From: Tom Lane <tgl@sss.pgh.pa.us>
To: "David G. Johnston" <david.g.johnston@gmail.com>
cc: "rekgrpth@gmail.com" <rekgrpth@gmail.com>,
        "pgsql-bugs@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Subject: Re: BUG #18307: system columns does not support using join
In-reply-to: 
 <CAKFQuwbpw_WhcB=w_MwUWjMGVYH=Hxr3bW0vOpbLH8QFMsgqxg@mail.gmail.com>
References: <18307-16a12cbeb212d742@postgresql.org>
 <CAKFQuwbpw_WhcB=w_MwUWjMGVYH=Hxr3bW0vOpbLH8QFMsgqxg@mail.gmail.com>
Comments: In-reply-to "David G. Johnston" <david.g.johnston@gmail.com>
	message dated "Wed, 24 Jan 2024 07:26:50 -0700"
MIME-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-ID: <264095.1706154850.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Wed, 24 Jan 2024 22:54:10 -0500
Message-ID: <264096.1706154850@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/264096.1706154850%40sss.pgh.pa.us>
Precedence: bulk

"David G. Johnston" <david.g.johnston@gmail.com> writes:
> On Tuesday, January 23, 2024, PG Bug reporting form <noreply@postgresql.=
org>
> wrote:
>> create table t(i int);
>> =

>> explain (costs off) select * from t join t tt on t.xmin =3D tt.xmin;
>> QUERY PLAN
>> ---------------------------------
>> Hash Join
>> Hash Cond: (t.xmin =3D tt.xmin)
>> ->  Seq Scan on t
>> ->  Hash
>> ->  Seq Scan on t tt
>> (5 rows)
>> =

>> explain (costs off) select * from t join t tt using (xmin);
>> ERROR:  column "xmin" specified in USING clause does not exist in left
>> table

> I don=E2=80=99t this being worth the effort to change, and really seems =
like
> completely expected behavior. =E2=80=9CSelect *=E2=80=9D doesn=E2=80=99t=
 output xmin, it requires
> explicit table qualification to see it.  This is the same thing.

Well, it is odd that "using (xmin)" isn't equivalent to the allegedly
equivalent "on t.xmin =3D tt.xmin".  This is down to the infrastructure
in transformFromClauseItem(), which searches the lists of (regular,
non-system) relation output column names to expand USING().  But like
you, I can't get excited about changing it.  There are a couple of
practical reasons why not:

* NATURAL JOIN is defined in terms of USING.  But we *certainly* don't
want "x NATURAL JOIN y" deciding that it should equate all the system
columns of x to those of y.  So there's going to be inconsistency at
one level or the other no matter what.

* I really find it hard to imagine a valid use case for joining on any
system column.  There are use-cases for joining on TID in an UPDATE
involving a self-join to the target table; but you can't write that
with JOIN USING syntax.

			regards, tom lane