Steve Crawford <scrawford@pinpointresearch.com> writes:
> But PG doesn't follow the same rules in subtracting timestamptz values
> so operations involving timestamps and intervals are (sometimes) not
> reversible:
Yeah. timestamp_mi is performing a justify_hours call, which it should
not, but removing that call changes a lot of the regression test
outputs. So we've been afraid to change it. You can find more about
that in the archives.
regards, tom lane