Forum: Ruby-core [ruby-trunk - Bug #7458][Open] Depending on value of BigDecimal::limit, addition can give inaccurate

Issue #7458 has been reported by bpow (Bradford Powell).

----------------------------------------
Bug #7458: Depending on value of BigDecimal::limit, addition can give 
inaccurate results
https://bugs.ruby-lang.org/issues/7458

Author: bpow (Bradford Powell)
Status: Open
Priority: Normal
Assignee:
Category:
Target version:
ruby -v: 2.0.0dev


=begin
This bug is present in ruby-trunk and also in 1.8.7 and 1.9.3p125.

I apologize that I cannot figure out what exactly is going wrong and how 
to patch, but at least I have a small reproducible test case.

When adding very small numbers (like 7E-19) to 1, the result should be 
very close to 1. However, depending on the limit to BigDecimal's 
precision, this is not always what is resulted, as this script 
demonstrates:


 require 'bigdecimal'

 one = BigDecimal('1')
 smallnum = BigDecimal('7E-19')
 0.upto(20) do |lim|
   BigDecimal::limit(lim)
   result = one + smallnum
   puts "result is #{result}, as float is #{result.to_f} with limit 
#{lim}"
 end

When run in trunk (or 1.8.7), the results it gives are:

 result is 0.10000000000000000007E1, as float is 1.0 with limit 0
 result is 0.1E1, as float is 1.0 with limit 1
 result is 0.1E1, as float is 1.0 with limit 2
 result is 0.1E1, as float is 1.0 with limit 3
 result is 0.1E1, as float is 1.0 with limit 4
 result is 0.1E1, as float is 1.0 with limit 5
 result is 0.1E1, as float is 1.0 with limit 6
 result is 0.1E1, as float is 1.0 with limit 7
 result is 0.1E1, as float is 1.0 with limit 8
 result is 0.1E1, as float is 1.0 with limit 9
 result is 0.2E1, as float is 2.0 with limit 10
 result is 0.2E1, as float is 2.0 with limit 11
 result is 0.2E1, as float is 2.0 with limit 12
 result is 0.2E1, as float is 2.0 with limit 13
 result is 0.2E1, as float is 2.0 with limit 14
 result is 0.2E1, as float is 2.0 with limit 15
 result is 0.2E1, as float is 2.0 with limit 16
 result is 0.2E1, as float is 2.0 with limit 17
 result is 0.2E1, as float is 2.0 with limit 18
 result is 0.1000000000000000001E1, as float is 1.0 with limit 19
 result is 0.10000000000000000007E1, as float is 1.0 with limit 20

So the results with limit <=9 or >=19 are as expected, but in between 
the result returned is equal to 2.0

A test that demonstrates this (suitable for inclusion in 
(({test/bigdecimal/test_bigdecimal.rb})) is:

 def test_limited_one_plus_small
    one = BigDecimal('1')
    smallnum = BigDecimal('0.7E-18')
    BigDecimal::save_limit do
      1.upto(20) do |lim|
        BigDecimal::limit(lim)
        result = (one + smallnum).to_f.to_s
        assert_equal("1.0", result, "Error adding a small number to one 
with BigDecimal::limit(#{lim})")
      end
    end
 end

I think this is different from #7404 (from what I can tell by google 
translate), but I apologize for the duplicate if not.
=end

Issue #7458 has been updated by mame (Yusuke Endoh).

Target version changed from 2.0.0 to next minor


----------------------------------------
Bug #7458: Depending on value of BigDecimal::limit, addition can give 
inaccurate results
https://bugs.ruby-lang.org/issues/7458#change-36546

Author: bpow (Bradford Powell)
Status: Assigned
Priority: Normal
Assignee: mrkn (Kenta Murata)
Category: ext
Target version: next minor
ruby -v: 2.0.0dev


=begin
This bug is present in ruby-trunk and also in 1.8.7 and 1.9.3p125.

I apologize that I cannot figure out what exactly is going wrong and how 
to patch, but at least I have a small reproducible test case.

When adding very small numbers (like 7E-19) to 1, the result should be 
very close to 1. However, depending on the limit to BigDecimal's 
precision, this is not always what is resulted, as this script 
demonstrates:


 require 'bigdecimal'

 one = BigDecimal('1')
 smallnum = BigDecimal('7E-19')
 0.upto(20) do |lim|
   BigDecimal::limit(lim)
   result = one + smallnum
   puts "result is #{result}, as float is #{result.to_f} with limit 
#{lim}"
 end

When run in trunk (or 1.8.7), the results it gives are:

 result is 0.10000000000000000007E1, as float is 1.0 with limit 0
 result is 0.1E1, as float is 1.0 with limit 1
 result is 0.1E1, as float is 1.0 with limit 2
 result is 0.1E1, as float is 1.0 with limit 3
 result is 0.1E1, as float is 1.0 with limit 4
 result is 0.1E1, as float is 1.0 with limit 5
 result is 0.1E1, as float is 1.0 with limit 6
 result is 0.1E1, as float is 1.0 with limit 7
 result is 0.1E1, as float is 1.0 with limit 8
 result is 0.1E1, as float is 1.0 with limit 9
 result is 0.2E1, as float is 2.0 with limit 10
 result is 0.2E1, as float is 2.0 with limit 11
 result is 0.2E1, as float is 2.0 with limit 12
 result is 0.2E1, as float is 2.0 with limit 13
 result is 0.2E1, as float is 2.0 with limit 14
 result is 0.2E1, as float is 2.0 with limit 15
 result is 0.2E1, as float is 2.0 with limit 16
 result is 0.2E1, as float is 2.0 with limit 17
 result is 0.2E1, as float is 2.0 with limit 18
 result is 0.1000000000000000001E1, as float is 1.0 with limit 19
 result is 0.10000000000000000007E1, as float is 1.0 with limit 20

So the results with limit <=9 or >=19 are as expected, but in between 
the result returned is equal to 2.0

A test that demonstrates this (suitable for inclusion in 
(({test/bigdecimal/test_bigdecimal.rb})) is:

 def test_limited_one_plus_small
    one = BigDecimal('1')
    smallnum = BigDecimal('0.7E-18')
    BigDecimal::save_limit do
      1.upto(20) do |lim|
        BigDecimal::limit(lim)
        result = (one + smallnum).to_f.to_s
        assert_equal("1.0", result, "Error adding a small number to one 
with BigDecimal::limit(#{lim})")
      end
    end
 end

I think this is different from #7404 (from what I can tell by google 
translate), but I apologize for the duplicate if not.
=end