No, you can't add pointers together. If you live at 1332 Lakeview Drive, and your neighbor lives at 1364 Lakeview, what's 1332+1364? It's a number, but it doesn't mean anything. If you try to perform this type of calculation with pointers in a C program, your compiler will complain.

The only time the addition of pointers might come up is if you try to add a pointer and the difference of two pointers:

p = p + p2 - p1;

which is the same thing as this:

p = (p + p2) - p1.

Here's a correct way of saying this:

p = p + ( p2 - p1 );

Or even better in this case would be this example:

p += p2 - p1;

The hardest part about using a pointer-to-function is declaring it. Consider an example. You want to create a pointer, pf, that points to the strcmp() function. The strcmp() function is declared in this way:

int strcmp( const char *, const char * )

To set up pf to point to the strcmp() function, you want a declaration that looks just like the strcmp() function's declaration, but that has *pf rather than strcmp:

int (*pf)( const char *, const char * );

Notice that you need to put parentheses around *pf. If you don't include parentheses, as in

int *pf( const char *, const char * ); /* wrong */

you'll get the same thing as this:

(int *) pf( const char *, const char * ); /* wrong */

That is, you'll have a declaration of a function that returns int*.

After you've gotten the declaration of pf, you can #include <string.h> and assign the address of strcmp() to pf:

pf = strcmp;

or

pf = & strcmp; /* redundant & */

You don't need to go indirect on pf to call it:

if ( pf( str1, str2 ) > 0 ) /* ... */

Pointers to functions are interesting when you pass them to other functions. A function that takes function pointers says, in effect, "Part of what I do can be customized. Give me a pointer to a function, and I'll call it when that part of the job needs to be done. That function can do its part for me." This is known as a "callback." It's used a lot in graphical user interface libraries, in which the style of a display is built into the library but the contents of the display are part of the application.

As a simpler example, say you have an array of character pointers (char*s), and you want to sort it by the value of the strings the character pointers point to. The standard qsort() function uses function pointers to perform that task. qsort() takes four arguments

1. a pointer to the beginning of the array,

2. the number of elements in the array,

3. the size of each array element, and

4. a comparison function.

and returns an int.

The comparison function takes two arguments, each a pointer to an element. The function returns 0 if the pointed-to elements compare equal, some negative value if the first element is less than the second, and some positive value if the first element is greater than the second. A comparison function for integers might look like this:

int icmp( const int *p1, const int *p2 )
{
     return *p1 - *p2;
}

The sorting algorithm is part of qsort(). So is the exchange algorithm; it just copies bytes, possibly by calling memcpy() or memmove(). qsort() doesn't know what it's sorting, so it can't know how to compare them. That part is provided by the function pointer.

You can't use strcmp() as the comparison function for this example, for two reasons. The first reason is that strcmp()'s type is wrong; more on that a little later. The second reason is that it won't work. strcmp() takes two pointers to char and treats them as the first characters of two strings. The example deals with an array of character pointers (char*s), so the comparison function must take two pointers to character pointers (char*s). In this case, the following code might be an example of a good comparison function:

int strpcmp( const void *p1, const void *p2 )
{
     char * const *sp1 = (char * const *) p1;
     char * const *sp2 = (char * const *) p2;
     return strcmp( *sp1, *sp2 );
}

The call to qsort() might look something like this:

qsort( array, numElements, sizeof( char * ), pf2 );

qsort() will call strpcmp() every time it needs to compare two character pointers (char*s).

Why can't strcmp() be passed to qsort(), and why were the arguments of strpcmp() what they were?

A function pointer's type depends on the return type of the pointed-to function, as well as the number and types of all its arguments. qsort() expects a function that takes two constant void pointers:

void qsort( void *base,
            size_t numElements,
            size_t sizeOfElement,
            int (*compFunct)( const void *, const void *) );

Because qsort() doesn't really know what it's sorting, it uses a void pointer in its argument (base) and in the arguments to the comparison function. qsort()'s void* argument is easy; any pointer can be converted to a void* without even needing a cast. The function pointer is harder.

For an array of character arrays, strcmp() would have the right algorithm but the wrong argument types. The simplest, safest way to handle this situation is to pass a function that takes the right argument types for qsort() and then casts them to the right argument types. That's what strpcmp() does.

If you have a function that takes a char*, and you know that a char* and a void* are the same in every environment your program might ever work in, you might cast the function pointer, rather than the pointed- to function's arguments, in this way:

char     table[ NUM_ELEMENTS ][ ELEMENT_SIZE ];
/* ... */
/* passing strcmp() to qsort for array of array of char */
qsort( table, NUM_ELEMENTS, ELEMENT_SIZE,
  ( int (*)( const void *, const void * ) ) strcmp );

Casting the arguments and casting the function pointer both can be error prone. In practice, casting the function pointer is more dangerous.

The basic problem here is using void* when you have a pointer to an unknown type. C++ programs sometime solve this problem with templates.

No. In an array declaration, the size must be known at compile time. You can't specify a size that's known only at runtime. For example, if i is a variable, you can't write code like this:

char array[i]; /* not valid C */

Some languages provide this latitude. C doesn't. If it did, the stack would be more complicated, function calls would be more expensive, and programs would run a lot slower.

If you know that you have an array but you won't know until runtime how big it will be, declare a pointer to it and use malloc() or calloc() to allocate the array from the heap.

If you know at compile time how big an array is, you can declare its size at compile time. Even if the size is some complicated expression, as long as it can be evaluated at compile time, it can be used.

/* A program that copies the argv array and all the pointed-to
strings. It also deallocates all the copies. */
#include <stdlib.h>
#include <string.h>
int main(int argc, char** argv)
{
    char** new_argv;
    int i;
    /* Since argv[0] through argv[argc] are all valid, the
    program needs to allocate room for argc+1 pointers. */
    new_argv = (char**) calloc(argc+1, sizeof (char*));
    /* or malloc((argc+1) * sizeof (char*)) */
    printf("allocated room for %d pointers starting at %P\n", argc+1, new_argv);
    /* now copy all the strings themselves
    (argv[0] through argv[argc-1]) */
    for (i = 0; i < argc; ++i) {
        /* make room for '\0' at end, too */
    new_argv[i] = (char*) malloc(strlen(argv[i]) + 1);
        strcpy(new_argv[i], argv[i]);
        printf("allocated %d bytes for new_argv[%d] at %P, ""copied \"%s\"\n",
                strlen(argv[i]) + 1, i, new_argv[i], new_argv[i]);
    }
    new_argv[argc] = NULL;
    /* To deallocate everything, get rid of the strings (in any
    order), then the array of pointers. If you free the array
    of pointers first, you lose all reference to the copied
    strings. */
    for (i = 0; i < argc; ++i) 
    {
        free(new_argv[i]);
        printf("freed new_argv[%d] at %P\n", i, new_argv[i]);
        argv[i] = NULL;
    }
    free(new_argv);
    printf("freed new_argv itself at %P\n", new_argv);
    return 0; 
}

Both the malloc() and the calloc() functions are used to allocate dynamic memory. Each operates slightly different from the other. malloc() takes a size and returns a pointer to a chunk of memory at least that big:

void *malloc( size_t size );

calloc() takes a number of elements, and the size of each, and returns a pointer to a chunk of memory at least big enough to hold them all:

void *calloc( size_t numElements, size_t sizeOfElement );

There's one major difference and one minor difference between the two functions. The major difference is that malloc() doesn't initialize the allocated memory. The first time malloc() gives you a particular chunk of memory, the memory might be full of zeros. If memory has been allocated, freed, and reallocated, it probably has whatever junk was left in it. That means, unfortunately, that a program might run in simple cases (when memory is never reallocated) but break when used harder (and when memory is reused).

calloc() fills the allocated memory with all zero bits. That means that anything there you're going to use as a char or an int of any length, signed or unsigned, is guaranteed to be zero. Anything you're going to use as a pointer is set to all zero bits. That's usually a null pointer, but it's not guaranteed.

Anything you're going to use as a float or double is set to all zero bits; that's a floating-point zero on some types of machines, but not on all.

The minor difference between the two is that calloc() returns an array of objects; malloc() returns one object. Some people use calloc() to make clear that they want an array. Other than initialization, most C programmers don't distinguish between

calloc( numElements, sizeOfElement)

and

malloc( numElements * sizeOfElement)

There's a nit, though. malloc() doesn't give you a pointer to an array. In theory (according to the ANSI C standard), pointer arithmetic works only within a single array. In practice, if any C compiler or interpreter were to enforce that theory, lots of existing C code would break. (There wouldn't be much use for realloc(), either, which also doesn't guarantee a pointer to an array.)

Don't worry about the array-ness of calloc(). If you want initialization to zeros, use calloc(); if not, use malloc().

The coward's answer is, you can't, portably. The ANSI/ISO C standard requires compilers to handle only single objects as large as (32KB - 1) bytes long.

Why is 64KB magic? It's the biggest number that needs more than 16 bits to represent it.

For some environments, to get an array that big, you just declare it. It works, no trouble. For others, you can't declare such an array, but you can allocate one off the heap, just by calling malloc() or calloc().

On a PC compatible, the same limitations apply, and more. You need to use at least a large data model. You might also need to call "far" variants of malloc() or calloc(). For example, with Borland C and C++ compilers, you could write

far char *buffer = farmalloc(70000L);

Or with Microsoft C and C++ compilers, you could write

far char *buffer = fmalloc(70000L);

to allocate 70,000 bytes of memory into a buffer. (The L in 70000L forces a long constant. An int constant might be only 15 bits long plus a sign bit, not big enough to store the value 70,000.)

Compilers for PC compatibles use two types of pointers.

near pointers are 16 bits long and can address a 64KB range. far pointers are 32 bits long and can address a 1MB range.

near pointers operate within a 64KB segment. There's one segment for function addresses and one segment for data.

far pointers have a 16-bit base (the segment address) and a 16-bit offset. The base is multiplied by 16, so a far pointer is effectively 20 bits long. For example, if a far pointer had a segment of 0x7000 and an offset of 0x1224, the pointer would refer to address 0x71224. A far pointer with a segment of 0x7122 and an offset of 0x0004 would refer to the same address.

Before you compile your code, you must tell the compiler which memory model to use. If you use a small- code memory model, near pointers are used by default for function addresses. That means that all the functions need to fit in one 64KB segment. With a large-code model, the default is to use far function addresses. You'll get near pointers with a small data model, and far pointers with a large data model. These are just the defaults; you can declare variables and functions as explicitly near or far.

far pointers are a little slower. Whenever one is used, the code or data segment register needs to be swapped out. far pointers also have odd semantics for arithmetic and comparison. For example, the two far pointers in the preceding example point to the same address, but they would compare as different! If your program fits in a small-data, small-code memory model, your life will be easier. If it doesn't, there's not much you can do.

If it sounds confusing, it is. There are some additional, compiler-specific wrinkles. Check your compiler manuals for details.

Sometimes you can get away with using a small memory model in most of a given program. There might be just a few things that don't fit in your small data and code segments.

When that happens, you can use explicit far pointers and function declarations to get at the rest of memory. A far function can be outside the 64KB segment most functions are shoehorned into for a small-code model. (Often, libraries are declared explicitly far, so they'll work no matter what code model the program uses.)

A far pointer can refer to information outside the 64KB data segment. Typically, such pointers are used with farmalloc() and such, to manage a heap separate from where all the rest of the data lives.

If you use a small-data, large-code model, you should explicitly make your function pointers far.

A "stack trace" is a list of which functions have been called, based on this information. When you start using a debugger, one of the first things you should learn is how to get a stack trace.

The stack is very inflexible about allocating memory; everything must be deallocated in exactly the reverse order it was allocated in. For implementing function calls, that is all that's needed. Allocating memory off the stack is extremely efficient. One of the reasons C compilers generate such good code is their heavy use of a simple stack.

There used to be a C function that any programmer could use for allocating memory off the stack. The memory was automatically deallocated when the calling function returned. This was a dangerous function to call; it's not available anymore.

The heap is where malloc(), calloc(), and realloc() get memory.

Getting memory from the heap is much slower than getting it from the stack. On the other hand, the heap is much more flexible than the stack. Memory can be allocated at any time and deallocated in any order. Such memory isn't deallocated automatically; you have to call free().

Recursive data structures are almost always implemented with memory from the heap. Strings often come from there too, especially strings that could be very long at runtime.

If you can keep data in a local variable (and allocate it from the stack), your code will run faster than if you put the data on the heap. Sometimes you can use a better algorithm if you use the heap�faster, or more robust, or more flexible. It's a tradeoff.

If memory is allocated from the heap, it's available until the program ends. That's great if you remember to deallocate it when you're done. If you forget, it's a problem. A "memory leak" is some allocated memory that's no longer needed but isn't deallocated. If you have a memory leak inside a loop, you can use up all the memory on the heap and not be able to get any more. (When that happens, the allocation functions return a null pointer.) In some environments, if a program doesn't deallocate everything it allocated, memory stays unavailable even after the program ends.

Some programming languages don't make you deallocate memory from the heap. Instead, such memory is "garbage collected" automatically. This maneuver leads to some very serious performance issues. It's also a lot harder to implement. That's an issue for the people who develop compilers, not the people who buy them. (Except that software that's harder to implement often costs more.) There are some garbage collection libraries for C, but they're at the bleeding edge of the state of the art.