Nobody understands templates?

[H. S. Teoh]

On Thu, Mar 06, 2014 at 12:02:42AM +0000, develop32 wrote:
> On Wednesday, 5 March 2014 at 23:47:33 UTC, H. S. Teoh wrote:
> >Whoa. What did you do with those arrays?? Either you did something
> >wrong, or there's a nasty bug somewhere in the compiler/language;
> >AFAIK static arrays are supposed to be value types so they shouldn't
> >generate any garbage at all.
> 
> I think it was the case of using array literals, like this (I didn't
> know much about D back then)
> 
> this(float x, float y, float z)
> {
>    this.vector = [x, y, z];
> }

I assume this.vector is a float[3]?

Yeah, this is one of the things in D that I find disappointing. Array
literals are a minefield of hidden allocations and runtime performance
hits. In theory, the compiler *should* realize that since this.vector is
a static array, it should just assign x, y, z directly to the array
elements. But IIRC (assuming the compiler hasn't changed in this
respect) what it actually does is to construct a *dynamic* array [x, y,
z] and then assign it to the static array. Which, of course, produces
huge amounts of garbage. And which kinda defeats the purpose of using
static arrays in the first place...  :-(


> And megabytes accumulated because there were hundreds of objects all
> doing complicated stuff every frame, passing and constructing
> vectors and matrices around.
> 
> Memory leaks could have been avoided, but still, one should be
> careful when using arrays.

A recent idiom of mine:

	struct Vector(T, size_t n) {
		T[n] impl;
		alias impl this;

		this(Args...)(Args args)
			if (Args.length == n)
		{
			// This is compile-time unrolled
			foreach (i, arg; args) {
				impl[i] = arg;
			}
		}
	}


T

-- 
The two rules of success: 1. Don't tell everything you know. -- YHL