Incorrect period index for Gregorian Dates at beginning of each CM_PERIOD
In Ruby 1.9-2.0 date_core, the year and chronological Julian day of a Date are stored as offsets from an indexed period relative to 0j, -4712-01-01 in the Julian calendar. Each period consists of exactly 194,796 Julian years, i.e. 194,800 Gregorian years.
Since 0j is -4713-11-24 in the proleptic Gregorian calendar, Gregorian calendar years do not align with these periods. A problem occurs when decode_years() uses the Gregorian calendar year of a Date in the the first 38 days of a period to assign the period index ("nth"). The calculation of nth ((y + 4712) / CM_PERIOD_GCY) is incorrect from November 24 until and including December 31. As a result of this error, an instance with period nth-1 is not equivalent to instances that represent the same Julian day, but which are constructed with an accurate period. These tests, which pass under Ruby 1.8, demonstrate the defect:
assert_equal(0, (Date.jd(0) <=> Date.civil(-4713, 11, 24, Date::GREGORIAN)))
assert_equal(0, (Date.jd(213447717) <=> Date.civil(579687, 11, 24)))
assert_equal(0, (Date.jd(-213447717) <=> Date.civil(-589113, 11, 24, Date::GREGORIAN)))
Attached is a patch against trunk which corrects this issue for methods which use decode_year() to assign the period index, including Date.civil, Date.ordinal and Date.commercial.
* ext/date/date_core.c: fixed a bug . reported by Riley Lynch. * ext/date/date_core.c: fixed coding error . reported by Riley Lynch.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/branches/ruby_2_0_0@41952 b2dd03c8-39d4-4d8f-98ff-823fe69b080e