Re: The Emacs Calculator and calendar

[Paul Eggert <eggert@cs.ucla.edu>]

On 10/07/2012 06:57 AM, Stefan Monnier wrote:

> To the extent possible, Emacs could let the user specify the calendar
> indirectly by instead specifying a "context" (a place, plus whatever
> else is needed to resolve ambiguities), and then let Emacs figure out
> which calendar was used at that time in that context.

Yes, for example Emacs could examine (say) the TZ variable
plus optional extra info.  This would work in theory, but in
practice there would be many problems.

For example, I live the Los Angeles area, and presumably
Emacs would infer its calendrical behavior from my TZ
setting 'America/Los_Angeles'.  But what behavior would that
be, exactly?  To help answer that, here's L.A.'s calendrical
history as best I know:

  Settled by Tongva and Chumash thousands of years ago;
  exact years not known.  These people used calendars, which
  most likely did not agree with each other and varied with
  time, but the details are not known.

  Area first visited by Europeans in 1742.  The Cabrillo
  Expedition anchored in San Pedro and Santa Monica bays for
  one day each, then left and never returned.

  Area visited again in 1602 by the Vizcaíno Expedition, which
  also anchored for a couple of days and then moved on.

  Next known visit by the Portolá Expedition of 1769.  These
  are the first Europeans who are known to have set foot
  near what became Los Angeles downtown, although there are
  rumors of other visits before then.

  City officially founded 4 Sept 1781 (Gregorian).

Given the above, there are several problems in deciding how
Emacs should behave for TZ='America/Los_Angeles':

  Would its calendar change from "unknown" status to Old
  Spanish status in 1542 and then switch to Emacs Julian in
  1556 when Spain switched to Julian and then switch to
  Gregorian in 1582, all because Spanish explorers' ships
  dropped anchor nearby for a couple of days in 1542 and 1602?

  Or should the Los Angeles entry stay "unknown" until 1781
  because the city didn't exist until then?

  Or should it do something else?

No answer is satisfactory here -- whatever we'd put into the
table would be wrong for some common uses.

And Los Angeles is one of the *easy* cases.  There are
hundreds of other locations to do, many of them much harder
than Los Angeles, where we'd have worse problems, some
technical and some political.

Some other questions would come up too.  For instance:

  What do we do when a calendar is partly known, but not
  completely, as is the case for the Chumash calendar,
  or for Julius Caesar's Julian calendar?

  Should Emacs distinguish between "unknown" (that is, there
  was a calendar but we don't know what it was exactly, as
  in Los Angeles circa 1700) and "none" (that is, the area
  was uninhabited and had no calendar, as in Los Angeles
  circa 15,000 BC)?

  We could make some simplifying assumptions, e.g., use the
  Gregorian calendar when the actual calendar isn't fully
  known, but how would this be reflected to the user?  And
  if we're going to do that, why not just use Gregorian
  everywhere, as that's simpler?

I hope this helps to explain why adding a calendrical/locale
database would be a big project, and why any attempts to
build such a thing would run the risk hurting users (by
giving them wrong or misleading answers) as much as help
them.