有点晚了，但希望这能帮助到一些人。

在C语言中，数组总是在堆栈上分配内存，因此函数不能返回 一个(非静态)数组，因为内存分配在堆栈上 当执行到达当前块的末尾时自动释放。 当你想处理二维数组时，这真的很烦人 (即矩阵)，并实现一些可以改变和返回矩阵的函数。 要实现这一点，可以使用指针对指针来实现矩阵 动态分配内存:

/* Initializes a matrix */
double** init_matrix(int num_rows, int num_cols){
    // Allocate memory for num_rows float-pointers
    double** A = calloc(num_rows, sizeof(double*));
    // return NULL if the memory couldn't allocated
    if(A == NULL) return NULL;
    // For each double-pointer (row) allocate memory for num_cols floats
    for(int i = 0; i < num_rows; i++){
        A[i] = calloc(num_cols, sizeof(double));
        // return NULL if the memory couldn't allocated
        // and free the already allocated memory
        if(A[i] == NULL){
            for(int j = 0; j < i; j++){
                free(A[j]);
            }
            free(A);
            return NULL;
        }
    }
    return A;
} 

这里有一个例子:

double**       double*           double
             -------------       ---------------------------------------------------------
   A ------> |   A[0]    | ----> | A[0][0] | A[0][1] | A[0][2] | ........ | A[0][cols-1] |
             | --------- |       ---------------------------------------------------------
             |   A[1]    | ----> | A[1][0] | A[1][1] | A[1][2] | ........ | A[1][cols-1] |
             | --------- |       ---------------------------------------------------------
             |     .     |                                    .
             |     .     |                                    .
             |     .     |                                    .
             | --------- |       ---------------------------------------------------------
             |   A[i]    | ----> | A[i][0] | A[i][1] | A[i][2] | ........ | A[i][cols-1] |
             | --------- |       ---------------------------------------------------------
             |     .     |                                    .
             |     .     |                                    .
             |     .     |                                    .
             | --------- |       ---------------------------------------------------------
             | A[rows-1] | ----> | A[rows-1][0] | A[rows-1][1] | ... | A[rows-1][cols-1] |
             -------------       ---------------------------------------------------------

The double-pointer-to-double-pointer A points to the first element A[0] of a memory block whose elements are double-pointers itself. You can imagine these double-pointers as the rows of the matrix. That's the reason why every double-pointer allocates memory for num_cols elements of type double. Furthermore A[i] points to the i-th row, i.e. A[i] points to A[i][0] and that's just the first double-element of the memory block for the i-th row. Finally, you can access the element in the i-th row and j-th column easily with A[i][j].

下面是一个完整的例子来演示它的用法:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

/* Initializes a matrix */
double** init_matrix(int num_rows, int num_cols){
    // Allocate memory for num_rows double-pointers
    double** matrix = calloc(num_rows, sizeof(double*));
    // return NULL if the memory couldn't allocated
    if(matrix == NULL) return NULL;
    // For each double-pointer (row) allocate memory for num_cols
    // doubles
    for(int i = 0; i < num_rows; i++){
        matrix[i] = calloc(num_cols, sizeof(double));
        // return NULL if the memory couldn't allocated
        // and free the already allocated memory
        if(matrix[i] == NULL){
            for(int j = 0; j < i; j++){
                free(matrix[j]);
            }
            free(matrix);
            return NULL;
        }
    }
    return matrix;
}

/* Fills the matrix with random double-numbers between -1 and 1 */
void randn_fill_matrix(double** matrix, int rows, int cols){
    for (int i = 0; i < rows; ++i){
        for (int j = 0; j < cols; ++j){
            matrix[i][j] = (double) rand()/RAND_MAX*2.0-1.0;
        }
    }
}


/* Frees the memory allocated by the matrix */
void free_matrix(double** matrix, int rows, int cols){
    for(int i = 0; i < rows; i++){
        free(matrix[i]);
    }
    free(matrix);
}

/* Outputs the matrix to the console */
void print_matrix(double** matrix, int rows, int cols){
    for(int i = 0; i < rows; i++){
        for(int j = 0; j < cols; j++){
            printf(" %- f ", matrix[i][j]);
        }
        printf("\n");
    }
}


int main(){
    srand(time(NULL));
    int m = 3, n = 3;
    double** A = init_matrix(m, n);
    randn_fill_matrix(A, m, n);
    print_matrix(A, m, n);
    free_matrix(A, m, n);
    return 0;
}

这里的大多数答案或多或少都与应用程序编程有关。下面是一个嵌入式系统编程的例子。例如，以下是NXP Kinetis KL13系列微控制器参考手册的摘录，此代码片段用于从固件中运行驻留在ROM中的引导加载程序:

＂ 为了获得入口点的地址，用户应用程序读取包含引导加载程序API树指针的单词，该指针位于引导加载程序向量表的0x1C偏移量处。向量表被放置在引导加载器地址范围的底部，ROM的地址范围是0x1C00_0000。因此，API树指针位于地址0x1C00_001C。

引导加载程序API树是一个包含指向其他结构的指针的结构，这些结构具有引导加载程序的函数和数据地址。引导加载程序入口点总是API树的第一个单词。 ＂

uint32_t runBootloaderAddress;
void (*runBootloader)(void * arg);
// Read the function address from the ROM API tree.
runBootloaderAddress = **(uint32_t **)(0x1c00001c);
runBootloader = (void (*)(void * arg))runBootloaderAddress;
// Start the bootloader.
runBootloader(NULL);

例如，如果您想随机访问不连续的数据。

p -> [p0, p1, p2, ...]  
p0 -> data1
p1 -> data2

——用C

T ** p = (T **) malloc(sizeof(T*) * n);
p[0] = (T*) malloc(sizeof(T));
p[1] = (T*) malloc(sizeof(T));

存储一个指针p，它指向一个指针数组。每个指针指向一段数据。

如果sizeof(T)很大，则可能无法分配sizeof(T) * n字节的连续块(即使用malloc)。

I have used double pointers today while I was programming something for work, so I can answer why we had to use them (it's the first time I actually had to use double pointers). We had to deal with real time encoding of frames contained in buffers which are members of some structures. In the encoder we had to use a pointer to one of those structures. The problem was that our pointer was being changed to point to other structures from another thread. In order to use the current structure in the encoder, I had to use a double pointer, in order to point to the pointer that was being modified in another thread. It wasn't obvious at first, at least for us, that we had to take this approach. A lot of address were printed in the process :)).

当你处理在应用程序其他地方被更改的指针时，你应该使用双指针。在处理返回和寻址给您的硬件时，您可能还会发现双指针是必须的。

字符串是使用双指针的一个很好的例子。字符串本身是一个指针，所以任何时候你需要指向一个字符串，你就需要一个双指针。

比较变量的修改值和指针的修改值:

#include <stdio.h>
#include <stdlib.h>

void changeA(int (*a))
{
  (*a) = 10;
}

void changeP(int *(*P))
{
  (*P) = malloc(sizeof((*P)));
}

int main(void)
{
  int A = 0;

  printf("orig. A = %d\n", A);
  changeA(&A);
  printf("modi. A = %d\n", A);

  /*************************/

  int *P = NULL;

  printf("orig. P = %p\n", P);
  changeP(&P);
  printf("modi. P = %p\n", P);

  free(P);

  return EXIT_SUCCESS;
}

这帮助我避免指针被调用函数修改时返回指针的值(用于单链表)。

古老的(坏的):

int *func(int *P)
{
  ...
  return P;
}

int main(void)
{
  int *pointer;
  pointer = func(pointer);
  ...
}    

新(更好的):

void func(int **pointer)
{
  ...
}

int main(void)
{
  int *pointer;
  func(&pointer);
  ...
}