Re: BUG #19053: Inconsistent arithmetic regarding TIMESTAMPTZ and INTERVAL

On Mon, 15 Sep 2025 11:41:29 -0400
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> AFAICS all of these are behaving as-expected.

Well, I would say each calculation by itself can be somewhat explained.
But it's the overall combination of behavior that seems inconsistent.

In particular:

TIMESTAMPTZ '2025-10-27' - TIMESTAMPTZ '2025-10-26'

This gives INTERVAL '1 day 01:00:00', which makes sense in some way.
However, knowing that '24 hours' are not the same as '1 day', I would
rather expect either '25 hours' or '1 day'.

The following does actually make sense:

TIMESTAMPTZ '2025-10-26' + INTERVAL '1 day 01:00:00'

But it's the previous substraction that, given how INTERVAL behaves
here, does not make sense.

> Yeah, it's confusing,
> but expecting calendar calculations to have mathematical rigor is
> a fool's errand.

I don't expect mathematical rigor, but I would like to have consistent
semantics of what INTERVAL '1 day' means, and when it is used and when
it is returned.

> The intent of what's implemented is to produce
> useful results for calculations like
>
> regression=# SELECT TIMESTAMPTZ '2025-03-30' + INTERVAL '1 day';
> ?column?
> ------------------------
> 2025-03-31 00:00:00+02
> (1 row)
>
> despite the intervening DST change.

Yes, I agree that the addition behaves correctly. I just wonder if the
substraction would need to return either '1 day' or '25 hours'. But not
'1 day 01:00:00'.

> regards, tom lane

Regards,
Jan Behrens