From: Tom Lane <tgl@sss.pgh.pa.us>
To: Masahiko Sawada <sawada.mshk@gmail.com>
cc: "Bossart, Nathan" <bossartn@amazon.com>,
        Michael Paquier <michael@paquier.xyz>,
        Neil Chen <carpenter.nail.cz@gmail.com>,
        "Boris P. Korzun" <drtr0jan@yandex.ru>,
        "pgsql-bugs@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Subject: Re: Inconsistent behavior of pg_dump/pg_restore on DEFAULT PRIVILEGES
In-reply-to: 
 <CAD21AoArNPsNKYE1b-oURYonUA7Zo-kaq-u661uWcZqTURb48A@mail.gmail.com>
References: 
 <CAA3qoJnr2+1dVJObNtfec=qW4Z0nz=A9+r5bZKoTSy5RDjskMw@mail.gmail.com>
 <8261632200581@mail.yandex.ru>
 <CAA3qoJ=f6jZAkkfwH+rtH_8C9z-nfcgNFYZnGYxVdBjz7Vkhig@mail.gmail.com>
 <CAD21AoBrCbjiJt3OCDQCLrM6OQOxjV9cAyM6bn4=SjHq0konBQ@mail.gmail.com>
 <CAD21AoC+m46JEsFE2d6JQuPuNN5ODj1vpZ3z2HtPdn+HFDJfdg@mail.gmail.com>
 <721942.1634180151@sss.pgh.pa.us>
 <CAD21AoAUCNCo77LJLAFGEkhtgruMpzPD153SC7dY7TyahvM_Fg@mail.gmail.com>
 <YWfh9O/36r+GlgNB@paquier.xyz>
 <CAD21AoDRg1E7AEM_5ifyjFbQ71J303OHA-R1HjzqC0FT306dKQ@mail.gmail.com>
 <AB89F451-ED7F-4FA6-ADF8-F8132220D6A2@amazon.com>
 <CAD21AoArNPsNKYE1b-oURYonUA7Zo-kaq-u661uWcZqTURb48A@mail.gmail.com>
Comments: In-reply-to Masahiko Sawada <sawada.mshk@gmail.com>
	message dated "Tue, 19 Oct 2021 11:04:32 +0900"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1737875.1634613552.1@sss.pgh.pa.us>
Content-Transfer-Encoding: quoted-printable
Date: Mon, 18 Oct 2021 23:19:12 -0400
Message-ID: <1737876.1634613552@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1737876.1634613552%40sss.pgh.pa.us>
Precedence: bulk

Masahiko Sawada <sawada.mshk@gmail.com> writes:
> I've looked at the patch proposed you proposed. If we can depend on
> acldefault() being STRICT (which is legitimate to me), I think we
> don't need to build an expression depending on the caller (i.g.,
> is_default_acl). If acldefault() were to become not STRICT, we could
> detect it by regression tests. What do you think?

FWIW, I'm working on a refactoring of this logic that will bring the
acldefault() call into the getDefaultACLs code, which would mean that
we won't need that assumption anymore anyway.  The code as I have it
produces SQL like

     acldefault(CASE WHEN defaclobjtype =3D 'S'
                THEN 's'::"char" ELSE defaclobjtype END, defaclrole) AS ac=
ldefault

and we could wrap the test-for-zero around that:

     CASE WHEN defaclnamespace =3D 0 THEN
       acldefault(CASE WHEN defaclobjtype =3D 'S'
                  THEN 's'::"char" ELSE defaclobjtype END, defaclrole)
     ELSE NULL END AS acldefault

(although I think it might be better to write ELSE '{}' not ELSE NULL).

So I think we don't need to worry about whether acldefault() will stay
strict.  This patch will only need to work in the back branches.

			regards, tom lane