Daylight savings time: Not that easy
The W3C working group note Working with Time Zones correctly identifes the problem that local time in a time zone with "daylight savings time" part of the year can not be specified by offset from UTC, because that offset is not constant. So the time zone must be identified by a TZID.
They suggest that two <datetime, TZID> pair can be compared,
by computing the offsets on that date and at those times, using the TZ database.
It is not that easy. Each spring there is an hour that does not exist (and all trains are suddenly one hour late). That is not a computational problem, because a datetime is still unique within its time zone.
But each fall there is an hour that appears twice (causing all the trains to be one hour delayed). No time zone database can possibly tell if a datetime falling within that hour is the first or the second appearance of that datetime.
So I see no other solution than this: Don't mess with time (leave that to Hermione Granger) but keep the offset from UTC constant in each time zone. If daylight savings time must be used, use it all year round. Anyone suggesting other solutions, should provide a computational method to convert local time to UTC.
If rational arguments are not enough, let me remind you that daylight savings time was introduced in Norway during World War II by Vidkun Quisling.