MIME-Version: 1.0
In-Reply-To: <49d5cbd1-7e47-4740-bc81-e574150f44cb@eisentraut.org>
References: 
 <CAHM0NXjz0kDwtzoe-fnHAqPB1qA8_VJN0XAmCgUZ+iPnvP5LbA@mail.gmail.com>
 <CAKFQuwY0vhNG8T+pC3BQJurFi3NN_L1KbEPGagRN3V5nKZcpDQ@mail.gmail.com>
 <931747.1721687375@sss.pgh.pa.us>
 <b46070c0-0e98-49a0-823f-fa7e13bf626b@eisentraut.org>
 <CAGECzQSGKWy1tcB8BB=p2suOvW0KLtLR+pm=XTagfp9WSA4HXQ@mail.gmail.com>
 <634aca95-6db5-4beb-b18d-67e65582817f@eisentraut.org>
 <4054709.1758895915@sss.pgh.pa.us>
 <3d01eb20-d587-4dec-9261-0c92240ce546@eisentraut.org>
 <4174779.1758917912@sss.pgh.pa.us>
 <87b40ca1-f935-4d71-9edb-9d9f1053cb45@eisentraut.org>
 <49d5cbd1-7e47-4740-bc81-e574150f44cb@eisentraut.org>
From: "David G. Johnston" <david.g.johnston@gmail.com>
Date: Tue, 14 Apr 2026 05:57:27 -0700
Message-ID: 
 <CAKFQuwZ2MoUqWv0oKX4M0d85k92HugdF=1qJzhoC1pO20ywWYA@mail.gmail.com>
Subject: Re: [PATCH] GROUP BY ALL
To: Peter Eisentraut <peter@eisentraut.org>
Cc: Tom Lane <tgl@sss.pgh.pa.us>,
 pgsql-hackers <pgsql-hackers@postgresql.org>,
	David Christensen <david@pgguru.net>, Jelte Fennema-Nio <postgres@jeltef.nl>
Content-Type: multipart/alternative; boundary="0000000000004d8c8d064f6b2438"
Archived-At: 
 <https://www.postgresql.org/message-id/CAKFQuwZ2MoUqWv0oKX4M0d85k92HugdF%3D1qJzhoC1pO20ywWYA%40mail.gmail.com>
Precedence: bulk

--0000000000004d8c8d064f6b2438
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tuesday, April 14, 2026, Peter Eisentraut <peter@eisentraut.org> wrote:
>
>
> I don't see any mention of using GROUP BY with window functions in our
> relevant documentation, for example
>
> https://www.postgresql.org/docs/devel/sql-expressions.html#
> SYNTAX-WINDOW-FUNCTIONS
> https://www.postgresql.org/docs/devel/functions-window.html


The select reference page covers this.  But the window clause could get
better treatment, it=E2=80=99s buried in step 5.

https://www.postgresql.org/docs/current/sql-select.html


> Commit ef38a4d9756 added a regression test
>
> EXPLAIN (COSTS OFF) SELECT a, COUNT(a) OVER (PARTITION BY a) FROM t1 GROU=
P
> BY ALL;
>
> but the test table contains no data, so I don't know if this kind of quer=
y
> produces interesting information.
>

Each non-null value of =E2=80=9Ca=E2=80=9D would have an output of 1, while=
 a null valued
=E2=80=9Ca=E2=80=9D would have an output of 0.  =E2=80=9CA=E2=80=9D is grou=
ped since all expressions
involving =E2=80=9Ca=E2=80=9D are non-aggregated.  The equivalent rewrite i=
s:

Select a, count(a_expr) over =E2=80=A6 from ( =E2=80=94 step 5
    select a, a as a_expr from tbl group by all. =E2=80=94 step 4
);

For purposes of group by all one would erase/ignore the actual window
wrapper while leaving the expressions it operates over in place.  This
extends from =E2=80=9Cwindow expressions are processed after group by/havin=
g=E2=80=9D
documented in SELECT.  IOW, group by all resolves during processing step 4
with intermediate results for the expressions within the window functions,
then step 5 removes the intermediate expressions that don=E2=80=99t appear =
in the
final output while adding in the results of processing the window functions=
.

David J.

--0000000000004d8c8d064f6b2438
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tuesday, April 14, 2026, Peter Eisentraut &lt;<a href=3D"mailto:peter@ei=
sentraut.org">peter@eisentraut.org</a>&gt; wrote:<blockquote class=3D"gmail=
_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:=
1ex">
<br>
I don&#39;t see any mention of using GROUP BY with window functions in our =
relevant documentation, for example<br>
<br>
<a href=3D"https://www.postgresql.org/docs/devel/sql-expressions.html#SYNTA=
X-WINDOW-FUNCTIONS" target=3D"_blank">https://www.postgresql.org/doc<wbr>s/=
devel/sql-expressions.html#<wbr>SYNTAX-WINDOW-FUNCTIONS</a><br>
<a href=3D"https://www.postgresql.org/docs/devel/functions-window.html" tar=
get=3D"_blank">https://www.postgresql.org/doc<wbr>s/devel/functions-window.=
html</a></blockquote><div><br></div><div>The select reference page covers t=
his.=C2=A0 But the window clause could get better treatment, it=E2=80=99s b=
uried in step 5.</div><div><br></div><div><a href=3D"https://www.postgresql=
.org/docs/current/sql-select.html">https://www.postgresql.org/docs/current/=
sql-select.html</a><br></div><div><br></div><blockquote class=3D"gmail_quot=
e" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
Commit ef38a4d9756 added a regression test<br>
<br>
EXPLAIN (COSTS OFF) SELECT a, COUNT(a) OVER (PARTITION BY a) FROM t1 GROUP =
BY ALL;<br>
<br>
but the test table contains no data, so I don&#39;t know if this kind of qu=
ery produces interesting information.<br>
</blockquote><div><br></div><div>Each non-null value of =E2=80=9Ca=E2=80=9D=
 would have an output of 1, while a null valued =E2=80=9Ca=E2=80=9D would h=
ave an output of 0. =C2=A0=E2=80=9CA=E2=80=9D is grouped since all expressi=
ons involving =E2=80=9Ca=E2=80=9D are non-aggregated.=C2=A0 The equivalent =
rewrite is:</div><div><br></div><div>Select a, count(a_expr) over =E2=80=A6=
 from ( =E2=80=94 step 5</div><div>=C2=A0 =C2=A0 select a, a as a_expr from=
 tbl group by all. =E2=80=94 step 4</div><div>);</div><div><br></div><div>F=
or purposes of group by all one would erase/ignore the actual window wrappe=
r while leaving the expressions it operates over in place.=C2=A0 This exten=
ds from =E2=80=9Cwindow expressions are processed after group by/having=E2=
=80=9D documented in SELECT.=C2=A0 IOW, group by all resolves during proces=
sing step 4 with intermediate results for the expressions within the window=
 functions, then step 5 removes the intermediate expressions that don=E2=80=
=99t appear in the final output while adding in the results of processing t=
he window functions.</div><div><br></div><div>David J.</div><div><br></div>

--0000000000004d8c8d064f6b2438--