Communication between D and C with dynamic arrays
via Digitalmars-d-learn
digitalmars-d-learn at puremagic.com
Sun Aug 10 08:42:20 PDT 2014
On Sunday, 10 August 2014 at 14:26:29 UTC, seany wrote:
> In D, arrays are dynamic. However, to the best of my knowledge,
> in C, they are static.
>
> I am having difficulty in imagining how to send D arrays to a C
> function.
>
> My first Idea was to make a pointer to the array. then find the
> size of the array, which itself is an array, therefore take the
> pointer to it, then get the dimension of the array, which is an
> integer, and send the trio of two pointers and the dimesion to
> the C code.
>
In D, there both dynamic and static arrays. The static ones are
those whose length is known at compile time, i.e.
int[10] my_numbers;
These are treated as value types, not references. When a function
takes a static array, it receives a copy of the entire array, not
a reference to it:
void foo(int[4] data);
This has no direct equivalent in C.
Then there are dynamic arrays. Under the hood, a dynamic array
with elements of type T is actually a struct (well, it behaves as
if it is one):
struct DynamicArray {
size_t length;
T* ptr;
}
With T being the specific type. (The length and ptr members might
be in the opposite order.) You can always access the length and
pointer of both types of arrays:
void main() {
import std.stdio;
int[2] static_array;
int[] dynamic_array;
writeln(static_array.length); // 2
writeln(dynamic_array.length); // 0
dynamic_array ~= [1,2,3,4]; // append elements
writeln(dynamic_array.length); // 4
// you can even set the length
writeln(dynamic_array); // [1, 2, 3, 4]
dynamic_array.length = 2;
writeln(dynamic_array); // [1, 2]
// growing fills the array with default value
dynamic_array.length = 4;
writeln(dynamic_array); // [1, 2, 0, 0]
// slicing
dynamic_array = static_array[]; // dyn.a. now points to
stat.a.
// dynamic_array.length == static_array.length
// dynamic_array.ptr == static_array.ptr
}
The details of arrays in D are described here:
http://dlang.org/arrays
As for communicating with C, it depends.
Some C functions only take a pointer to an array. They either
already know the length from somewhere else, or they use a
special value in the array as a signal that the array ends there.
The latter is typically used for strings, which are defined to
end with a 0 byte. For passing arrays to these C functions, you
can use the `ptr` property; alternatively, you can take the
address of the first element.
extern(C) my_c_func(int* data);
int[] a = [1,2,3,4,5];
my_c_func(a.ptr);
my_c_func(&(a[0]));
Of course, you may need to make sure that the array actually
contains the magic value at its end. For strings, this can be
done with std.string.toStringz:
http://dlang.org/phobos/std_string.html#.toStringz
In most cases, however, the C function also accepts the length of
the array. For these, you need to use the `length` property:
extern(C) my_other_c_func(int* data, size_t length);
int[] b = [6,7,8,9];
my_other_c_func(b.ptr, b.length);
> Now this arises two questions :
>
> 1. I assume that D allocates a particular space to the array,
> say N elements, and up to these N elements, you can increment
> the pointers, to jump to next array element.
>
> What happens if the N element space is used up? Does the
> pointer incrementing method break down?
Yes, the array elements are stored next to each other in memory,
without gaps. By default, D checks whether you are still inside
the bounds of the array, and raises an error if you access an
element outside of the bounds. But this only works if you access
the array directly using an index, or slicing:
int[] a = [10,11,12,13];
writeln(a[0]); // 10
writeln(a[3]); // 13
//writeln(a[4]); // ERROR
// slicing: indices 1 (incl.) up to 3 (excl.)
writeln(a[1 .. 3]); // [11, 12]
// length is represented by $
writeln(a[2 .. $]); // [12, 13]
//writeln(a[0 .. 5]); // ERROR
You mentioned incrementing a pointer. This does work, just like
in C, but if you do this, you are responsible for checking the
array bounds. The language cannot help you there. If you access
something outside of the array, it's undefined behaviour:
int[] a = [20,30,40,50];
int* b = a.ptr;
writeln(*b); // 20
writeln(b[0]); // 20
writeln(b[0 .. 4]); // [20, 30, 40, 50]
//writeln(b[0 .. $]); // doesn't compiler, length is unknown
b++;
writeln(b[0]); // 30
//writeln(b[100]); // ???
// might print garbage, might crash, might eat your
harddisk...
>
> 2. in D pointers are being converted to void * as I read in the
> reply to another post of mine. I dont remember in my knowledge
> of C, that they are accepted in C, are they? Do I have to meake
> a type retrospection every time I get send something to C?
>
> Any help is appreciated.
I think this is a misunderstanding. If you take a pointer to a
variable of type `T` (or to an element of an array of `T`s), the
pointer's type is `T*`. This is just like in C. `void*` is a
special pointer type that you can use if you don't already know
the real type it is pointing to, or if it is supposed to be
opaque. This also exists in C.
Which pointer type you need to use when calling a C function
depends of course on that function: Just use whatever the
function accepts.
>
> PS: is there a built in size operator for arrays in D?
For the `length` property (the number of elements), see above.
There is also the `sizeof` property, which returns the actual
size that a variable occupies in memory. But be careful: for
dynamic arrays, `sizeof` returns either 8 or 16, depending on
whether you are compiling 32bit or 64bit programs. This is
because, as explained above, dynamic array variables are
internally just a pair of length and pointer.
int[] a;
int[16] b;
writeln(a.sizeof); // 16
writeln(b.sizeof); // 64
More information about the Digitalmars-d-learn
mailing list