From: Tom Lane <tgl@sss.pgh.pa.us>
To: =?utf-8?Q?Dagfinn_Ilmari_Manns=C3=A5ker?= <ilmari@ilmari.org>
cc: Matthias van de Meent <boekewurm+postgres@gmail.com>,
        Artur Formella <artur.formella3@gmail.com>,
        pgsql-hackers@postgresql.org
Subject: Re: Allowing additional commas between columns, and at the end of the SELECT clause
In-reply-to: <877cfxhdpd.fsf@wibble.ilmari.org>
References: <0368c60f-abe2-4f5f-972d-7cd1e6db2382@gmail.com> <CAEze2WhL_A9QO1xHQnEoYN1g5wsigqrL2+FLELcjFwaA-nP-Ug@mail.gmail.com> <87bk59hp5o.fsf@wibble.ilmari.org> <738732.1715609494@sss.pgh.pa.us> <877cfxhdpd.fsf@wibble.ilmari.org>
Comments: In-reply-to =?utf-8?Q?Dagfinn_Ilmari_Manns=C3=A5ker?= <ilmari@ilmari.org>
	message dated "Mon, 13 May 2024 17:35:42 +0100"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <762524.1715620475.1@sss.pgh.pa.us>
Date: Mon, 13 May 2024 13:14:35 -0400
Message-ID: <762525.1715620475@sss.pgh.pa.us>
Archived-At: <https://www.postgresql.org/message-id/762525.1715620475%40sss.pgh.pa.us>
Precedence: bulk

=?utf-8?Q?Dagfinn_Ilmari_Manns=C3=A5ker?= <ilmari@ilmari.org> writes:
> Tom Lane <tgl@sss.pgh.pa.us> writes:
>> I'm fairly down on this idea for SQL, because I think it creates
>> ambiguity for the ROW() constructor syntax.  That is:
>> 	(x,y) is understood to be shorthand for ROW(x,y)
>> 	(x) is not ROW(x), it's just x
>> 	(x,) means what?

> Python has a similar issue: (x, y) is a tuple, but (x) is just x, and
> they use the trailing comma to disambiguate, so (x,) creates a
> single-item tuple.  AFAIK it's the only place where the trailing comma
> is significant.

Ugh :-(.  The semantic principle I'd prefer to have here is "a trailing
comma is ignored", but what they did breaks that.  But then again,
I'm not particularly a fan of anything about Python's syntax.

> Yeah, a more principled approach would be to not special-case target
> lists, but to allow one (and only one) trailing comma everywhere:
> select, order by, group by, array constructors, row constructors,
> everything that looks like a function call, etc.

If it can be made to work everywhere, that would get my vote.
I'm not sure if any other ambiguities arise, though.  SQL has
a lot of weird syntax corners (and the committee keeps adding
more :-().

			regards, tom lane