Ruby float calculations


#1

Could someone please explain to me why the following happens:

irb(main):001:0> 0.29 - 0.38 + 0.1
=> 0.00999999999999998
irb(main):002:0> 0.29 + (-0.38 + 0.1)
=> 0.00999999999999995
irb(main):003:0> 0.29 + 0.1 - 0.38
=> 0.01

How can I prevent something like this from happening, or at the very
least how can I work around such errors?

Many thanks in advance.


#2

On Fri, 10 Mar 2006, Alexandru T. wrote:

least how can I work around such errors?

Many thanks in advance.

it is a hardware limitation, even a C program will do the same:

 harp:~ > cat a.c
 #include <stdlib.h>
 #include <stdio.h>

 main () {
   printf ("%32.32f\n", 0.29 - 0.38 + 0.1);
   printf ("%32.32f\n", 0.29 + (-0.38 + 0.1));
   printf ("%32.32f\n", 0.29 + 0.1 - 0.38);
 }


 harp:~ > gcc a.c && a.out
 0.00999999999999998112620858137234
 0.00999999999999995337063296574343
 0.01000000000000000888178419700125

with ruby, at least, you can use BigDecimal or the like - but with a
serious
speed penalty. in short this is just the way computers work:

http://en.wikipedia.org/wiki/Floating_point

hth.

-a


#3

On 3/9/06, Alexandru T. removed_email_address@domain.invalid wrote:

Could someone please explain to me why the following happens:

irb(main):001:0> 0.29 - 0.38 + 0.1
=> 0.00999999999999998
irb(main):002:0> 0.29 + (-0.38 + 0.1)
=> 0.00999999999999995
irb(main):003:0> 0.29 + 0.1 - 0.38
=> 0.01

Check this out:

d1 = 0.29 - 0.38 + 0.1
d2 = 0.29 + (-0.38 + 0.1)
d3 = 0.29 + 0.1 - 0.38
p (d3 - d2)
p Float::EPSILON
(d3 - d2) < Float::EPSILON #=> true

Computers do not store numbers indefinitely, so there is an “accuracy”
associated with floating point operations. For more information,
check out:
http://en.wikipedia.org/wiki/Floating_point

Cameron


#4

My question now would be if there is an easy way to implement
floating-point decimal arithmetic in Ruby.

Forget about this… it has already been answered… I should have been
more careful

with ruby, at least, you can use BigDecimal or the like - but with a
serious
speed penalty. in short this is just the way computers work:

Thanks again


#5

http://en.wikipedia.org/wiki/Floating_point

Thanks for your answers. It’s clear now. Expecially this link on the IBM
site made it very clear:
http://www2.hursley.ibm.com/decimal/decifaq1.html#inexact

My question now would be if there is an easy way to implement
floating-point decimal arithmetic in Ruby.