Bug #13279: Math.log10 accuracy regression

• Status: Closed
• Assignee: -
• Target version: -
• Backport: 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE

[jgoizueta (Javier Goizueta)]

I've notice a degradation in the precision of Math.log10 in Ruby 2.4 with respect to 2.1:

2.1.2 :136 > Math.log10 10**1000
 => 1000.0

2.4.0 :009 > Math.log10 10**1000
 => 999.9999999999999

The problem has been introduced in this commit: git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@55641 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
(https://github.com/ruby/ruby/commit/fd2c613b1cec46d0b02df722bb00ae01154d5dc3)

The use of arithmetic to compute the log10(2) introduces some roundoff error in that constant:

2.1.2 :140 > M_LN2 = 0.693147180559945309417232121458176568
 => 0.6931471805599453
2.1.2 :141 > M_LN10 = 2.30258509299404568401799145468436421
 => 2.302585092994046
2.1.2 :142 > l2 == M_LN2/M_LN10
 => false
2.1.2 :144 > l2
 => 0.3010299956639812
2.1.2 :145 > M_LN2/M_LN10
 => 0.30102999566398114

• The correct value of the constant (base-10 log of 2) is 0.301029995663981195213738894724...
• The log10(2) function yields a correctly rounded approximation of 0.3010299956639812
• But the computed division of constants results in 0.30102999566398114

[nobu (Nobuyoshi Nakada)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN to 2.2: DONTNEED, 2.3: DONTNEED, 2.4: REQUIRED

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset r57779.

---

math.c: more accuracy

• math.c (math_log10): calculate log10(2) for more accuracy.
  [ruby-core:79907] [Bug #13279]

[naruse (Yui NARUSE)]

• Backport changed from 2.2: DONTNEED, 2.3: DONTNEED, 2.4: REQUIRED to 2.2: DONTNEED, 2.3: DONTNEED, 2.4: DONE

ruby_2_4 r57928 merged revision(s) 57778,57779.