Implementing Sparse Vectors With Associative Arrays/Compiler Bug?

[Ed]

I'm new to D and am trying to implement simple sparse vectors 
using associative arrays, but I'm getting fairly large floating 
point errors. Example code for sparse dot product:

import std.stdio;
import std.math;
import std.random;
static import std.datetime;

int main(string[] args) {
   double[int] v1;
   double[int] v2;
   Random gen;
   gen.seed(cast(uint)std.datetime.Clock.currTime().stdTime());

   double accum = 0;
   double val;
   foreach(i;1 .. 1000) {
     val = uniform(-1000.0,1000.0,gen);
     accum += val * val;
     v1[i] = val;
     v2[i] = val;
   }

   double accum2 = 0;
   double v2Val;
   foreach(k;v1.byKey()) {
     v2Val= v2.get(k,0);
     if(v2Val != 0) {
       accum2 += v1.get(k,0) * v2Val;
     }
   }

   writefln("accum - accum2 = %e", accum - accum2);

   return 0;
}

This outputs values such as:
accum - accum2 = -4.172325e-07
accum - accum2 = 2.384186e-07
accum - accum2 = 4.172325e-07

Are errors of this magnitude to be expected using doubles, or is 
this a compiler bug?