Content-Type: text/plain;
	charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3864.400.21\))
Subject: Re: ALTER TABLE: warn when actions do not recurse to partitions
From: Chao Li <li.evan.chao@gmail.com>
In-Reply-To: 
 <CAKAnmmLWdEvFR+-+5HXzMEnd8fLfyvkK4QxtSHMBhoXC5Qvcug@mail.gmail.com>
Date: Thu, 12 Mar 2026 09:14:20 +0800
Cc: Jim Jones <jim.jones@uni-muenster.de>,
 "David G. Johnston" <david.g.johnston@gmail.com>,
 Postgres hackers <pgsql-hackers@lists.postgresql.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <A762E5CC-2BC3-4215-8A70-48281FEE2234@gmail.com>
References: 
 <CAEoWx2=SLga-xH09Cq_PAvsHhQHrBK+V0vF821JKgzS=Bm0haA@mail.gmail.com>
 <CAKFQuwZXt2dAuHFs6MtRCzfMP6YZFzssuzdW2woJBVmco=CDdQ@mail.gmail.com>
 <CAEoWx2nic2MaPUEKwb0Me1iwO3LsCf8wUHm55pjqaTRgJEZpcQ@mail.gmail.com>
 <CAKFQuwahxpPKT-LXGJ34BO5cVBUxFxzgNPTY8E_VW-cuNU_DkQ@mail.gmail.com>
 <CAEoWx2kQ9aSn-0LGPKc=woj+h-HR4uBV9TL20aeM3HzLmFqgFQ@mail.gmail.com>
 <6eff5e43-cacd-4a2a-ad1d-e3b313c86050@uni-muenster.de>
 <950BB7B5-0180-4C36-82A0-7E17B920F740@gmail.com>
 <CAEoWx2mZsWnyS_XLDL5mL+yuU_g65p_tmHNesyfngqow7gtS3A@mail.gmail.com>
 <fb70022a-5403-4e16-9efc-28b041f35d9a@uni-muenster.de>
 <8ECD9403-F0BB-4971-94CF-2709EEB4E3B9@gmail.com>
 <f4f70647-8738-48b3-abbd-5b52cde97374@uni-muenster.de>
 <9174F0CF-2F70-4B4F-AED4-CAF113B7F093@gmail.com>
 <14669a83-c7b4-4cdf-890c-dceecd025ee1@uni-muenster.de>
 <3F59D90E-9A47-4C1C-B330-D62D668A462E@gmail.com>
 <5e01263a-1994-44d5-9e98-7212acf9c985@uni-muenster.de>
 <CAKAnmmKzzOWKQVhaAt8LrBmUW=FLWBqS4v8cwKOG53hBToQZRA@mail.gmail.com>
 <D1AB25A4-B3D2-4E57-AE0F-1CBD3823EB19@gmail.com>
 <CAKAnmmLWdEvFR+-+5HXzMEnd8fLfyvkK4QxtSHMBhoXC5Qvcug@mail.gmail.com>
To: Greg Sabino Mullane <htamfids@gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/A762E5CC-2BC3-4215-8A70-48281FEE2234%40gmail.com>
Precedence: bulk


> On Mar 12, 2026, at 00:39, Greg Sabino Mullane <htamfids@gmail.com> =
wrote:
>=20
> On Wed, Mar 11, 2026 at 3:05=E2=80=AFAM Chao Li =
<li.evan.chao@gmail.com> wrote:
> Are you concerning that rendering the full message text is the extra =
work? This is not a hot path, so I don=E2=80=99t think that would be a =
big deal. Actually, adding two more fields sounds more expensive
>=20
> Well, the recurring creation and freeing of the strings is the part =
that seems inefficient. But you don't even need to store the strings at =
all if you are tracking the action+rel. In such a case, the final =
strings can be created on the fly inside of =
EmitPartitionNoRecurseNotice, right? Then you just need a list to store =
the combos of action+relation.
>=20

Fully understood your point. My considerations are:

* This is not on a hot path, and that=E2=80=99s a very trivial =
performance impact.
* I would believe in most of use cases, ALTER TABLE won=E2=80=99t take =
duplicate sub-commands, thus duplicated messages should rarely happen.
* If we take your approach, actually, we don=E2=80=99t have to store =
action+relation in the list, only action is okay. But, if we defer =
building the notice message to EmitPartitionNoRecurseNotice, then we =
have to leave relation open till EmitPartitionNoRecurseNotice, which =
feels a worse burden. Looking at ATController(), rel is closed early. An =
alternative is to store the relation name in a temp variable, which also =
introduces extra code. As a trade-off, I think building the notice =
message in CollectPartitionNoRecurseNotice() is an easy implementation.

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/