Except when it is the operand of the sizeof, _Alignof, or unary & operator, or is a string literal being used to initialize another array in a declaration, an expression of type "N-element array of T" will be converted ("decay") to an expression of type "pointer to T", and its value will be the address of the first element in the array (6.3.2.1/3).
In the line return hws; in populate, the type of the expression hws is "100-element array of 20-element array of int"; by the rule above, it will be converted to type "pointer to 20-element array of int", or int (*)[20]. Thus, the correct declaration for populate would need to be
int (*populate())[20]
{
...
return hws;
}
which reads as
populate -- populate
populate() -- is a function
*populate() -- returning a pointer
(*populate())[20] -- to a 20-element array
int (*populate())[20] -- of int.
and the type of whatever you return the result to would need to be int (*)[20] as well.
Having said that...
Using global variables this way is highly discouraged for a number of reasons. It would be better to pass the array in to populate as a parameter, like so:
void populate(int (*arr)[20], size_t rows)
{
size_t i, j;
for (i = 0; i < rows; i++)
{
for (j = 0; j < 20; j++)
{
arr[i][j] = randomize();
}
}
}
You would call this as simply
populate(hws, sizeof hws / sizeof hws[0]);
If you're using a compiler that supports variable length arrays (either a C99 compiler or a C2011 compiler that does not define __STDC_NO_VLA__ or defines it to 0), you could define the function as
void populate(size_t cols, size_t rows, int (*arr)[cols]) // cols must be declared
{ // before it can be used
size_t i, j; // to declare arr
for (i = 0; i < rows; i++)
{
for (j = 0; j < cols; j++)
{
arr[i][j] = randomize();
}
}
}
and call it as
size_t rows = sizeof hws[0] / sizeof hws[0][0]; // dividing the size of the array
size_t cols = sizeof hws / sizeof hws[0]; // by the size of the first element
populate(cols, rows, hws); // gives the number of elements
so you're not hardcoding the 20 anywhere.
If you aren't using a compiler that supports variable length arrays, and you don't want to hardcode the number of rows in the function prototype, you can do something like this:
void populate(int *arr, size_t rows, size_t cols)
{
size_t i, j;
for (i = 0; i < rows; i++)
{
for (j = 0; j < cols; j++)
{
arr[i * rows + j] = randomize();
}
}
}
and call it as
// rows and cols calculated as above
populate(&hws[0][0], rows, cols);
In this case, instead of passing a pointer to the array, we pass a pointer to the first element (the address value is the same, but the type is int * instead of int (*)[20]. So the populate function treats it like a 1D array, and computes the offset with i * rows + j. Note that this trick will only work for 2D arrays that have been allocated contiguously.
This also means that populate can work with arrays of any size, not just Nx20.
hwsis not an array. it's an array of arrays.