Forum: Ruby Unexpected BigDecimal rounding issue

Announcement (2017-05-07): www.ruby-forum.com is now read-only since I unfortunately do not have the time to support and maintain the forum any more. Please see rubyonrails.org/community and ruby-lang.org/en/community for other Rails- und Ruby-related community platforms.
0eaa4f32363d027f0d905d36272d98c9?d=identicon&s=25 Samuel Lown (samlown)
on 2009-05-06 03:01
Hi,

Some more not so fun rounding issues.

Using Floats:

>> ((30 / 1.16) * 1.16) == 30
=> true

Cool, so you'd think it would work with BigDecimal:

>> ((30 / BigDecimal("1.16")) * BigDecimal("1.16")) == 30
=> false
>> ((30 / BigDecimal("1.16")) * BigDecimal("1.16")).to_s
=> "30.0000000000000000000000002"

I'm clearly missing something here as I thought BigDecimal was supposed
to fix this type of thing.

Tested in:
ruby 1.8.6 (2008-03-03 patchlevel 114) [universal-darwin9.0]
ruby 1.9.1p0 (2009-01-30 revision 21907) [i386-darwin9.6.0]

Can anyone offer an explanation?

Cheers, sam
F1d6cc2b735bfd82c8773172da2aeab9?d=identicon&s=25 Nobuyoshi Nakada (nobu)
on 2009-05-06 05:06
(Received via mailing list)
Hi,

At Wed, 6 May 2009 10:01:54 +0900,
Samuel Lown wrote in [ruby-talk:335867]:
> >> ((30 / BigDecimal("1.16")) * BigDecimal("1.16")) == 30
> => false
> >> ((30 / BigDecimal("1.16")) * BigDecimal("1.16")).to_s
> => "30.0000000000000000000000002"
>
> I'm clearly missing something here as I thought BigDecimal was supposed
> to fix this type of thing.

BigDecimal is another kind of floating point number, which uses
decimal base instead of binary, but has finite digits.
Therefore, it is impossible to represent exactly a recurring
decimal theoretically.
8f6f95c4bd64d5f10dfddfdcd03c19d6?d=identicon&s=25 Rick Denatale (rdenatale)
on 2009-05-06 05:30
(Received via mailing list)
On Tue, May 5, 2009 at 9:01 PM, Samuel Lown <me@samlown.com> wrote:
>
> ruby 1.9.1p0 (2009-01-30 revision 21907) [i386-darwin9.6.0]
>
> Can anyone offer an explanation?

BigDecimal is still a kind of float, albeit a decimal float rather
than a binary float.

So it helps with problems when the fractional part of a number can be
expressed exactly as a finite string of decimal digits, but just as
certain fractional parts can't be expressed exactly as a finite string
of binary digits, there are some which have the same problem when the
base is 10.

For example 1/3 cannot be expressed as a decimal float, no matter how
many digits 0.33333.....

Similarly 3000/116 (which is the same as 30/1.16) can't be expressed
by a finite sequence of decimal digits.

It comes out as

25.862068965517241379310344827[5862068965517241379310344827]...

where the digits in the brackets repeat infinitely.


--
Rick DeNatale

Blog: http://talklikeaduck.denhaven2.com/
Twitter: http://twitter.com/RickDeNatale
WWR: http://www.workingwithrails.com/person/9021-rick-denatale
LinkedIn: http://www.linkedin.com/in/rickdenatale
0eaa4f32363d027f0d905d36272d98c9?d=identicon&s=25 Samuel Lown (samlown)
on 2009-05-06 12:08
Rick Denatale wrote:
> On Tue, May 5, 2009 at 9:01 PM, Samuel Lown <me@samlown.com> wrote:
>>
>> Can anyone offer an explanation?
>
> BigDecimal is still a kind of float, albeit a decimal float rather
> than a binary float.
>

Many thanks Nobuyoshi and Rick, that makes complete sense now.

I guess the safest rule of thumb is if you're doing comparisons for
humans, use humanized numbers (i.e. integers :-)

Cheers, sam
3131fcea0a711e5ad89c8d49cc9253b4?d=identicon&s=25 Julian Leviston (Guest)
on 2009-05-06 12:11
(Received via mailing list)
Yeah I think they would be better off with rationals

Blog: http://random8.zenunit.com/
Learn: http://sensei.zenunit.com/
Twitter: http://twitter.com/random8r

On 06/05/2009, at 1:30 PM, Rick DeNatale <rick.denatale@gmail.com>
E0d864d9677f3c1482a20152b7cac0e2?d=identicon&s=25 Robert Klemme (Guest)
on 2009-05-06 13:07
(Received via mailing list)
2009/5/6 Samuel Lown <me@samlown.com>:
> Many thanks Nobuyoshi and Rick, that makes complete sense now.
>
> I guess the safest rule of thumb is if you're doing comparisons for
> humans, use humanized numbers (i.e. integers :-)

Or define a maximum error, i.e. do not rely on identical values but
rather the difference of the quotient.

Kind regards

robert
This topic is locked and can not be replied to.