Greg Stark <gsstark(at)mit(dot)edu> writes:
> But that still seems a bit more complex than ideal. Would it be
> reasonable to have a tinterval() constructor which takes timestamptz
> data types?
No, because that would be encouraging people to use tinterval ;-).
That type needs to die.
If Jeff doesn't finish his range-type stuff soon, maybe a tinterval
replacement that hasn't got a Y2038 problem would be in order. But
we shouldn't put any more effort into tinterval as such.
(BTW, tinterval hasn't got a gist opclass either, so the constructor
is the least of the missing pieces here.)
regards, tom lane