Use array annotation with pointers - c

Question

I have currently trouble understanding the following scenario:
I have a multidimensional array of Strings and I want to address it by using pointers only but I always get a Segmentation Fault when using the array annotation on the pointer. This is just an example code I want to use the 3D array in a pthread so I want to pass it in via a structure as a pointer but it just doesn't work and I would like to know why? I thought pointers and arrays are functionally equivalent? Here is the sample code:
#include <stdio.h>
void func(unsigned char ***ptr);
int main() {
// Image of dimension 10 times 10
unsigned char image[10][10][3];
unsigned char ***ptr = image;
memcpy(image[0][0], "\120\200\12", 3);
// This works as expected
printf("Test: %s", image[0][0]);
func(image);
return 0;
}
void func(unsigned char ***ptr) {
// But here I get a Segmentation Fault but why??
printf("Ptr: %s", ptr[0][0]);
}
Thanks in advance for your help :)

Answer 1

I think maybe strdup confuses the issue. Pointers and arrays are not always equivalent. Let me try to demonstrate. I always avoid actual multi-dimension arrays, so I may make a mistake here, but:
int main()
{
char d3Array[10][10][4]; //creates a 400-byte contiguous memory area
char ***d3Pointer; //a pointer to a pointer to a pointer to a char.
int i,j;
d3Pointer = malloc(sizeof(char**) * 10);
for (i = 0; i < 10; ++i)
{
d3Pointer[i] = malloc(sizeof(char*) * 10);
for (j = 0; j < 4; ++j)
{
d3Pointer[i][j] = malloc(sizeof(char) * 4);
}
}
//this
d3Pointer[2][3][1] = 'a';
//is equivalent to this
char **d2Pointer = d3Pointer[2];
char *d1Pointer = d2Pointer[3];
d1Pointer[1] = 'a';
d3Array[2][3][1] = 'a';
//is equivalent to
((char *)d3Array)[(2 * 10 * 4) + (3 * 4) + (1)] = 'a';
}
Generally, I use the layered approach. If I want contiguous memory, I handle the math myself..like so:
char *psuedo3dArray = malloc(sizeof(char) * 10 * 10 * 4);
psuedo3dArray[(2 * 10 * 4) + (3 * 4) + (1)] = 'a';
Better yet, I use a collection library like uthash.
Note that properly encapsulating your data makes the actual code incredibly easy to read:
typedef unsigned char byte_t;
typedef struct
{
byte_t r;
byte_t g;
byte_t b;
}pixel_t;
typedef struct
{
int width;
int height;
pixel_t * pixelArray;
}screen_t;
pixel_t *getxyPixel(screen_t *pScreen, int x, int y)
{
return pScreen->pixelArray + (y*pScreen->width) + x;
}
int main()
{
screen_t myScreen;
myScreen.width = 1024;
myScreen.height = 768;
myScreen.pixelArray = (pixel_t*)malloc(sizeof(pixel_t) * myScreen.height * myScreen.width);
getxyPixel(&myScreen, 150, 120)->r = 255;
}

Answer 2

In C, you should allocate space for your 2D array one row at a time. Your definition of test declares a 10 by 10 array of char pointers, so you don't need to call malloc for it. But to store a string you need to allocate space for the string. Your call to strcpy would crash. Use strdup instead. One way to write your code is as follows.
char ***test = NULL;
char *ptr = NULL;
test = malloc(10 * sizeof(char **));
for (int i = 0; i < 10; i++) {
test[i] = malloc(10 * sizeof(char *));
}
test[0][0] = strdup("abc");
ptr = test[0][0];
printf("%s\n", ptr);
test[4][5] = strdup("efg");
ptr = test[4][5];
printf("%s\n", ptr);
Alternatively, if you want to keep your 10 by 10 definition, you could code it like this:
char *test[10][10];
char *ptr = NULL;
test[0][0] = strdup("abc");
ptr = test[0][0];
printf("%s\n", ptr);
test[4][5] = strdup("efg");
ptr = test[4][5];
printf("%s\n", ptr);

Answer 3

Your problem is, that a char[10][10][3] is something very different from a char***: The first is an array of arrays of arrays, the later is a pointer to a pointer to a pointer. The confusions arises because both can be dereferenced with the same syntax. So, here is a bit of an explanation:
The syntax a[b] is nothing but a shorthand for *(a + b): First you perform pointer arithmetic, then you dereference the resulting pointer.
But, how come you can use a[b] when a is an array instead of a pointer? Well, because...
Arrays decay into pointers to their first element: If you have an array declared like int array[10], saying array + 3 results in array decaying to a pointer of type int*.
But, how does that help to evaluate a[b]? Well, because...
Pointer arithmetic takes the size of the target into account: The expression array + 3 triggers a calculation along the lines of (size_t)array + 3*sizeof(*array). In our case, the pointer that results from the array-pointer-decay points to an int, which has a size, say 4 bytes. So, the pointer is incremented by 3*4 bytes. The result is a pointer that points to the fourths int in the array, the first three elements are skipped by the pointer arithmetic.
Note, that this works for arrays of any element type. Arrays can contain bytes, or integers, or floats, or structs, or other arrays. The pointer arithmetic is the same.
But, how does that help us with multidimensional arrays? Well, because...
Multidimensional arrays are just 1D arrays that happen to contain arrays as elements: When you declare an array with char image[256][512]; you are declaring a 1D array of 256 elements. These 256 elements are all arrays of 512 characters, each. Since the sizeof(char) == 1, the size of an element of the outer array is 512*sizeof(char) = 512, and, since we have 256 such arrays, the total size of image is 256*512. Now, I can declare a 3D array with char animation[24][256][512];...
So, going back to your example that uses
char image[10][10][3]
what happens when you say image[1][2][1] is this: The expression is equivalent to this one:
*(*(*(image + 1) + 2) + 3)
image being of type char[10][10][3] decays into a pointer to its first element, which is of type char(*)[10][3] The size of that element is 10*3*1 = 30 bytes.
image + 1: Pointer arithmetic is performed to add 1 to the resulting pointer, which increments it by 30 bytes.
*(image + 1): The pointer is dereferenced, we are now talking directly about the element, which is of type char[10][3].
This array again decays into a pointer to its first element, which is of type char(*)[3]. The size of the element is 3*1 = 3. This pointer points at the same byte in memory as the pointer that resulted from step 2. The only difference is, that it has a different type!
*(image + 1) + 2: Pointer arithmetic is performed to add 2 to the resulting pointer, which increments it by 2*3 = 6 bytes. Together with the increment in step 2, we now have an offset of 36 bytes, total.
*(*(image + 1) + 2): The pointer is dereferenced, we are now talking directly about the element, which is of type char[3].
This array again decays into a pointer to its first element, which is of type char*. The size of the element is now just a single byte. Again, this pointer has the same value as the pointer resulting from step 5, but a different type.
*(*(image + 1) + 2) + 1: Pointer arithmetic again, adding 1*1 = 1 bytes to the total offset, which increases to 37 bytes.
*(*(*(image + 1) + 2) + 1): The pointer is dereferenced the last time, we are now talking about the char at an offset of 37 bytes into the image.
So, what's the difference to a char***? When you dereference a char***, you do not get any array-pointer-decay. When you try to evaluate the expression pointers[1][2][1] with a variable declared as
char*** pointers;
the expression is again equivalent to:
*(*(*(pointers + 1) + 2) + 3)
pointers is a pointer, so no decay happens. Its type is char***, and it points to a value of type char**, which likely has a size of 8 bytes (assuming a 64 bit system).
pointers + 1: Pointer arithmetic is performed to add 1 to the resulting pointer, which increments it by 1*8 = 8 bytes.
*(pointers + 1): The pointer is dereferenced, we are now talking about the pointer value that is found in memory at an offset of 8 bytes of where pointers points.
Further steps depending on what actually happened to be stored at pointers[1]. These steps do not involve any array-pointer-decay, and thus load pointers from memory instead.
You see, the difference between a char[10][10][3] and a char*** is profound. In the first case, the array-pointer-decay transforms the process into a pure offset computation into a multidimensional array. In the later case, we repeatedly load pointers from memory when accessing elements, all we ever have are 1D arrays of pointers. And it's all down to the types of pointers!