally used to create macros and to substitute values. Because the preprocessor simply does text replacement and has no concept nor facility for type checking, preprocessor value substitution introduces subtle problems that can be avoided in C++ by using const values.

The typical use of the preprocessor to substitute values for names in C looks like this:

#define BUFSIZE 100

BUFSIZE is a name that only exists during preprocessing, therefore it doesn’t occupy storage and can be placed in a header file to provide a single value for all translation units that use it. It’s very important for code maintenance to use value substitution instead of so-called “magic numbers.” If you use magic numbers in your code, not only does the reader have no idea where the numbers come from or what they represent, but if you decide to change a value, you must perform hand editing, and you have no trail to follow to ensure you don’t miss one of your values (or accidentally change one you shouldn’t).

Most of the time, BUFSIZE will behave like an ordinary variable, but not all the time. In addition, there’s no type information. This can hide bugs that are very difficult to find. C++ uses const to eliminate these problems by bringing value substitution into the domain of the compiler. Now you can say

const int bufsize = 100;

You can use bufsize anyplace where the compiler must know the value at compile time. The compiler can use bufsize to perform constant folding, which means the compiler will reduce a complicated constant expression to a simple one by performing the necessary calculations at compile time. This is especially important in array definitions:

char buf[bufsize];

You can use const for all the built-in types (char, int, float, and double) and their variants (as well as class objects, as you’ll see later in this chapter). Because of subtle bugs that the preprocessor might introduce, you should always use const instead of #define value substitution.

const in header files

To use const instead of #define, you must be able to place const definitions inside header files as you can with #define. This way, you can place the definition for a const in a single place and distribute it to translation units by including the header file. A const in C++ defaults to internal linkage; that is, it is visible only within the file where it is defined and cannot be seen at link time by other translation units. You must always assign a value to a const when you define it, except when you make an explicit declaration using extern:

extern const int bufsize;

Normally, the C++ compiler avoids creating storage for a const, but instead holds the definition in its symbol table. When you use extern with const, however, you force storage to be allocated (this is also true for certain other cases, such as taking the address of a const). Storage must be allocated because extern says “use external linkage,” which means that several translation units must be able to refer to the item, which requires it to have storage.

In the ordinary case, when extern is not part of the definition, no storage is allocated. When the const is used, it is simply folded in at compile time.

The goal of never allocating storage for a const also fails with complicated structures. Whenever the compiler must allocate storage, constant folding is prevented (since there’s no way for the compiler to know for sure what the value of that storage is – if it could know that, it wouldn’t need to allocate the storage).

Because the compiler cannot always avoid allocating storage for a const, const definitions must default to internal linkage, that is, linkage only within that particular translation unit. Otherwise, linker errors would occur with complicated consts because they cause storage to be allocated in multiple cpp files. The linker would then see the same definition in multiple object files, and complain. Because a const defaults to internal linkage, the linker doesn’t try to link those definitions across translation units, and there are no collisions. With built-in types, which are used in the majority of cases involving constant expressions, the compiler can always perform constant folding.

Safety consts

The use of const is not limited to replacing #defines in constant expressions. If you initialize a variable with a value that is produced at runtime and you know it will not change for the lifetime of that variable, it is good programming practice to make it a const so the compiler will give you an error message if you accidentally try to change it. Here’s an example:

//: C08:Safecons.cpp
// Using const for safety
#include <iostream>
using namespace std;

const int i = 100;  // Typical constant
const int j = i + 10; // Value from const expr
long address = (long)&j; // Forces storage
char buf[j + 10]; // Still a const expression

int main() {
  cout << "type a character & CR:";
  const char c = cin.get(); // Can't change
  const char c2 = c + 'a';
  cout << c2;
  // ...
} ///:~

You can see that i is a compile-time const, but j is calculated from i. However, because i is a const, the calculated value for j still comes from a constant expression and is itself a compile-time constant. The very next line requires the address of j and therefore forces the compiler to allocate storage for j. Yet this doesn’t prevent the use of j in the determination of the size of buf because the compiler knows j is const and that the value is valid even if storage was allocated to hold that value at some point in the program.

In main( ), you see a different kind of const in the identifier c because the value cannot be known at compile time. This means storage is required, and the compiler doesn’t attempt to keep anything in its symbol table (the same behavior as in C). The initialization must still happen at the point of definition, and once the initialization occurs, the value cannot be changed. You can see that c2 is calculated from c and also that scoping works for consts as it does for any other type – yet another improvement over the use of #define.

As a matter of practice, if you think a value shouldn’t change, you should make it a const. This not only provides insurance against inadvertent changes, it also allows the compiler to generate more efficient code by eliminating storage and memory reads.

Aggregates

It’s possible to use const for aggregates, but you’re virtually assured that the compiler will not be sophisticated enough to keep an aggregate in its symbol table, so storage will be allocated. In these situations, const means “a piece of storage that cannot be changed.” However, the value cannot be used at compile time because the compiler is not required to know the contents of the storage at compile time. In the following code, you can see the statements that are illegal:

//: C08:Constag.cpp
// Constants and aggregates
const int i[] = { 1, 2, 3, 4 };
//! float f[i[3]]; // Illegal
struct S { int i, j; };
const S s[] = { { 1, 2 }, { 3, 4 } };
//! double d[s[1].j]; // Illegal
int main() {} ///:~

In an array definition, the compiler must be able to generate code that moves the stack pointer to accommodate the array. In both of the illegal definitions above, the compiler complains because it cannot find a constant expression in the array definition.

Differences with C

Constants were introduced in early versions of C++ while the Standard C specification was still being finished. Although the C committee then decided to include const in C, somehow it came to mean for them “an ordinary variable that cannot be changed.” In C, a const always occupies storage and its name is global. The C compiler cannot treat a const as a compile-time constant. In C, if you say

const int bufsize = 100;
char buf[bufsize];

you will get an error, even though it seems like a rational thing to do. Because bufsize occupies storage somewhere, the C compiler cannot know the value at compile time. You can optionally say

const int bufsize;

in C, but not in C++, and the C compiler accepts it as a declaration indicating there is storage allocated elsewhere. Because C defaults to external linkage for consts, this makes sense. C++ defaults to internal linkage for consts so if you want to accomplish the same thing in C++, you must explicitly change the linkage to external using extern:

extern const int bufsize; // Declaration only

This line also works in C.

In C++, a const doesn’t necessarily create storage. In C a const always creates storage. Whether or not storage is reserved for a const in C++ depends on how it is used. In general, if a const is used simply to replace a name with a value (just as you would use a #define), then storage doesn’t have to be created for the const. If no storage is created (this depends on the complexity of the data type and the sophistication of the compiler), the values may be folded into the code for greater efficiency after type checking, not before, as with #define. If, however, you take an address of a const (even unknowingly, by passing it to a function that takes a reference argument) or you define it as extern, then storage is created for the const.

In C++, a const that is outside all functions has file scope (i.e., it is invisible outside the file). That is, it defaults to internal linkage. This is very different from all other identifiers in C++ (and from const in C!) that default to external linkage. Thus, if you declare a const of the same name in two different files and you don’t take the address or define that name as extern, the ideal C++ compiler won’t allocate storage for the const, but simply fold it into the code. Because const has implied file scope, you can put it in C++ header files with no conflicts at link time.

Since a const in C++ defaults to internal linkage, you can’t just define a const in one file and reference it as an extern in another file. To give a const external linkage so it can be referenced from another file, you must explicitly define it as extern, like this:

extern const int x = 1;

Notice that by giving it an initializer and saying it is extern, you force storage to be created for the const (although the compiler still has the option of doing constant folding here). The initialization establishes this as a definition, not a declaration. The declaration:

extern const int x;

in C++ means that the definition exists elsewhere (again, this is not necessarily true in C). You can now see why C++ requires a const definition to have an initializer: the initializer distinguishes a declaration from a definition (in C it’s always a definition, so no initializer is necessary). With an extern const declaration, the compiler cannot do constant folding because it doesn’t know the value.

The C approach to const is not very useful, and if you want to use a named value inside a constant expression (one that must be evaluated at compile time), C almost forces you to use #define in the preprocessor.

Pointers

Pointers can be made const. The compiler will still endeavor to prevent storage allocation and do constant folding when dealing with const pointers, but these features seem less useful in this case. More importantly, the compiler will tell you if you attempt to change a const pointer, which adds a great deal of safety.

When using const with pointers, you have two options: const can be applied to what the pointer is pointing to, or the const can be applied to the address stored in the pointer itself. The syntax for these is a little confusing at first but becomes comfortable with practice.

Pointer to const

The trick with a pointer definition, as with any complicated definition, is to read it starting at the identifier and work your way out. The const specifier binds to the thing it is “closest to.” So if you want to prevent any changes to the element you are pointing to, you write a definition like this:

const int* u;

Starting from the identifier, we read “u is a pointer, which points to a const int.” Here, no initialization is required because you’re saying that u can point to anything (that is, it is not const), but the thing it points to cannot be changed.

Here’s the mildly confusing part. You might think that to make the pointer itself unchangeable, that is, to prevent any change to the address contained inside u, you would simply move the const to the other side of the int like this:

int const* v;

It’s not all that crazy to think that this should read “v is a const pointer to an int.” However, the way it actually reads is “v is an ordinary pointer to an int that happens to be const.” That is, the const has bound itself to the int again, and the effect is the same as the previous definition. The fact that these two definitions are the same is the confusing point; to prevent this confusion on the part of your reader, you should probably stick to the first form.

const pointer

To make the pointer itself a const, you must place the const specifier to the right of the *, like this:

int d = 1;
int* const w = &d;

Now it reads: “w is a pointer, which is const, that points to an int.” Because the pointer itself is now the const, the compiler requires that it be given an initial value that will be unchanged for the life of that pointer. It’s OK, however, to change what that value points to by saying

*w = 2;

You can also make a const pointer to a const object using either of two legal forms:

int d = 1;
const int* const x = &d;  // (1)
int const* const x2 = &d; // (2)

Now neither the pointer nor the object can be changed.

Some people argue that the second form is more consistent because the const is always placed to the right of what it modifies. You’ll have to decide which is clearer for your particular coding style.

Here are the above lines in a compileable file:

//: C08:ConstPointers.cpp
const int* u;
int const* v;
int d = 1;
int* const w = &d;
const int* const x = &d;  // (1)
int const* const x2 = &d; // (2)
int main() {} ///:~

Formatting

This book makes a point of only putting one pointer definition on a line, and initializing each pointer at the point of definition whenever possible. Because of this, the formatting style of “attaching” the ‘*’ to the data type is possible:

int* u = &i;

as if int* were a discrete type unto itself. This makes the code easier to understand, but unfortunately that’s not actually the way things work. The ‘*’ in fact binds to the identifier, not the type. It can be placed anywhere between the type name and the identifier. So you could do this:

int *u = &i, v = 0;

which creates an int* u, as before, and a non-pointer int v. Because readers often find this confusing, it is best to follow the form shown in this book.

Assignment and type checking

C++ is very particular about type checking, and this extends to pointer assignments. You can assign the address of a non-const object to a const pointer because you’re simply promising not to change something that is OK to change. However, you can’t assign the address of a const object to a non-const pointer because then you’re saying you might change the object via the pointer. Of course, you can always use a cast to force such an assignment, but this is bad programming practice because you are then breaking the constness of the object, along with any safety promised by the const. For example:

//: C08:PointerAssignment.cpp
int d = 1;
const int e = 2;
int* u = &d; // OK -- d not const
//! int* v = &e; // Illegal -- e const
int* w = (int*)&e; // Legal but bad practice
int main() {} ///:~

Although C++ helps prevent errors it does not protect you from yourself if you want to break the safety mechanisms.

Character array literals

The place where strict constness is not enforced is with character array literals. You can say

char* cp = "howdy";

and the compiler will accept it without complaint. This is technically an error because a character array literal (“howdy” in this case) is created by the compiler as a constant character array, and the result of the quoted character array is its starting address in memory. Modifying any of the characters in the array is a runtime error, although not all compilers enforce this correctly.

So character array literals are actually constant character arrays. Of course, the compiler lets you get away with treating them as non-const because there’s so much existing C code that relies on this. However, if you try to change the values in a character array literal, the behavior is undefined, although it will probably work on many machines.

If you want to be able to modify the string, put it in an array:

char cp[] = "howdy";

Since compilers often don’t enforce the difference you won’t be reminded to use this latter form and so the point becomes rather subtle.

Function arguments
& return values

The use of const to specify function arguments and return values is another place where the concept of constants can be confusing. If you are passing objects by value, specifying const has no meaning to the client (it means that the passed argument cannot be modified inside the function). If you are returning an object of a user-defined type by value as a const, it means the returned value cannot be modified. If you are passing and returning addresses, const is a promise that the destination of the address will not be changed.

Passing by const value

You can specify that function arguments are const when passing them by value, such as

void f1(const int i) {
  i++; // Illegal -- compile-time error
} 

but what does this mean? You’re making a promise that the original value of the variable will not be changed by the function f1( ). However, because the argument is passed by value, you immediately make a copy of the original variable, so the promise to the client is implicitly kept.

Inside the function, the const takes on meaning: the argument cannot be changed. So it’s really a tool for the creator of the function, and not the caller.

To avoid confusion to the caller, you can make the argument a const inside the function, rather than in the argument list. You could do this with a pointer, but a nicer syntax is achieved with the reference, a subject that will be fully developed in Chapter 11. Briefly, a reference is like a constant pointer that is automatically dereferenced, so it has the effect of being an alias to an object. To create a reference, you use the & in the definition. So the non-confusing function definition looks like this:

void f2(int ic) {
  const int& i = ic;
  i++;  // Illegal -- compile-time error
} 

Again, you’ll get an error message, but this time the constness of the local object is not part of the function signature; it only has meaning to the implementation of the function and therefore it’s hidden from the client.

Returning by const value

A similar truth holds for the return value. If you say that a function’s return value is const:

const int g();

you are promising that the original variable (inside the function frame) will not be modified. And again, because you’re returning it by value, it’s copied so the original value could never be modified via the return value.

At first, this can make the specification of const seem meaningless. You can see the apparent lack of effect of returning consts by value in this example:

//: C08:Constval.cpp
// Returning consts by value
// has no meaning for built-in types

int f3() { return 1; }
const int f4() { return 1; }

int main() {
  const int j = f3(); // Works fine
  int k = f4(); // But this works fine too!
} ///:~

For built-in types, it doesn’t matter whether you return by value as a const, so you should avoid confusing the client programmer and leave off the const when returning a built-in type by value.

Returning by value as a const becomes important when you’re dealing with user-defined types. If a function returns a class object by value as a const, the return value of that function cannot be an lvalue (that is, it cannot be assigned to or otherwise modified). For example:

//: C08:ConstReturnValues.cpp
// Constant return by value
// Result cannot be used as an lvalue

class X {
  int i;
public:
  X(int ii = 0);
  void modify();
};

X::X(int ii) { i = ii; }

void X::modify() { i++; }

X f5() {
  return X();
}

const X f6() {
  return X();
}

void f7(X& x) { // Pass by non-const reference
  x.modify();
}

int main() {
  f5() = X(1); // OK -- non-const return value
  f5().modify(); // OK
// Causes compile-time errors:
//!  f7(f5());
//!  f6() = X(1);
//!  f6().modify();
//!  f7(f6());
} ///:~

f5( ) returns a non-const X object, while f6( ) returns a const X object. Only the non-const return value can be used as an lvalue. Thus, it’s important to use const when returning an object by value if you want to prevent its use as an lvalue.

The reason const has no meaning when you’re returning a built-in type by value is that the compiler already prevents it from being an lvalue (because it’s always a value, and not a variable). Only when you’re returning objects of user-defined types by value does it become an issue.

The function f7( ) takes its argument as a non-const reference (an additional way of handling addresses in C++ and the subject of Chapter 11). This is effectively the same as taking a non-const pointer; it’s just that the syntax is different. The reason this won’t compile in C++ is because of the creation of a temporary.

Temporaries

Sometimes, during the evaluation of an expression, the compiler must create temporary objects. These are objects like any other: they require storage and they must be constructed and destroyed. The difference is that you never see them – the compiler is responsible for deciding that they’re needed and the details of their existence. But there is one thing about temporaries: they’re automatically const. Because you usually won’t be able to get your hands on a temporary object, telling it to do something that will change that temporary is almost certainly a mistake because you won’t be able to use that information. By making all temporaries automatically const, the compiler informs you when you make that mistake.

In the above example, f5( ) returns a non-const X object. But in the expression:

f7(f5());

the compiler must manufacture a temporary object to hold the return value of f5( ) so it can be passed to f7( ). This would be fine if f7( ) took its argument by value; then the temporary would be copied into f7( ) and it wouldn’t matter what happened to the temporary X. However, f7( ) takes its argument by reference, which means in this example takes the address of the temporary X. Since f7( ) doesn’t take its argument by const reference, it has permission to modify the temporary object. But the compiler knows that the temporary will vanish as soon as the expression evaluation is complete, and thus any modifications you make to the temporary X will be lost. By making all temporary objects automatically const, this situation causes a compile-time error so you don’t get caught by what would be a very difficult bug to find.

However, notice the expressions that are legal:

  f5() = X(1);
  f5().modify();

Although these pass muster for the compiler, they are actually problematic. f5( ) returns an X object, and for the compiler to satisfy the above expressions it must create a temporary to hold that return value. So in both expressions the temporary object is being modified, and as soon as the expression is over the temporary is cleaned up. As a result, the modifications are lost so this code is probably a bug – but the compiler doesn’t tell you anything about it. Expressions like these are simple enough for you to detect the problem, but when things get more complex it’s possible for a bug to slip through these cracks.

The way the constness of class objects is preserved is shown later in the chapter.

Passing and returning addresses

If you pass or return an address (either a pointer or a reference), it’s possible for the client programmer to take it and modify the original value. If you make the pointer or reference a const, you prevent this from happening, which may save you some grief. In fact, whenever you’re passing an address into a function, you should make it a const if at all possible. If you don’t, you’re excluding the possibility of using that function with anything that is a const.

The choice of whether to return a pointer or reference to a const depends on what you want to allow your client programmer to do with it. Here’s an example that demonstrates the use of const pointers as function arguments and return values:

//: C08:ConstPointer.cpp
// Constant pointer arg/return

void t(int*) {}

void u(const int* cip) {
//!  *cip = 2; // Illegal -- modifies value
  int i = *cip; // OK -- copies value
//!  int* ip2 = cip; // Illegal: non-const
}

const char* v() {
  // Returns address of static character array:
  return "result of function v()";
}

const int* const w() {
  static int i;
  return &i;
}

int main() {
  int x = 0;
  int* ip = &x;
  const int* cip = &x;
  t(ip);  // OK
//!  t(cip); // Not OK
  u(ip);  // OK
  u(cip); // Also OK
//!  char* cp = v(); // Not OK
  const char* ccp = v(); // OK
//!  int* ip2 = w(); // Not OK
  const int* const ccip = w(); // OK
  const int* cip2 = w(); // OK
//!  *w() = 1; // Not OK
} ///:~

The function t( ) takes an ordinary non-const pointer as an argument, and u( ) takes a const pointer. Inside u( ) you can see that attempting to modify the destination of the const pointer is illegal, but you can of course copy the information out into a non-const variable. The compiler also prevents you from creating a non-const pointer using the address stored inside a const pointer.

The functions v( ) and w( ) test return value semantics. v( ) returns a const char* that is created from a character array literal. This statement actually produces the address of the character array literal, after the compiler creates it and stores it in the static storage area. As mentioned earlier, this character array is technically a constant, which is properly expressed by the return value of v( ).

The return value of w( ) requires that both the pointer and what it points to must be const. As with v( ), the value returned by w( ) is valid after the function returns only because it is static. You never want to return pointers to local stack variables because they will be invalid after the function returns and the stack is cleaned up. (Another common pointer you might return is the address of storage allocated on the heap, which is still valid after the function returns.)

In main( ), the functions are tested with various arguments. You can see that t( ) will accept a non-const pointer argument, but if you try to pass it a pointer to a const, there’s no promise that t( ) will leave the pointer’s destination alone, so the compiler gives you an error message. u( ) takes a const pointer, so it will accept both types of arguments. Thus, a function that takes a const pointer is more general than one that does not.

As expected, the return value of v( ) can be assigned only to a pointer to a const. You would also expect that the compiler refuses to assign the return value of w( ) to a non-const pointer, and accepts a const int* const, but it might be a bit surprising to see that it also accepts a const int*, which is not an exact match to the return type. Once again, because the value (which is the address contained in the pointer) is being copied, the promise that the original variable is untouched is automatically kept. Thus, the second const in const int* const is only meaningful when you try to use it as an lvalue, in which case the compiler prevents you.

Standard argument passing

In C it’s very common to pass by value, and when you want to pass an address your only choice is to use a pointer[43]. However, neither of these approaches is preferred in C++. Instead, your first choice when passing an argument is to pass by reference, and by const reference at that. To the client programmer, the syntax is identical to that of passing by value, so there’s no confusion about pointers – they don’t even have to think about pointers. For the creator of the function, passing an address is virtually always more efficient than passing an entire class object, and if you pass by const reference it means your function will not change the destination of that address, so the effect from the client programmer’s point of view is exactly the same as pass-by-value (only more efficient).

Because of the syntax of references (it looks like pass-by-value to the caller) it’s possible to pass a temporary object to a function that takes a const reference, whereas you can never pass a temporary object to a function that takes a pointer – with a pointer, the address must be explicitly taken. So passing by reference produces a new situation that never occurs in C: a temporary, which is always const, can have its address passed to a function. This is why, to allow temporaries to be passed to functions by reference, the argument must be a const reference. The following example demonstrates this:

//: C08:ConstTemporary.cpp
// Temporaries are const

class X {};

X f() { return X(); } // Return by value

void g1(X&) {} // Pass by non-const reference
void g2(const X&) {} // Pass by const reference

int main() {
  // Error: const temporary created by f():
//!  g1(f());
  // OK: g2 takes a const reference:
  g2(f());
} ///:~


f( ) returns an object of class X by value. That means when you immediately take the return value of f( ) and pass it to another function as in the calls to g1( ) and g2( ), a temporary is created and that temporary is const. Thus, the call in g1( ) is an error because g1( ) doesn’t take a const reference, but the call to g2( ) is OK.

Classes

This section shows the ways you can use const with classes. You may want to create a local const in a class to use inside constant expressions that will be evaluated at compile time. However, the meaning of const is different inside classes, so you must understand the options in order to create const data members of a class.

You can also make an entire object const (and as you’ve just seen, the compiler always makes temporary objects const). But preserving the constness of an object is more complex. The compiler can ensure the constness of a built-in type but it cannot monitor the intricacies of a class. To guarantee the constness of a class object, the const member function is introduced: only a const member function may be called for a const object.

const in classes

One of the places you’d like to use a const for constant expressions is inside classes. The typical example is when you’re creating an array inside a class and you want to use a const instead of a #define to establish the array size and to use in calculations involving the array. The array size is something you’d like to keep hidden inside the class, so if you used a name like size, for example, you could use that name in another class without a clash. The preprocessor treats all #defines as global from the point they are defined, so this will not achieve the desired effect.

You might assume that the logical choice is to place a const inside the class. This doesn’t produce the desired result. Inside a class, const partially reverts to its meaning in C. It allocates storage within each object and represents a value that is initialized once and then cannot change. The use of const inside a class means “This is constant for the lifetime of the object.” However, each different object may contain a different value for that constant.

Thus, when you create an ordinary (non-static) const inside a class, you cannot give it an initial value. This initialization must occur in the constructor, of course, but in a special place in the constructor. Because a const must be initialized at the point it is created, inside the main body of the constructor the const must already be initialized. Otherwise you’re left with the choice of waiting until some point later in the constructor body, which means the const would be un-initialized for a while. Also, there would be nothing to keep you from changing the value of the const at various places in the constructor body.

The constructor initializer list

The special initialization point is called the constructor initializer list, and it was originally developed for use in inheritance (covered in Chapter 14). The constructor initializer list – which, as the name implies, occurs only in the definition of the constructor – is a list of “constructor calls” that occur after the function argument list and a colon, but before the opening brace of the constructor body. This is to remind you that the initialization in the list occurs before any of the main constructor code is executed. This is the place to put all const initializations. The proper form for const inside a class is shown here:

//: C08:ConstInitialization.cpp
// Initializing const in classes
#include <iostream>
using namespace std;

class Fred {
  const int size;
public:
  Fred(int sz);
  void print();
};

Fred::Fred(int sz) : size(sz) {}
void Fred::print() { cout << size << endl; }

int main() {
  Fred a(1), b(2), c(3);
  a.print(), b.print(), c.print();
} ///:~

The form of the constructor initializer list shown above is confusing at first because you’re not used to seeing a built-in type treated as if it has a constructor.

“Constructors” for built-in types

As the language developed and more effort was put into making user-defined types look like built-in types, it became apparent that there were times when it was helpful to make built-in types look like user-defined types. In the constructor initializer list, you can treat a built-in type as if it has a constructor, like this:

//: C08:BuiltInTypeConstructors.cpp
#include <iostream>
using namespace std;

class B {
  int i;
public:
  B(int ii);
  void print();
};

B::B(int ii) : i(ii) {}
void B::print() { cout << i << endl; }

int main() {
  B a(1), b(2);
  float pi(3.14159);
  a.print(); b.print();
  cout << pi << endl;
} ///:~

This is especially critical when initializing const data members because they must be initialized before the function body is entered.

It made sense to extend this “constructor” for built-in types (which simply means assignment) to the general case, which is why the float pi(3.14159) definition works in the above code.

It’s often useful to encapsulate a built-in type inside a class to guarantee initialization with the constructor. For example, here’s an Integer class:

//: C08:EncapsulatingTypes.cpp
#include <iostream>
using namespace std;

class Integer {
  int i;
public:
  Integer(int ii = 0);
  void print();
};

Integer::Integer(int ii) : i(ii) {}
void Integer::print() { cout << i << ' '; }

int main() {
  Integer i[100];
  for(int j = 0; j < 100; j++)
    i[j].print();
} ///:~

The array of Integers in main( ) are all automatically initialized to zero. This initialization isn’t necessarily more costly than a for loop or memset( ). Many compilers easily optimize this to a very fast process.

Compile-time constants in classes

The above use of const is interesting and probably useful in cases, but it does not solve the original problem which is: “how do you make a compile-time constant inside a class?” The answer requires the use of an additional keyword which will not be fully introduced until Chapter 10: static. The static keyword, in this situation, means “there’s only one instance, regardless of how many objects of the class are created,” which is precisely what we need here: a member of a class which is constant, and which cannot change from one object of the class to another. Thus, a static const of a built-in type can be treated as a compile-time constant.

There is one feature of static const when used inside classes which is a bit unusual: you must provide the initializer at the point of definition of the static const. This is something that only occurs with the static const; as much as you might like to use it in other situations it won’t work because all other data members must be initialized in the constructor or in other member functions.

Here’s an example that shows the creation and use of a static const called size inside a class that represents a stack of string pointers[44]:

//: C08:StringStack.cpp
// Using static const to create a 
// compile-time constant inside a class
#include <string>
#include <iostream>
using namespace std;

class StringStack {
  static const int size = 100;
  const string* stack[size];
  int index;
public:
  StringStack();
  void push(const string* s);
  const string* pop();
};

StringStack::StringStack() : index(0) {
  memset(stack, 0, size * sizeof(string*));
}

void StringStack::push(const string* s) {
  if(index < size)
    stack[index++] = s;
}

const string* StringStack::pop() {
  if(index > 0) {
    const string* rv = stack[--index];
    stack[index] = 0;
    return rv;
  }
  return 0;
}

string iceCream[] = {
  "pralines & cream",
  "fudge ripple",
  "jamocha almond fudge",
  "wild mountain blackberry",
  "raspberry sorbet",
  "lemon swirl",
  "rocky road",
  "deep chocolate fudge"
};

const int iCsz = 
  sizeof iceCream / sizeof *iceCream;

int main() {
  StringStack ss;
  for(int i = 0; i < iCsz; i++)
    ss.push(&iceCream[i]);
  const string* cp;
  while((cp = ss.pop()) != 0)
    cout << *cp << endl;
} ///:~

Since size is used to determine the size of the array stack, it is indeed a compile-time constant, but one that is hidden inside the class.

Notice that push( ) takes a const string* as an argument, pop( ) returns a const string*, and StringStack holds const string*. If this were not true, you couldn’t use a StringStack to hold the pointers in iceCream. However, it also prevents you from doing anything that will change the objects contained by StringStack. Of course, not all containers are designed with this restriction.

The “enum hack” in old code

In older versions of C++, static const was not supported inside classes. This meant that const was useless for constant expressions inside classes. However, people still wanted to do this so a typical solution (usually referred to as the “enum hack”) was to use an untagged enum with no instances. An enumeration must have all its values established at compile time, it’s local to the class, and its values are available for constant expressions. Thus, you will commonly see:

//: C08:EnumHack.cpp
#include <iostream>
using namespace std;

class Bunch {
  enum { size = 1000 };
  int i[size];
};

int main() {
  cout << "sizeof(Bunch) = " << sizeof(Bunch) 
       << ", sizeof(i[1000]) = " 
       << sizeof(int[1000]) << endl;
} ///:~

The use of enum here is guaranteed to occupy no storage in the object, and the enumerators are all evaluated at compile time. You can also explicitly establish the values of the enumerators:

enum { one = 1, two = 2, three };

With integral enum types, the compiler will continue counting from the last value, so the enumerator three will get the value 3.

In the StringStack.cpp example above, the line:

static const int size = 100;

would be instead:

enum { size = 100 };

Although you’ll often see the enum technique in legacy code, the static const feature was added to the language to solve just this problem. However, there is no overwhelming reason that you must choose static const over the enum hack, and in this book the enum hack is used because it is supported by more compilers at the time this book was written.

const objects & member functions

Class member functions can be made const. What does this mean? To understand, you must first grasp the concept of const objects.

A const object is defined the same for a user-defined type as a built-in type. For example:

const int i = 1;
const blob b(2);

Here, b is a const object of type blob. Its constructor is called with an argument of two. For the compiler to enforce constness, it must ensure that no data members of the object are changed during the object’s lifetime. It can easily ensure that no public data is modified, but how is it to know which member functions will change the data and which ones are “safe” for a const object?

If you declare a member function const, you tell the compiler the function can be called for a const object. A member function that is not specifically declared const is treated as one that will modify data members in an object, and the compiler will not allow you to call it for a const object.

It doesn’t stop there, however. Just claiming a member function is const doesn’t guarantee it will act that way, so the compiler forces you to reiterate the const specification when defining the function. (The const becomes part of the function signature, so both the compiler and linker check for constness.) Then it enforces constness during the function definition by issuing an error message if you try to change any members of the object or call a non-const member function. Thus, any member function you declare const is guaranteed to behave that way in the definition.

To understand the syntax for declaring const member functions, first notice that preceding the function declaration with const means the return value is const, so that doesn’t produce the desired results. Instead, you must place the const specifier after the argument list. For example,

//: C08:ConstMember.cpp
class X {
  int i;
public:
  X(int ii);
  int f() const;
};

X::X(int ii) : i(ii) {}
int X::f() const { return i; }

int main() {
  X x1(10);
  const X x2(20);
  x1.f();
  x2.f();
} ///:~

Note that the const keyword must be repeated in the definition or the compiler sees it as a different function. Since f( ) is a const member function, if it attempts to change i in any way or to call another member function that is not const, the compiler flags it as an error.

You can see that a const member function is safe to call with both const and non-const objects. Thus, you could think of it as the most general form of a member function (and because of this, it is unfortunate that member functions do not automatically default to const). Any function that doesn’t modify member data should be declared as const, so it can be used with const objects.

Here’s an example that contrasts a const and non-const member function:

//: C08:Quoter.cpp
// Random quote selection
#include <iostream>
#include <cstdlib> // Random number generator
#include <ctime> // To seed random generator
using namespace std;

class Quoter {
  int lastquote;
public:
  Quoter();
  int lastQuote() const;
  const char* quote();
};

Quoter::Quoter(){
  lastquote = -1;
  srand(time(0)); // Seed random number generator
}

int Quoter::lastQuote() const {
  return lastquote;
}

const char* Quoter::quote() {
  static const char* quotes[] = {
    "Are we having fun yet?",
    "Doctors always know best",
    "Is it ... Atomic?",
    "Fear is obscene",
    "There is no scientific evidence "
    "to support the idea "
    "that life is serious",
    "Things that make us happy, make us wise",
  };
  const int qsize = sizeof quotes/sizeof *quotes;
  int qnum = rand() % qsize;
  while(lastquote >= 0 && qnum == lastquote)
    qnum = rand() % qsize;
  return quotes[lastquote = qnum];
}

int main() {
  Quoter q;
  const Quoter cq;
  cq.lastQuote(); // OK
//!  cq.quote(); // Not OK; non const function
  for(int i = 0; i < 20; i++)
    cout << q.quote() << endl;
} ///:~

Neither constructors nor destructors can be const member functions because they virtually always perform some modification on the object during initialization and cleanup. The quote( ) member function also cannot be const because it modifies the data member lastquote (see the return statement). However, lastQuote( ) makes no modifications, and so it can be const and can be safely called for the const object cq.

mutable: bitwise vs. logical const

What if you want to create a const member function, but you’d still like to change some of the data in the object? This is sometimes referred to as the difference between bitwise const and logical const (also sometimes called memberwise const). Bitwise const means that every bit in the object is permanent, so a bit image of the object will never change. Logical const means that, although the entire object is conceptually constant, there may be changes on a member-by-member basis. However, if the compiler is told that an object is const, it will jealously guard that object to ensure bitwise constness. To effect logical constness, there are two ways to change a data member from within a const member function.

The first approach is the historical one and is called casting away constness. It is performed in a rather odd fashion. You take this (the keyword that produces the address of the current object) and cast it to a pointer to an object of the current type. It would seem that this is already such a pointer. However, inside a const member function it’s actually a const pointer, so by casting it to an ordinary pointer, you remove the constness for that operation. Here’s an example:

//: C08:Castaway.cpp
// "Casting away" constness

class Y {
  int i;
public:
  Y();
  void f() const;
};

Y::Y() { i = 0; }

void Y::f() const {
//!  i++; // Error -- const member function
  ((Y*)this)->i++; // OK: cast away const-ness
  // Better: use C++ explicit cast syntax:
  (const_cast<Y*>(this))->i++;
}

int main() {
  const Y yy;
  yy.f(); // Actually changes it!
} ///:~

This approach works and you’ll see it used in legacy code, but it is not the preferred technique. The problem is that this lack of constness is hidden away in a member function definition, and you have no clue from the class interface that the data of the object is actually being modified unless you have access to the source code (and you must suspect that constness is being cast away, and look for the cast). To put everything out in the open, you should use the mutable keyword in the class declaration to specify that a particular data member may be changed inside a const object:

//: C08:Mutable.cpp
// The "mutable" keyword

class Z {
  int i;
  mutable int j;
public:
  Z();
  void f() const;
};

Z::Z() : i(0), j(0) {}

void Z::f() const {
//! i++; // Error -- const member function
    j++; // OK: mutable
}

int main() {
  const Z zz;
  zz.f(); // Actually changes it!
} ///:~

This way, the user of the class can see from the declaration which members are likely to be modified in a const member function.

ROMability

If an object is defined as const, it is a candidate to be placed in read-only memory (ROM), which is often an important consideration in embedded systems programming. Simply making an object const, however, is not enough – the requirements for ROMability are much stricter. Of course, the object must be bitwise-const, rather than logical-const. This is easy to see if logical constness is implemented only through the mutable keyword, but probably not detectable by the compiler if constness is cast away inside a const member function. In addition,

  1. The class or struct must have no user-defined constructors or destructor.
  2. There can be no base classes (covered in Chapter 14) or member objects with user-defined constructors or destructors.

The effect of a write operation on any part of a const object of a ROMable type is undefined. Although a suitably formed object may be placed in ROM, no objects are ever required to be placed in ROM.

volatile

The syntax of volatile is identical to that for const, but volatile means “This data may change outside the knowledge of the compiler.” Somehow, the environment is changing the data (possibly through multitasking, multithreading or interrupts), and volatile tells the compiler not to make any assumptions about that data, especially during optimization.

If the compiler says, “I read this data into a register earlier, and I haven’t touched that register,” normally it wouldn’t need to read the data again. But if the data is volatile, the compiler cannot make such an assumption because the data may have been changed by another process, and it must reread that data rather than optimizing the code to remove what would normally be a redundant read.

You create volatile objects using the same syntax that you use to create const objects. You can also create const volatile objects, which can’t be changed by the client programmer but instead change through some outside agency. Here is an example that might represent a class associated with some piece of communication hardware:

//: C08:Volatile.cpp
// The volatile keyword

class Comm {
  const volatile unsigned char byte;
  volatile unsigned char flag;
  enum { bufsize = 100 };
  unsigned char buf[bufsize];
  int index;
public:
  Comm();
  void isr() volatile;
  char read(int index) const;
};

Comm::Comm() : index(0), byte(0), flag(0) {}

// Only a demo; won't actually work
// as an interrupt service routine:
void Comm::isr() volatile {
  flag = 0;
  buf[index++] = byte;
  // Wrap to beginning of buffer:
  if(index >= bufsize) index = 0;
}

char Comm::read(int index) const {
  if(index < 0 || index >= bufsize)
    return 0;
  return buf[index];
}

int main() {
  volatile Comm Port;
  Port.isr(); // OK
//!  Port.read(0); // Error, read() not volatile
} ///:~

As with const, you can use volatile for data members, member functions, and objects themselves. You can only call volatile member functions for volatile objects.

The reason that isr( ) can’t actually be used as an interrupt service routine is that in a member function, the address of the current object (this) must be secretly passed, and an ISR generally wants no arguments at all. To solve this problem, you can make isr( ) a static member function, a subject covered in Chapter 10.

The syntax of volatile is identical to const, so discussions of the two are often treated together. The two are referred to in combination as the c-v qualifier.

Summary

The const keyword gives you the ability to define objects, function arguments, return values and member functions as constants, and to eliminate the preprocessor for value substitution without losing any preprocessor benefits. All this provides a significant additional form of type checking and safety in your programming. The use of so-called const correctness (the use of const anywhere you possibly can) can be a lifesaver for projects.

Although you can ignore const and continue to use old C coding practices, it’s there to help you. Chapters 11 and on begin using references heavily, and there you’ll see even more about how critical it is to use const with function arguments.

Exercises

Solutions to selected exercises can be found in the electronic document The Thinking in C++ Annotated Solution Guide, available for a small fee from www.BruceEckel.com.

  1. Create three const int values, then add them together to produce a value that determines the size of an array in an array definition. Try to compile the same code in C and see what happens (you can generally force your C++ compiler to run as a C compiler by using a command-line flag).
  2. Prove to yourself that the C and C++ compilers really do treat constants differently. Create a global const and use it in a global constant expression; then compile it under both C and C++.
  3. Create example const definitions for all the built-in types and their variants. Use these in expressions with other consts to make new const definitions. Make sure they compile successfully.
  4. Create a const definition in a header file, include that header file in two .cpp files, then compile those files and link them. You should not get any errors. Now try the same experiment with C.
  5. Create a const whose value is determined at runtime by reading the time when the program starts (you’ll have to use the <ctime> standard header). Later in the program, try to read a second value of the time into your const and see what happens.
  6. Create a const array of char, then try to change one of the chars.
  7. Create an extern const declaration in one file, and put a main( ) in that file that prints the value of the extern const. Provide an extern const definition in a second file, then compile and link the two files together.
  8. Write two pointers to const long using both forms of the declaration. Point one of them to an array of long. Demonstrate that you can increment or decrement the pointer, but you can’t change what it points to.
  9. Write a const pointer to a double, and point it at an array of double. Show that you can change what the pointer points to, but you can’t increment or decrement the pointer.
  10. Write a const pointer to a const object. Show that you can only read the value that the pointer points to, but you can’t change the pointer or what it points to.
  11. Remove the comment on the error-generating line of code in PointerAssignment.cpp to see the error that your compiler generates.
  12. Create a character array literal with a pointer that points to the beginning of the array. Now use the pointer to modify elements in the array. Does your compiler report this as an error? Should it? If it doesn’t, why do you think that is?
  13. Create a function that takes an argument by value as a const; then try to change that argument in the function body.
  14. Create a function that takes a float by value. Inside the function, bind a const float& to the argument, and only use the reference from then on to ensure that the argument is not changed.
  15. Modify ConstReturnValues.cpp removing comments on the error-causing lines one at a time, to see what error messages your compiler generates.
  16. Modify ConstPointer.cpp removing comments on the error-causing lines one at a time, to see what error messages your compiler generates.
  17. Make a new version of ConstPointer.cpp called ConstReference.cpp which demonstrates references instead of pointers (you may need to look forward to Chapter 11).
  18. Modify ConstTemporary.cpp removing the comment on the error-causing line to see what error messages your compiler generates.
  19. Create a class containing both a const and a non-const float. Initialize these using the constructor initializer list.
  20. Create a class called MyString which contains a string and has a constructor that initializes the string, and a print( ) function. Modify StringStack.cpp so that the container holds MyString objects, and main( ) so it prints them.
  21. Create a class containing a const member that you initialize in the constructor initializer list and an untagged enumeration that you use to determine an array size.
  22. In ConstMember.cpp, remove the const specifier on the member function definition, but leave it on the declaration, to see what kind of compiler error message you get.
  23. Create a class with both const and non-const member functions. Create const and non-const objects of this class, and try calling the different types of member functions for the different types of objects.
  24. Create a class with both const and non-const member functions. Try to call a non-const member function from a const member function to see what kind of compiler error message you get.
  25. In Mutable.cpp, remove the comment on the error-causing line to see what sort of error message your compiler produces.
  26. Modify Quoter.cpp by making quote( ) a const member function and lastquote mutable.
  27. Create a class with a volatile data member. Create both volatile and non-volatile member functions that modify the volatile data member, and see what the compiler says. Create both volatile and non-volatile objects of your class and try calling both the volatile and non-volatile member functions to see what is successful and what kind of error messages the compiler produces.
  28. Create a class called bird that can fly( ) and a class rock that can’t. Create a rock object, take its address, and assign that to a void*. Now take the void*, assign it to a bird* (you’ll have to use a cast), and call fly( ) through that pointer. Is it clear why C’s permission to openly assign via a void* (without a cast) is a “hole” in the language, which couldn’t be propagated into C++?


[43] Some folks go as far as saying that everything in C is pass by value, since when you pass a pointer a copy is made (so you’re passing the pointer by value). However precise this might be, I think it actually confuses the issue.

[44] At the time of this writing, not all compilers supported this feature.

[ Previous Chapter ] [ Table of Contents ] [ Index ] [ Next Chapter ]
Last Update:02/01/2000