Is the BigDecimal class broken? It seems like the following should never, ever occur: Note that a.to_f != a.to_s.to_f
a.class => BigDecimal
a.to_f => 18658.1072928
a.to_s => "10865.81072928"
b.class => BigDecimal
b.to_f => 10000.0
b.to_s => "10000.0"
(a - b).to_f => 865.81072928
a.to_f - b.to_f => 8658.1072928
Any ideas as to what might be going wrong? We are running ruby 1.8.7p72 on our servers and 1.8.7p173 on our local machines.
As the docs say:
to_f:
Returns a new Float object having approximately the same value as the BigDecimal number. Normal accuracy limits and built-in errors of binary Float arithmetic apply.
That means that Float in ruby behaves just like a float everywhere: http://en.wikipedia.org/wiki/Floating_point#Accuracy_problems
Not every number is representable by a Float. In such cases the number is rounded.
If you're running Mac OS X 10.6.3 then this might be an explanation: http://www.abletech.co.nz/2010/03/osx-10-6-3-installs-new-ruby-with-faulty-bigdecimal/
To get rid of the problem you could use RVM.