Bug #12878: 0.49999999999999994.round is 1 and not 0 on Windows 64

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.3.1p112 (2016-04-26 revision 54768) [x64-mingw32]
• Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN

[Eregon (Benoit Daloze)]

0.49999999999999994.round produces 1 on this platform, while all other platforms seem to produce the expected 0 (round-to-nearest).
Maybe this is due to some slightly different floating-point representation?

See https://ci.appveyor.com/project/eregon/spec-x948i/build/190/job/jk8eb1u2kq6npwn4 which reproduced it in a run of ruby/spec.

[shyouhei (Shyouhei Urabe)]

JFYI if we stick to 64bit representation, 0.49999999999999994 is 0011111111011111111111111111111111111111111111111111111111111111.

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset r56534.

---

configure.in: no round in x64-mingw

• configure.in (ac_cv_func_round): round(3) in x86_64-w64-mingw32
  is not accurate in an edge case. [ruby-core:77794] [Bug #12878]