public inbox for [email protected]  

help / color / mirror / Atom feed

From: Viliam Ďurina <[email protected]>
To: Tom Lane <[email protected]>
Cc: Aleksander Alekseev <[email protected]>
Cc: [email protected]
Subject: Re: MIN/MAX functions for a record
Date: Fri, 22 Mar 2024 16:50:01 +0100
Message-ID: <CAO=iB8Lx6R5mYdL4wfcx-0h6RfPD7SjR_sfu8nB3ubbXPcKELA@mail.gmail.com> (raw)
In-Reply-To: <[email protected]>
References: <CAO=iB8L4WYSNxCJ8GURRjQsrXEQ2-zn3FiCsh2LMqvWq2WcONg@mail.gmail.com>
	<CAJ7c6TNpEfJJSrJvQJmVT5z8=8cPaq-JHNY3ZbmP63rif5Kiow@mail.gmail.com>
	<[email protected]>

Exactly Tom, I see no fundamental problem for it not to be implemented,
since comparison operator is already implemented. In fact, MIN/MAX should
work for all types for which comparison operator is defined.

Regarding index support, there should not be an issue if the index is
defined for the record (e.g. `CREATE INDEX ON my_table(ROW(field_a,
field_b))`). However such indexes seem not to be supported. Whether a
composite index is compatible with a record created on the indexed fields
in every edge case I'm not sure...

Alexander, rewriting the year-month example is easy, but how would you
rewrite this query?

CREATE TABLE events(event_time TIMESTAMP, message VARCHAR, user_id VARCHAR);

You want a newest message for each user. It's easy with MAX(record):

SELECT user_id, MAX(ROW(event_time, message)).message
FROM events
GROUP BY user_id;

One option is to rewrite to a subquery with LIMIT 1

SELECT user_id, (SELECT message FROM events e2 WHERE e1.user_id=e2.user_id
ORDER BY event_time DESC LIMIT 1)
FROM events e1
GROUP BY user_id;

If your query already has multiple levels of grouping, multiple joins,
UNIONs etc., it gets much more complex. I also wonder if the optimizer
would pick the same plan as it would be if the MAX(record) is supported.

Viliam

On Fri, Mar 22, 2024 at 4:12 PM Tom Lane <[email protected]> wrote:

> Aleksander Alekseev <[email protected]> writes:
> >> In my queries I often need to do MIN/MAX for tuples, for example:
> >> SELECT MAX(row(year, month))
> >> FROM (VALUES(2025, 1), (2024,2)) x(year, month);
> >> This query throws:
> >> ERROR: function max(record) does not exist
> >> Was this ever discussed or is there something preventing the
> implementation?
>
> > I believe it would be challenging to implement max(record) that would
> > work reasonably well in a general case.
>
> As long as you define it as "works the same way record comparison
> does", ie base it on record_cmp(), I don't think it would be much
> more than a finger exercise [*].  And why would you want it to act
> any differently from record_cmp()?  Those semantics have been
> established for a long time.
>
>                         regards, tom lane
>
> [*] Although conceivably there are some challenges in getting
> record_cmp's caching logic to work in the context of an aggregate.
>

view thread (3+ messages)

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected], [email protected], [email protected]
  Subject: Re: MIN/MAX functions for a record
  In-Reply-To: <CAO=iB8Lx6R5mYdL4wfcx-0h6RfPD7SjR_sfu8nB3ubbXPcKELA@mail.gmail.com>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox