MIME-Version: 1.0
References: 
 <CADkLM=coCVy92QkVUUTLdo5eO2bMDtwMrzRn_8miAhX+uPaqXg@mail.gmail.com>
 <CAA5RZ0vY5jHXQEOyUdjW7tPrXb9TY_bdr8ZpCRuALj1zU5DD_w@mail.gmail.com>
In-Reply-To: 
 <CAA5RZ0vY5jHXQEOyUdjW7tPrXb9TY_bdr8ZpCRuALj1zU5DD_w@mail.gmail.com>
From: Corey Huinker <corey.huinker@gmail.com>
Date: Sun, 1 Mar 2026 23:03:42 -0500
Message-ID: 
 <CADkLM=d8vuFzQSrV+8MfviK3-TyG8r45yEbnOUd6VJGEoyNmHA@mail.gmail.com>
Subject: Re: Add starelid, attnum to pg_stats and leverage this in pg_dump
To: Sami Imseih <samimseih@gmail.com>
Cc: nathandbossart@gmail.com, pgsql-hackers@lists.postgresql.org
Content-Type: multipart/alternative; boundary="000000000000dca5a8064c02ac33"
Archived-At: 
 <https://www.postgresql.org/message-id/CADkLM%3Dd8vuFzQSrV%2B8MfviK3-TyG8r45yEbnOUd6VJGEoyNmHA%40mail.gmail.com>
Precedence: bulk

--000000000000dca5a8064c02ac33
Content-Type: text/plain; charset="UTF-8"

> Ideally, we would want the OID and attnum to be first columns to match
> the pg_stat_activity/pg_stat_progress style, but that will be too invasive.
>

I don't think there is an order that will make anyone happy, let alone
everyone.

The existing columns (schemaname, tablename, attname, inherited) constitute
a grain, so its hard to separate them.

Once we add attnum, that can replace attname in the above grain. Once we
add starelid/tableid, then that can be combined win inherited and either
attnum/attname for a grain. The ordering of these columns will largely fall
down to the perspective of the person seeking to use it.


>
> - This <type> should be int2
> - Maybe for the describtion, it should be something like:
>

+1


>
>   "The number of the column, as in
> <structname>pg_attribute</structname>.<structfield>attnum</structfield>"
>

We don't do that for attname, why would attnum be different?


>
>
> +     <row>
> +      <entry role="catalog_table_entry"><para role="column_definition">
> +       <structfield>starelid</structfield> <type>oid</type>
> +      </para>
> +      <para>
> +       ID of the relation
> +      </para></entry>
> +     </row>
>
> This should indicate it is a reference to pg_class.oid, like so:
>

I went with pg_attribute, but the reference is there now.


>
> Maybe "ID of the table or index" is better, since this can only be a
> table or index
> for pg_stats.
>

I went with table, because it matches the tablename definition (i.e. it is
equally inaccurate).


>
> I dislike the existing "pg_stats.tablename", since this can also be an
> expression index.
> "pg_stats.relation" with a description of "Name of table or index" is
> more appropriate.
> It is a change that we can possibly make in a major version. Looked
> through the archives,
> and did not see this being reported/discussed.
>

I don't see it changing in any version, minor or major.


>
> v1-0002:
>
> I examined the version variants of getAttributeStats and all looks good,
> and also ran a test on a 18 and 19 server version with the patched pg_dump
> client and all looks good.
>
> One minor comment is:
>
>
> +                    pg_fatal("statistics table oid information missing");
>
>
> I noticed in other pg_fatal messages, we include OIDs
>
>             pg_fatal("could not find function definition for function
> with OID %u",
>                      cast->castfunc);
>
> Should we do the same here?
>

If I had the oid, I wouldn't have the error. :)

I've modified the error message to give the namespace/tag of the entry.
Will post revised patch later.

--000000000000dca5a8064c02ac33
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr"><br></div><br><div class=3D"gmail_quote g=
mail_quote_container"><blockquote class=3D"gmail_quote" style=3D"margin:0px=
 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br=
>
Ideally, we would want the OID and attnum to be first columns to match<br>
the pg_stat_activity/pg_stat_progress style, but that will be too invasive.=
<br></blockquote><div><br></div><div>I don&#39;t think there is an order th=
at will make anyone happy, let alone everyone.</div><div><br></div><div>The=
 existing columns (schemaname, tablename, attname, inherited) constitute a =
grain, so its hard to separate them.</div><div><br></div><div>Once we add a=
ttnum, that can replace attname in the above grain. Once we add starelid/ta=
bleid, then that can be combined win inherited and either attnum/attname fo=
r a grain. The ordering of these columns will largely fall down to the pers=
pective of the person seeking to use it.</div><div><br></div><div>=C2=A0</d=
iv><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0.8ex;bord=
er-left:1px solid rgb(204,204,204);padding-left:1ex"><br>
- This &lt;type&gt; should be int2<br>
- Maybe for the describtion, it should be something like:<br></blockquote><=
div><br></div><div>+1</div><div>=C2=A0</div><blockquote class=3D"gmail_quot=
e" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204)=
;padding-left:1ex">
<br>
=C2=A0 &quot;The number of the column, as in<br>
&lt;structname&gt;pg_attribute&lt;/structname&gt;.&lt;structfield&gt;attnum=
&lt;/structfield&gt;&quot;<br></blockquote><div><br></div><div>We don&#39;t=
 do that for attname, why would attnum be different?</div><div></div><div>=
=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0=
.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br>
<br>
+=C2=A0 =C2=A0 =C2=A0&lt;row&gt;<br>
+=C2=A0 =C2=A0 =C2=A0 &lt;entry role=3D&quot;catalog_table_entry&quot;&gt;&=
lt;para role=3D&quot;column_definition&quot;&gt;<br>
+=C2=A0 =C2=A0 =C2=A0 =C2=A0&lt;structfield&gt;starelid&lt;/structfield&gt;=
 &lt;type&gt;oid&lt;/type&gt;<br>
+=C2=A0 =C2=A0 =C2=A0 &lt;/para&gt;<br>
+=C2=A0 =C2=A0 =C2=A0 &lt;para&gt;<br>
+=C2=A0 =C2=A0 =C2=A0 =C2=A0ID of the relation<br>
+=C2=A0 =C2=A0 =C2=A0 &lt;/para&gt;&lt;/entry&gt;<br>
+=C2=A0 =C2=A0 =C2=A0&lt;/row&gt;<br>
<br>
This should indicate it is a reference to pg_class.oid, like so:<br></block=
quote><div><br></div><div>I went with pg_attribute, but the reference is th=
ere now.</div><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"m=
argin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left=
:1ex"><br>
Maybe &quot;ID of the table or index&quot; is better, since this can only b=
e a<br>
table or index<br>
for pg_stats.<br></blockquote><div><br></div><div>I went with table, becaus=
e it matches the tablename definition (i.e. it is equally inaccurate).</div=
><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px=
 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<br>
I dislike the existing &quot;pg_stats.tablename&quot;, since this can also =
be an<br>
expression index.<br>
&quot;pg_stats.relation&quot; with a description of &quot;Name of table or =
index&quot; is<br>
more appropriate.<br>
It is a change that we can possibly make in a major version. Looked<br>
through the archives,<br>
and did not see this being reported/discussed.<br></blockquote><div><br></d=
iv><div>I don&#39;t see it changing in any version, minor or major.</div><d=
iv><br></div><div>=C2=A0</div><blockquote class=3D"gmail_quote" style=3D"ma=
rgin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:=
1ex">
<br>
v1-0002:<br>
<br>
I examined the version variants of getAttributeStats and all looks good,<br=
>
and also ran a test on a 18 and 19 server version with the patched pg_dump<=
br>
client and all looks good.<br>
<br>
One minor comment is:<br>
<br>
<br>
+=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 pg_f=
atal(&quot;statistics table oid information missing&quot;);<br>
<br>
<br>
I noticed in other pg_fatal messages, we include OIDs<br>
<br>
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 pg_fatal(&quot;could not find fun=
ction definition for function<br>
with OID %u&quot;,<br>
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0cast-&gt;castfunc);<br>
<br>
Should we do the same here?<br></blockquote><div><br></div><div>If I had th=
e oid, I wouldn&#39;t have the error. :)<br><br></div><div>I&#39;ve modifie=
d the error message to give the namespace/tag of the entry. Will post revis=
ed patch later.</div></div></div>

--000000000000dca5a8064c02ac33--