From: Tom Lane <tgl@sss.pgh.pa.us>
To: "Euler Taveira" <euler@eulerto.com>
cc: "Kong Man" <kong_mansatiansin@hotmail.com>,
        "pgsql-bugs@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Subject: Re: pg_restore 14 skips ACL COLUMN when --schema is used
In-reply-to: <90659bbc-e0cb-407d-b42b-5931346735ad@app.fastmail.com>
References: 
 <DM4PR11MB73976902DBBA10B1D652F9498B06A@DM4PR11MB7397.namprd11.prod.outlook.com>
 <2427882.1690590240@sss.pgh.pa.us>
 <DM4PR11MB7397E0F1BFD8A839199B59838B05A@DM4PR11MB7397.namprd11.prod.outlook.com>
 <DM4PR11MB73970D090F35B7E1C2EEEB108B05A@DM4PR11MB7397.namprd11.prod.outlook.com>
 <90659bbc-e0cb-407d-b42b-5931346735ad@app.fastmail.com>
Comments: In-reply-to "Euler Taveira" <euler@eulerto.com>
	message dated "Wed, 02 Aug 2023 16:30:02 -0300"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3317366.1691006807.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Wed, 02 Aug 2023 16:06:47 -0400
Message-ID: <3317367.1691006807@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/3317367.1691006807%40sss.pgh.pa.us>
Precedence: bulk

"Euler Taveira" <euler@eulerto.com> writes:
> You didn't provide a test case as requested but I investigated this issu=
e
> according to your description. The following commands is sufficient to p=
roduce
> the analysis below.

Thanks for looking into it!  The ingredient I missed while studying this
earlier is that the table must have *both* table-level and column-level
ACLs to provoke the issue.

> It means that an ACL for columns has 2 dependencies (nDeps =3D 2) the fu=
nction
> _tocEntryRequired() returns 0 (see code below).

D'oh.  I think this code was correct when written, but we added an extra
dependency from column ACL to table ACL later because restoring them in
the wrong order causes problems.

> The discussion [1] in the commit 0d4e6ed3085 does not explain if the 'if=
' logic
> covers all cases. It certainly doesn't for the OP case. The only (hackis=
h)
> suggestion I have ATM is to detect ACL for columns and returns 1.

Yeah, ignoring dependencies on ACLs in this logic seems like the best
way forward.  Do you want to write a patch?

			regards, tom lane