int[3]   a = [1,2,4];
    float[3] b = [1,2,4];
    float[3] c;
    // Why doesn't this work?
    c = a[] + b[]; // Error: incompatible types for ((a[]) + (b[])): 'int[]' and 'float[]'
    // When this works?
    c[0] = a[0] + b[0];
    c[1] = a[1] + b[1];
    c[2] = a[2] + b[2];
    assert(c == [2,4,8]);

On 16.04.2011 12:12, simendsjo wrote:
> int[3] a = [1,2,4];
> float[3] b = [1,2,4];
> float[3] c;
> // Why doesn't this work?
> c = a[] + b[]; // Error: incompatible types for ((a[]) + (b[])): 'int[]'
> and 'float[]'
> // When this works?
> c[0] = a[0] + b[0];
> c[1] = a[1] + b[1];
> c[2] = a[2] + b[2];
> assert(c == [2,4,8]);

I tried using a template mixin to avoid having to type this all over, but I cannot get it to work.. I think I have something wrong with my alias usage, but that's just speculating :)

mixin template ArrayVectorOp(string op, alias dest, alias a, alias b, int I)
if(dest.length == a.length && dest.length == b.length && I >= 1 && I <= dest.length) {
    // dest[I-1] = a[I-1] op b[I-1]
    mixin(dest.stringof~"["~(I-1).stringof~"]"
          " = "
          ~a.stringof~"["~(I-1).stringof~"]"
          ~op
          ~b.stringof~"["~(I-1).stringof~"];");

    static if(I > 1)
        mixin ArrayVectorOp!(op, dest, a, b, I-1);
}

void main() {
    int[3]   a = [1,   2,   3];
    float[3] b = [2.1, 3.2, 4.3];
    float[3] c;
    mixin ArrayVectorOp!("+", c, a, b, a.length);
    assert(c == [3.1, 5.2, 7.3]);
}


Gives the following output:

t.d(4): no identifier for declarator c[3 - 1]
t.d(4): Error: c is used as a type
t.d(4): Error: cannot implicitly convert expression (cast(float)a[2u] + b[2u]) o
f type float to const(_error_[])
t.d(4): no identifier for declarator c[2 - 1]
t.d(4): Error: c is used as a type
t.d(4): Error: cannot implicitly convert expression (cast(float)a[1u] + b[1u]) o
f type float to const(_error_[])
t.d(4): no identifier for declarator c[1 - 1]
t.d(4): Error: c is used as a type
t.d(4): Error: cannot implicitly convert expression (cast(float)a[0u] + b[0u]) o
f type float to const(_error_[])
t.d(11): Error: mixin t.main.ArrayVectorOp!("+",c,a,b,3u).ArrayVectorOp!(op,c,a,
b,2).ArrayVectorOp!(op,c,a,b,1) error instantiating
t.d(11): Error: mixin t.main.ArrayVectorOp!("+",c,a,b,3u).ArrayVectorOp!(op,c,a,
b,2) error instantiating
t.d(18): Error: mixin t.main.ArrayVectorOp!("+",c,a,b,3u) error instantiating

simendsjo:

>      int[3]   a = [1,2,4];
>      float[3] b = [1,2,4];
>      float[3] c;
>      // Why doesn't this work?
>      c = a[] + b[]; // Error: incompatible types for ((a[]) + (b[])):
> 'int[]' and 'float[]'

>      // When this works?
>      c[0] = a[0] + b[0];
>      c[1] = a[1] + b[1];
>      c[2] = a[2] + b[2];
>      assert(c == [2,4,8]);

Vector ops are often implemented in assembly, and despite they are less flexible, they sometimes lead to more efficiency (if the arrays are large). The second example uses normal D code, that's much more flexible.

Bye,
bearophile

On Sat, Apr 16, 2011 at 13:37, simendsjo <simen.endsjo@pandavre.com> wrote:
> I tried using a template mixin to avoid having to type this all over, but I cannot get it to work.. I think I have something wrong with my alias usage, but that's just speculating :)

Try this:


auto arrayOp(string op, A, B, int N, int n = 0)(A[N] a, B[N] b)
{
    static if (n < N - 1)
        mixin("return [a[n] " ~ op ~ " b[n]] ~ arrayOp!(op,A,B,N,n+1)(a, b);");
    else
        mixin("return [a[n] " ~ op ~ " b[n]];");
}

void main()
{
    int[3]   a = [1,2,4];
    float[3] b = [1,2,4];
    auto c = arrayOp!"+"(a,b);
    writeln(c);
}

I had to resort to the N/n kludge, because of A[N] <-> A[] incompatibilities.

Note that this could easily be extend to accept any string function (ala std.algorithm) and any number of arrays.

Philippe