Sorting floating-point values, and NaN

[tn]

On Tuesday, 12 November 2013 at 09:40:52 UTC, Walter Bright wrote:
> On 11/12/2013 1:33 AM, tn wrote:
>> I could not find any documentation on how the unordered
>> comparison operators (<>, !<>=, !<=, !<, !>=, !>, !<>) 
>> translate into opCmp calls.
>
> That's because they just don't translate to opCmp calls.

Well, they seem to translate to something, because this works as 
expected:

   struct S {
       private double v;
       auto opCmp(S rhs) { return v - rhs.v; }
   }

   S v1 = S(1);
   S v2 = S(2);
   S vn = S(double.nan);
   assert((v2 < v1) == false);
   assert((v1 < v2) == true);
   assert((v1 < v1) == false);
   assert((v1 < vn) == false);
   assert((v2 !>= v1) == false);
   assert((v1 !>= v2) == true);
   assert((v1 !>= v1) == false);
   assert((v1 !>= vn) == true);


> but then we went ahead and deprecated those operators

Maybe the spec then needs to be updated:

http://dlang.org/expression.html#RelExpression
http://dlang.org/expression.html#floating_point_comparisons