On 7/23/15 5:58 PM, jmh530 wrote:
I was looking at
http://dlang.org/concepts.html
where it discusses overloading templates based on constraints. I wanted
to try to overload a template function with another version that does
everything in place. So basically, one has a return and the other
doesn't. However, when I run the code, it only calls the first function.
The second one is ignored. I tried a number of adjustments, but the only
thing that worked is to re-name the function something else, remove the
U's, and just have it be a void function.
Just so you know, *every* template is tried for every call. So the if
statements are not dependent on each other. Something I've wanted for a
long time is an if-else mechanism for template constraints.
What ends up happening is you have to repeat your constraints on every
overload, but negate the previous ones.
In any case, to your code:
import std.stdio : writeln;
import std.traits;
T test(T)(T x)
if (isNumeric!(T))
{
writeln("calling test without void");
T y = x;
y += 1;
return y;
}
U test(T, U)(ref T x)
if (isNumeric!(T) && is(U == void))
this will NEVER be called via IFTI, because U cannot be determined from
the parameters. Return type (or how you use the result) does not play
into IFTI at all.
This is likely why it's never used, because it fails to instantiate.
And note that regular overloading does not work like this. You can't
overload on return type. Overloading on ref works, but only makes it so
you can dictate how to handle rvalues vs. lvalues differently.
-Steve
