Re: [icu-support] Universal Time Scale: Julian vs. Gregorian until 1582?

[Markus S. <mar...@gm...>]

Thanks, Eric. What I found out is quite simple:

- .Net uses proleptic time. As they write [1],
  "Time values are measured in 100-nanosecond units called ticks, and
a particular date is the number of ticks since 12:00 midnight, January
1, 0001 A.D. (C.E.) in the GregorianCalendar calendar."
  and it looks like they mean Gregorian for all dates past and future.

- By default, an ICU Calendar is a GregorianCalendar with a
  change date of October 15, 1582, that is, it's a Julian calendar
before that. [2]

- There are 2 more days in the ICU (Julian/Gregorian) calendar between
0001 and 1582 than in a purely (proleptic) Gregorian one: ICU has to
skip 10 days 1582-Oct-5..14 (inclusive) but add 12 leap days for the
non-multiple-of-400 centuries in between.

- ICU's Universal Time Scale is supposed to match .Net's
System.DateTime. [3] It's off by 2 days.

As a fix, I think EpochOffsets.java needs to be modified by replacing
  Calendar cal = Calendar.getInstance(utc, Locale.ENGLISH);
with
  // use a proleptic Gregorian calendar for 0001AD and later by setting
  // the Gregorian change date before 0001AD
  // with a value that is safely before that date by any measure, i.e.,
  // more than 719164 days before 1970
  long before0001AD = -750000 * 86400 * 1000;
  Calendar cal = new GregorianCalendar(utc,
Locale.ENGLISH).setGregorianChange(new Date(before0001AD));

>From the looks of it, once this code is fixed, the trio of tools in
com.ibm.icu.dev.tool.timescale needs to be re-run, and the results
placed into the library source code, and we should be fine. Right?

markus

[1] .Net System.DateTime documentation:
http://msdn2.microsoft.com/en-us/library/03ybds8y.aspx
[2] ICU4J GregorianCalendar API ref:
http://icu.sourceforge.net/apiref/icu4j/com/ibm/icu/util/GregorianCalendar.html
[3] http://icu.sourceforge.net/userguide/universalTimeScale.html

On 6/6/06, Eric Mader <em...@ic...> wrote:
> The values in utmscale.c were generated using the following tool:
> package com.ibm.icu.dev.tool.timescale;
> public class EpochOffsets

-- 
Opinions expressed here may not reflect my company's positions unless
otherwise noted.