Re: Improve timestamp substraction to be DST-aware

Piotr Marcinczyk <pmarcinc(at)gmail(dot)com> writes:
> I would like to implement item from TODO list: "Improve TIMESTAMP WITH
> TIME ZONE subtraction to be DST-aware".

> To compute interval properly, we need time zone. Currently in timestamp
> we don't have time zone information, so I would use actual session time
> zone for computations. This can cause problems in the future, when TZ
> info will be saved in DB (this is also on TODO list). In such case, user
> may suspect, that stored TZ will be used, and probably behavior of
> subtraction will change. Anyway, I think, that using session TZ now is
> better than not using it at all.

TBH, that doesn't sound like an especially great idea from here. The
value of a timestamp difference is currently exact, but you're proposing
to make it fuzzy and context-dependent.

The alternative proposal that's been on the table for awhile (see the
preceding entry in the TODO list) is to remove the interval_justify_hours
call in timestamp_mi, which would also have the effect of fixing the
inconsistency that T1 + (T2 - T1) doesn't necessarily yield T2. And it
would do that a lot more straightforwardly, with less risk that there's
still corner cases that would misbehave.

If it's not the T1 + (T2 - T1) issue that's bothering you, perhaps
you should explain exactly what results you're hoping to get by changing
this behavior.

regards, tom lane