[MacRuby-devel] difference in float behaviour

[Laurent Sansonetti]

Hi Emil,

This is sort-of expected. MacRuby floats have less precision than MRI because we pack them as immediate values, for performance reasons. As a more generic issue, floating point arithmetic is almost never accurate.

Laurent

On Jun 5, 2010, at 9:59 AM, Emil Tin wrote:

> 
> hey, 
> thanks for all the great work with macruby! 
> 
> i'm experiencing a difference in floating point behaviour between mri and the latest macruby 0.7 nightly:
> 
> ~$ ruby -v
> ruby 1.8.7 (2009-06-08 patchlevel 173) [universal-darwin10.0]
> 
> ~$ ruby -e "z = 0.034228469425; puts Math.sqrt(z*z) < z"
> false
> 
> ~$ ruby -e "z = 0.034228469424; puts Math.sqrt(z*z) < z"
> false
> 
> 
> ~$ macruby -v
> MacRuby 0.7 (ruby 1.9.2) [universal-darwin10.0, x86_64]
> 
> ~$ macruby -e "z = 0.034228469425; puts Math.sqrt(z*z) < z"
> false
> 
> ~$ macruby -e "z = 0.034228469424; puts Math.sqrt(z*z) < z"
> true
> 
> 
> notice how that last script outputs true instead of false. also i'm seeing this difference with floats:
> 
> ~$ ruby -e "puts 8.1"
> 8.1
> 
> ~$ macruby -e "puts 8.1"
> 8.09999999999999
> 
> 
> 
> 
> this difference is causing trouble because asin() might fail even when the input seems to be 1.0:
> 
> z = 0.034228469424
> r = Math.sqrt z*z
> puts d = z/r # => 1.0
> puts d>1.0 # false with mri, but true with macruby
> puts Math.asin d # => 1.5707963267949 with mri, but error with macruby (since d>1.0)
> 
> 
> 
> -emil
> 
> _______________________________________________
> MacRuby-devel mailing list
> MacRuby-devel at lists.macosforge.org
> http://lists.macosforge.org/mailman/listinfo.cgi/macruby-devel