Feedback-ID: i0fe9450f:Fastmail
Date: Wed, 28 Jan 2026 11:46:04 +0900
From: Michael Paquier <michael@paquier.xyz>
To: Corey Huinker <corey.huinker@gmail.com>
Cc: Chao Li <li.evan.chao@gmail.com>, Tender Wang <tndrwang@gmail.com>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: Extended Statistics set/restore/clear functions.
Message-ID: <aXl4bMfSTQUxM_yy@paquier.xyz>
References: 
 <CADkLM=c3JivzHNXLt-X_JicYknRYwLTiOCHOPiKagm2_vdrFUg@mail.gmail.com>
 <aXVw4LGO4fWunn4n@paquier.xyz>
 <CADkLM=f4+3_HrXSAaEic5VGGd-fOjS4jGnr1WGJYrL7vf3ZJnw@mail.gmail.com>
 <aXWBhkru03FnrfIW@paquier.xyz>
 <CADkLM=c7DY3Jv6ef0n_MGUJ1FyTMUoT697LbkST05nraVGNHYg@mail.gmail.com>
 <aXca7jnwtzFSYr4s@paquier.xyz>
 <aXhFKleXWlooHcy3@paquier.xyz>
 <CADkLM=dZr_Ut3jKw94_BisyyDtNZPRJWeOALXVzcJz=ZFTAhvQ@mail.gmail.com>
 <CADkLM=fTgxGL+pcw7RFHe3EP+rmc-Dk_pwh-HWJkonF+xDPqMQ@mail.gmail.com>
 <aXlPml6eaoU8lM5f@paquier.xyz>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha512;
	protocol="application/pgp-signature"; boundary="bmP2HWKQkVU8cMBv"
Content-Disposition: inline
In-Reply-To: <aXlPml6eaoU8lM5f@paquier.xyz>
Archived-At: 
 <https://www.postgresql.org/message-id/aXl4bMfSTQUxM_yy%40paquier.xyz>
Precedence: bulk


--bmP2HWKQkVU8cMBv
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Wed, Jan 28, 2026 at 08:51:54AM +0900, Michael Paquier wrote:
> On Tue, Jan 27, 2026 at 11:14:12AM -0500, Corey Huinker wrote:
> >> 0001 - adds a test left out of dependencies
>=20
> Applied this one now, you are right that it should have been added in
> 302879bd68d1, so my mistake I guess.

This code in v32-0002 has given me a pause, because we define numexprs
and use it down the road for some input validation:
+    /* decode expression (if any) */
+    exprdatum =3D SysCacheGetAttr(STATEXTOID,
+                                tup,
+                                Anum_pg_statistic_ext_stxexprs,
+                                &isnull);
+
+    if (!isnull)
+    {
[...]
+    }
+    numexprs =3D list_length(exprs);

Actually, I think that the patch split has been done incorrectly
because missing this code on HEAD also means that we do not perform a=20
correct validation of ndistinct and dependencies data when any of
these include expressions, because we are missing the negative
attribute numbers, numexprs remaining at 0 all the time.  This can be
easily demonstrated with a statext object like that;
CREATE STATISTICS stats_obj (dependencies) ON lower(name), upper(name)
  FROM tab_obj;

The restore function would reject input like that, but it should be
valid:
[{"attributes": [-1], "dependency": -2, "degree": 1.000000},
 {"attributes": [-2], "dependency": -1, "degree": 1.000000}]

I'll go fix that now with some tests to cover things, after extracting
the relevant portion of the code from v32-0002.
--
Michael

--bmP2HWKQkVU8cMBv
Content-Type: application/pgp-signature; name="signature.asc"

-----BEGIN PGP SIGNATURE-----
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=GaaU
-----END PGP SIGNATURE-----

--bmP2HWKQkVU8cMBv--