Have you considered doing a survey of other Date/Time/Calendar
implementations to find out how they solve this problem? For example, a
lot of careful thought and research has gone into the Perl DateTime
modules.
http://search.cpan.org/dist/DateTime/lib/DateTime.pm#Adding_a_Duration_to_a_Datetime
Very interesting. Thanks for the hint! :)
Yes, nice!
It seems that the perl modules uses the same approach I was
suggesting:
"DateTime.pm always adds (or subtracts) days, then months, minutes,
and then seconds and nanoseconds. If there are any boundary
overflows, these are normalized at each step. For the days and
months (the calendar) the local (not UTC) values are used. For
minutes and seconds, the local values are used. This generally just
works.
This means that adding one month and one day to February 28, 2003
will produce the date April 1, 2003, not March 29, 2003."
Ok, so let's follow this approach. I will explain it in detail in the
API reference.
