2020-12-11 Alan E. Davis wrote:
>
> I had hoped that subtracting 10 hours from 06:44 UTC would get me at
> least -04:44.

I am in doubts how to present negative time correctly. Having in mind
wall clocks with hands, your expectation has some sense.

I think, the source of confusion is that you are trying to use
facilities intended for TimeInterval + TimeInterval arithmetic, where
TimeInterval denotes a type. Actually you need to compute Time +
TimeInterval. Sorry, I am unaware if Emacs provides proper functions.

There are a lot of subtle issues with time-related operations, but most
of DST complications pointed out by Tim could be handled with IANA
(Olson) DB for time zones, on Linux it is almost always installed as
tzdata package and is getting regular updates. It is rather precise and
contains history of DST and other transitions. If people complains on
incorrect results, usually they have wrong timezone set on their computers.

For astronomy the best representation should be timestamp (seconds since
epoch) converted to local date time when necessary. Unsure concerning
leap seconds.

Human-related future events mostly should be stored as date + local time
+ label of administrative region that allows to get time zone ID (namely
ID, not offset from UTC). Time offset could be changed after event has
been planned, time zone could be split into several ones with distinct
offsets.

- LocalTime + TimeInterval operation could give incorrect result unless
LocalTime is augmented with date and TimeZone, the latter is the history
of transitions.
- TimeZone (e.g. Europe/Berlin) as full transition history is not the
same as TimeOffset at particular moment (+0100).

Carefully choose storage format:
- just Date e.g. for date of birth (adding time or time zone could
distort results),
- Date + LocalTime + Place for planned events
- Timestamp for most events in past. Use it for future if local time is
irrelevant, astronomical events is an example.

Conversion to local time could evolve due to administrative decisions.

Examples when Olson DB is not enough and additional negotiations or
justifications are necessary:

- (March, 31) - (1 month)
- Conversion from Date + LocalTime to Timestamp around time transition
when non-existing or ambiguous local time is possible.