Sorting floating-point values, and NaN

[Walter Bright]

On 11/12/2013 12:24 PM, Andrei Alexandrescu wrote:
> You're young. I guarantee you were to hit NaN whether or not D had anything to
> do with it.

I want to add that NaN has been a reality on x86 machines since about 1983. C 
(and C++) compilers for decades have utterly ignored its existence - but that 
didn't make it go away. It just meant that things like the fp functions in the 
Standard library exhibited undefined behavior with NaNs, which is far worse than 
at least having sensible documented behavior.

The C compilers I wrote (and C++) always made an effort to handle NaN correctly 
(and overflows, underflows, and subnormals), and I often felt that I was the 
only one who cared about it :-)

What D does is simply recognize that NaNs are an inevitable characteristic of 
the floating point hardware, and deal with them in the most straightforward 
manner practical. Deciding and documenting what .sort should do with NaNs is 
part of that. Trying to pretend that NaNs aren't a fact of life with IEEE 
floating point hardware is not going to work.

Note:

http://dlang.org/phobos/std_math.html#.exp

and how the behavior with NaN arguments is all carefully documented.