Bug #12805: MRI BigDecimal#div produces grossly wrong result

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-linux]
• Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN

[felixvf (Felix von Ferey)]

Expected

(It works in JRuby-9.1.5.0)

# ruby -e 'require "bigdecimal"; (1...50).each do |i| puts (BigDecimal.new("1").div(BigDecimal.new("0.22E0"),i)).to_s; end'
0.5E1
0.45E1
0.455E1
0.4545E1
0.45455E1
0.454545E1
0.4545455E1
0.45454545E1
0.454545455E1
0.4545454545E1
0.45454545455E1
0.454545454545E1
0.4545454545455E1
0.45454545454545E1
0.454545454545455E1
0.4545454545454545E1
0.45454545454545455E1
0.454545454545454545E1
0.4545454545454545455E1
0.45454545454545454545E1
0.454545454545454545455E1
0.4545454545454545454545E1
0.45454545454545454545455E1
0.454545454545454545454545E1
0.4545454545454545454545455E1
0.45454545454545454545454545E1
0.454545454545454545454545455E1
0.4545454545454545454545454545E1
0.45454545454545454545454545455E1
0.454545454545454545454545454545E1
0.4545454545454545454545454545455E1
0.45454545454545454545454545454545E1
0.454545454545454545454545454545455E1
0.4545454545454545454545454545454545E1
0.45454545454545454545454545454545455E1
0.454545454545454545454545454545454545E1
0.4545454545454545454545454545454545455E1
0.45454545454545454545454545454545454545E1
0.454545454545454545454545454545454545455E1
0.4545454545454545454545454545454545454545E1
0.45454545454545454545454545454545454545455E1
0.454545454545454545454545454545454545454545E1
0.4545454545454545454545454545454545454545455E1
0.45454545454545454545454545454545454545454545E1
0.454545454545454545454545454545454545454545455E1
0.4545454545454545454545454545454545454545454545E1
0.45454545454545454545454545454545454545454545455E1
0.454545454545454545454545454545454545454545454545E1
0.4545454545454545454545454545454545454545454545455E1

Actual

(It does not work in MRI-2.3.1p112)

# ruby -e 'require "bigdecimal"; (1...50).each do |i| puts (BigDecimal.new("1").div(BigDecimal.new("0.22E0"),i)).to_s; end'
0.4E1
0.4E1
0.4E1
0.4E1
0.4E1
0.4E1
0.4E1
0.4E1
0.4E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454E1
0.4545454545454545454545454545454545454545454545E1
0.4545454545454545454545454545454545454545454545E1
0.4545454545454545454545454545454545454545454545E1
0.4545454545454545454545454545454545454545454545E1

Additional observations

Note that dividing by "0.22" shows the bug, but dividing by "2.2" does not show the bug:

# ruby -e 'require "bigdecimal"; (1...50).each do |i| puts (BigDecimal.new("1").div(BigDecimal.new("0.22E1"),i)).to_s; end'
0.5E0
0.45E0
0.455E0
0.4545E0
0.45455E0
0.454545E0
0.4545455E0
0.45454545E0
0.454545454E0
0.4545454545E0
0.45454545455E0
0.454545454545E0
0.4545454545455E0
0.45454545454545E0
0.454545454545455E0
0.4545454545454545E0
0.45454545454545455E0
0.454545454545454545E0
0.4545454545454545455E0
0.45454545454545454545E0
0.454545454545454545455E0
0.4545454545454545454545E0
0.45454545454545454545455E0
0.454545454545454545454545E0
0.4545454545454545454545455E0
0.45454545454545454545454545E0
0.454545454545454545454545454E0
0.4545454545454545454545454545E0
0.45454545454545454545454545455E0
0.454545454545454545454545454545E0
0.4545454545454545454545454545455E0
0.45454545454545454545454545454545E0
0.454545454545454545454545454545455E0
0.4545454545454545454545454545454545E0
0.45454545454545454545454545454545455E0
0.454545454545454545454545454545454545E0
0.4545454545454545454545454545454545455E0
0.45454545454545454545454545454545454545E0
0.454545454545454545454545454545454545455E0
0.4545454545454545454545454545454545454545E0
0.45454545454545454545454545454545454545455E0
0.454545454545454545454545454545454545454545E0
0.4545454545454545454545454545454545454545455E0
0.45454545454545454545454545454545454545454545E0
0.454545454545454545454545454545454545454545454E0
0.4545454545454545454545454545454545454545454545E0
0.45454545454545454545454545454545454545454545455E0
0.454545454545454545454545454545454545454545454545E0
0.4545454545454545454545454545454545454545454545455E0

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset r56336.

---

bigdecimal.c: more precision

• ext/bigdecimal/bigdecimal.c (BigDecimal_div2): need more room
  for precision to round. [ruby-core:77475] [Bug #12805]