Oh dear. The following is an excerpt from a proposal that I initially
posted to the internal WWP list in Oct 96, then again with the note
"I want to post this to TEI-TECH soon" in Aug 97. But it looks like I
never actually did post it to TEI-TECH. Sigh.

| B.C. dates
| ---- -----
| The following is a quotation from "The ISO Date Format" by Peter
| Meyer
| (
|    The year is expressed in the manner of the astronomers. Year 0 is
|    the year preceding year 1 (a.k.a. 1 B.C.). Year -1 is the year
|    preceding year 0 (a.k.a. 2 B.C.). Generally, year -n is the year
|    which began n+1 years before the beginning of year 1 (so year -n
|    = year n+1 B.C.). With the use of this astronomical system of
|    year-numbering there are no limits to the dates which can be
|    expressed in this date format.[3]
| I think that the only way to make this work is to be consistent
| about using 4 digits for the year. Thus the Julian calendar was
| adopted in <date calendar="Gregorian" value="-0045">46 B.C.</date>.[6]
| Notes
| -----
| [3] However, since Mr. Meyer incorrectly says that ISO 8601 applies
|     to Julian dates as well as Gregorian, we need to verify this
|     information elsewhere.
| [6] The correct term for the calendar being used is "Gregorian
|     proleptic", because we are extending the system to a date before
|     it was invented. I see no reason for the WWP to differentiate
|     between proleptic and normal uses.

> If I sort these (using 'sort' on my Linux machine), I get:

In general, I would not use the capability to use cheap-and-dirty
tools (like an alphabetic sort) as the determining factor in the
format of your data. (Perhaps it should influence your decision, but
should not be the most important thing by far.) In this case, though,
won't a cheap-and-easy tool, i.e. numeric sort, do the right thing if
you first remove the internal hyphens? I.e.
  perl -pe's/([0-9])-/\1/g;' < date_list | sort -n > sorted_date_list

Furthermore I think that using the above suggested "astronomical"
version of the proleptic dates will again make the "date arithmetic"
you were using work, no? (But realize, of course, that such "date
arithmetic" is a bit tricky. E.g., the Members' meeting starts on 22
Oct, 169 days from today. (If I have the date right :-). But
which needs to be interpreted as "5 months and 16 days"
(P0000-05-16), not "516 days". Even so, some might argue that, using
an average length month, it should really be 5 months and 17 days.