> > Oh yeah. We don't have a date_part(units, time) function defined, so it
> > is getting converted to interval (which in other contexts *does* have
> > some usefulness as a "time equivalent").
> You're going to have an extremely hard time convincing me of that.
OK, thanks for the warning. I'll try later when I have more time...
> > We could fairly easily define a date_part() for the time and timetz data
> > types.
> I had figured that time would be cast to timestamp. Which is probably
> what it used to do.
Tom Lane pointed out the problem of inferring an appropriate date for
the upcast.
- Thomas