views:

151

answers:

7

How do I return a multidimensional array hidden in a private field?

class Myclass {
private:
 int myarray[5][5];
public:
 int **get_array();
};

........

int **get_array() {
 return grid;
}

cannot convert 'int ()[5][5]' to 'int*' in return test.cpp /Polky/src line 73 C/C++ Problem

A: 

Change your int's to int[][]'s or try using int[,] instead?

Harold
`int[,]` is not valid C++
David Rodríguez - dribeas
... and `int[][]` shouldn't be valid either. In this case it is important to understand the difference between arrays and pointers.
visitor
@David: Actually, `new int[5, 5]` is valid C++, it just doesn't do what one might think -- it's the same as `new int[5]` thanks to the comma operator :)
FredOverflow
@FredOverflow: right! :) Not just in that context...
David Rodríguez - dribeas
Ah, sorry, what am I thinking of? C# possibly? Haha, my bad
Harold
A: 
int **Myclass::get_array() { 
 return (int**)myarray; 
} 
rursw1
No. If you use that, you will be reinterpreting elements of the array as pointers.
Mike Seymour
@Mike Seymour, and that's true if you consider each row to be int* (like each vector of integers that is int*).
rursw1
+1  A: 

To return a pointer to your array of array member, the type needed is int (*)[5], not int **:

class Myclass {
private:
    int myarray[5][5];
public:
    int (*get_array())[5];
};

int (*Myclass::get_array())[5] {
    return myarray;
}
Charles Bailey
+1  A: 

I managed to make this function work in C++0x using automatic type deduction. However, I can't make it work without that. Native C arrays are not supported very well in C++ - their syntax is exceedingly hideous. You should use a wrapper class.

template<typename T, int firstdim, int seconddim> class TwoDimensionalArray {
    T data[firstdim][seconddim];
public:
    T*& operator[](int index) {
        return data[index];
    }
    const T*& operator[](int index) const {
        return data[index];
    }
};
class Myclass {
public:
    typedef TwoDimensionalArray<int, 5, 5> arraytype;
private:
    arraytype myarray;
public:
    arraytype& get_array() {
        return myarray;
    }
};

int main(int argc, char **argv) {
    Myclass m;
    Myclass::arraytype& var = m.get_array();
    int& someint = var[0][0];
}

This code compiles just fine. You can get pre-written wrapper class inside Boost (boost::array) that supports the whole shebang.

DeadMG
Doesn't compile with GCC: the first return data[index] makes a non-const reference from an rvalue, it says.
Cubbi
And that is the case, a bare pointer should be returned, not a reference to a pointer. Otherwise, we have at least gain much in readability. One minor point: why using `int` for the actual template types ? Something that cannot possibly be negative would be better expressed with an unsigned integer type I think.
Matthieu M.
data[index] is an lvalue, not an rvalue, much like *ptr is an lvalue. Not that the reference is really necessary, I think it's from a previous version of the code, you could probably remove it.
DeadMG
+8  A: 

A two-dimensional array does not decay to a pointer to pointer to ints. It decays to a pointer to arrays of ints - that is, only the first dimension decays to a pointer. The pointer does not point to int pointers, which when incremented advance by the size of a pointer, but to arrays of 5 integers.

class Myclass {
private:
    int myarray[5][5];
public:
    typedef int (*pointer_to_arrays)[5]; //typedefs can make things more readable with such awkward types

    pointer_to_arrays get_array() {return myarray;}
};

int main()
{
    Myclass o;
    int (*a)[5] = o.get_array();
    //or
    Myclass::pointer_to_arrays b = o.get_array();
}

A pointer to pointer (int**) is used when each subarray is allocated separately (that is, you originally have an array of pointers)

int* p[5];
for (int i = 0; i != 5; ++i) {
    p[i] = new int[5];
}

Here we have an array of five pointers, each pointing to the first item in a separate memory block, altogether 6 distinct memory blocks.

In a two-dimensional array you get a single contiguous block of memory:

int arr[5][5]; //a single block of 5 * 5 * sizeof(int) bytes

You should see that the memory layout of these things are completely different, and therefore these things cannot be returned and passed the same way.

visitor
+1 thanks, great answer
Sanjay Manohar
+2  A: 

How do I return a multidimensional array hidden in a private field?

If it's supposed to be hidden, why are you returning it in the first place?

Anyway, you cannot return arrays from functions, but you can return a pointer to the first element. What is the first element of a 5x5 array of ints? An array of 5 ints, of course:

int (*get_grid())[5]
{
    return grid;
}

Alternatively, you could return the entire array by reference:

int (&get_grid())[5][5]
{
    return grid;
}

...welcome to C declarator syntax hell ;-)

May I suggest std::vector<std::vector<int> > or boost::multi_array<int, 2> instead?

FredOverflow
+5  A: 

There are two possible types that you can return to provide access to your internal array. The old C style would be returning int *[5], as the array will easily decay into a pointer to the first element, which is of type int[5].

int (*foo())[5] {
   static int array[5][5] = {};
   return array;
}

Now, you can also return a proper reference to the internal array, the simplest syntax would be through a typedef:

typedef int (&array5x5)[5][5];
array5x5 foo() {
   static int array[5][5] = {};
   return array;
}

Or a little more cumbersome without the typedef:

int (&foo())[5][5] {
   static int array[5][5] = {};
   return array;
}

The advantage of the C++ version is that the actual type is maintained, and that means that the actual size of the array is known at the callers side.

David Rodríguez - dribeas
+1 for the C++ version (with the proper typedef)
Matthieu M.