Hacker News Reader

Comment by FreeFull | Hacker News Reader

    red>> .1 + .2 + .3 = .6
    == true
    red>> .1 + .2 + .3 - .6
    == 1.110223024625157e-16

This doesn't feel right to me. I usually would expect equality to be a strictly exact thing. This is bound to cause some problems in code that assumes IEEE-754 has been implemented exactly.

View on HN

The language distinguishes between equal and "strictly equal":

    red>> .1 + .2 + .3 == .6
    == false
    red>> .1 + .2 + .3 = .6
    == true

I'm not sure if that syntax is the wisest choice, but in the absence of arbitrary precision arithmetic it seems a very good concept, to avoid common IEEE-754 surprises.

They appear to be implementing it along these lines: http://randomascii.wordpress.com/2012/02/25/comparing-floati...

It should be noted that Rebol comes with an arbitrary precision (bignum) arithmetic option via it's money! datatype - http://www.rebol.net/wiki/Money

  >> .1 + .2 + .3 
  == 0.6000000000000001
  
  >> type? 0.1    
  == decimal!
  
  >> $.1 + $.2 + $.3
  == $0.6
  
  >> type? $0.1
  == money!

Red will most likely get this soon but perhaps with a different nomenclature (Red as already renamed decimal! to float!).

It also doesn't feel right to me when I learned it for the first time. But if you know that "=" means "almost equal" when doing float comparision, it is very convenient in practical use. For example, if we want to compare two float variables. In Red:

  if float-1 = float-2 [
      ; do something
  ]

In other languages:

  if (float-1 - float-2) < EPS {
      // do something
  }

Obviously, Red's way is more natural and readable. :)

Choosing the epsilon requires numerical analyis expertise or lucky cargo-cuting: https://randomascii.wordpress.com/2012/02/25/comparing-float...

Thank you for reminding us about those great articles! We were using a Ulp-based method only, but Qtxie proceeded with some improvements [1] for close to zero values, as suggested in the article. Thanks again. :-)

[1] https://github.com/red/red/pull/986

Reply on news.ycombinator.com