From: Tom Lane <tgl@sss.pgh.pa.us>
To: remi.cura@gmail.com
cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #18464: Replacing a SQL function silently drops the generated
 columns that use this function
In-reply-to: <18464-9f1edc4037b65003@postgresql.org>
References: <18464-9f1edc4037b65003@postgresql.org>
Comments: In-reply-to PG Bug reporting form <noreply@postgresql.org>
	message dated "Tue, 14 May 2024 15:26:15 -0000"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1104764.1715702118.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Tue, 14 May 2024 11:55:18 -0400
Message-ID: <1104765.1715702118@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1104765.1715702118%40sss.pgh.pa.us>
Precedence: bulk

PG Bug reporting form <noreply@postgresql.org> writes:
> I have a SQL function that I defined myself. Let's call it MyFunction().
> I use MyFunction in generated columns of 2 distinct tables.
> Whe trying to delete MyFunction, the DB correctly raises an error.
> However, when I `CREATE OR REPLACE` this existing function in place, the=
 DB
> silently drop all the generated columns depending on this function.
> This is not the expected behavior.

Indeed not, and it does not happen like that for me:

$ psql
psql (14.12)
Type "help" for help.

regression=3D# create function testfunc(int) returns int strict immutable =
language sql as 'select $1+1';
CREATE FUNCTION
regression=3D# create table foo (f1 int, f2 int generated always as (testf=
unc(f1)) stored);
CREATE TABLE
regression=3D# insert into foo values(1),(10);
INSERT 0 2
regression=3D# table foo;
 f1 | f2 =

----+----
  1 |  2
 10 | 11
(2 rows)

regression=3D# create or replace function testfunc(int) returns int strict=
 immutable language sql as 'select $1+2';
CREATE FUNCTION
regression=3D# table foo;
 f1 | f2 =

----+----
  1 |  2
 10 | 11
(2 rows)

Now you could make a reasonable argument that the generated column
should have auto-updated.  We don't implement that (at least not
yet), on the grounds that what I did above broke the function's
promise of immutability.  The same change would also have broken
indexes using the function, so generated columns are not any worse.

In any case, please provide a self-contained example demonstrating
your claim.

			regards, tom lane