Bug #10512: Time.at produces imprecise nsec values

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: ruby 2.2.0preview1 (2014-09-17 trunk 47616) [x86_64-linux]
• Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN

[cheald (Chris Heald)]

I'm working on improving JRuby parity with MRI, and I've found what I think is a bug in MRI and its test suite. test_time.rb, in test_at asserts:

assert_equal(199, Time.at(0.0000002).nsec)

This passes under MRI, but fails under JRuby, which produces (the correct) value of 200.

It looks like there is a precision issue in MRI which is causing this behavior:

2.2.0-preview1 :008 > Time.at(0.000000197).nsec
=> 197
2.2.0-preview1 :009 > Time.at(0.000000198).nsec
=> 197
2.2.0-preview1 :010 > Time.at(0.000000199).nsec
=> 198
2.2.0-preview1 :011 > Time.at(0.000000200).nsec
=> 199
2.2.0-preview1 :012 > Time.at(0.000000201).nsec
=> 201

First and foremost, why does MRI assert that Time.at(0.0000002).nsec == 199, rather than 200? Second, if this is not a bug, why not (so that I can make sure JRuby performs equivalently)?

[akr (Akira Tanaka)]

• Status changed from Open to Rejected

It is because the float value 0.0000002 is less than 0.0000002.

% ruby -e 'p((Rational(0.0000002) * 1000000000).floor)'
199

Time records 0.0000002 as Rational(0.0000002).
It is the most precious record according to the given argument.

% ruby -e 'p Rational(0.0000002), Time.at(0.0000002).subsec == Rational(0.0000002)' 
(944473296573929/4722366482869645213696)
true