struct vector (T, uint N)
{

    T[N] S;

    .vector!(T,N*2) opCat(vector a);

}

alias vector!(float,3) float3;

void main()
{
}

>dmd.exe test.d
test.d(5): index -1073741824 overflow for static array

XP DMD 0.97

Bent Rasmussen wrote:

> struct vector (T, uint N)
> {
> 
>     T[N] S;
> 
>     .vector!(T,N*2) opCat(vector a);
> 
> }
> 
> alias vector!(float,3) float3;
> 
> void main()
> {
> }
> 
> 
>>dmd.exe test.d
> 
> test.d(5): index -1073741824 overflow for static array
> 
> XP DMD 0.97
> 
> 

I dont't think it's a bug except for the signed output. What happens here is that vector is instantiated recursively by having N = 3, then 3*2, 3*4, 3*8, 3*16, 3*32, ... until it reaches 3*2^30. It's equal 3221225472 then. 3*2^30 when stored in a signed int == -1073741824. If that 3*2^30 were to be multiplied by 2 once more, it would overflow the dword storage it's given and be equal 6*2^30 == 3*2^31 and thats more than 2^32-1 which cannot be handled by a dword. Here, the compiler properly signals that something is wrong. It could give some more meaningful information though, at least by printing the number as an unsigned int.

I'm not sure but I think it's a downside of the en-masse template instantiation we have in D. havent tried this kind of thing in C++ though...

regards, Tom

You're right, I was too quick on that one.