Date: Mon, 24 Nov 2025 13:27:04 -0600
From: Nico Williams <nico@cryptonector.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Laurenz Albe <laurenz.albe@cybertec.at>,
	Calvin Guo <newoakllc2023@gmail.com>,
	pgsql-general@lists.postgresql.org
Subject: Re: set role command
Message-ID: <aSSxiCArsuPkf/I5@ubby>
References: <CA+bysH_or91tt7r0gKLJtw5Wp+DEYwnaRJoTvLfKO33dcdQ_rQ@mail.gmail.com>
 <464edf94a18cf09a49e275d9e9d723949d7682d3.camel@cybertec.at>
 <955750.1764001100@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <955750.1764001100@sss.pgh.pa.us>
Archived-At: <https://www.postgresql.org/message-id/aSSxiCArsuPkf/I5%40ubby>
Precedence: bulk

On Mon, Nov 24, 2025 at 11:18:20AM -0500, Tom Lane wrote:
> Laurenz Albe <laurenz.albe@cybertec.at> writes:
> > On Mon, 2025-11-24 at 16:15 +0800, Calvin Guo wrote:
> >> I really feel, once you "set role usera", you should behave like usera, you should
> >> NOT have the power say: hi, I can assume my super user power whenever I want.
> >> As this make the "set role usera" pretty much useless.
> 
> > I respect your feelings, but that is not how SET ROLE works.
> > The current behavior is intentional and documented in
> > https://www.postgresql.org/docs/current/sql-set-role.html
> 
> And it's also required by the SQL standard, which is very clear
> that "user identifier" and "role" are different things, and
> SET ROLE only changes the latter.
> 
> > There is SET SESSION AUTHORIZATION, which acts somewhet more like you want,
> > except that you can become a superuser again with RESET SESSION AUTHORIZATION.
> 
> In the standard, the privileges required to do SET SESSION
> AUTHORIZATION are "implementation defined", which means we could
> change how it works without breaking standards conformance.
> We'd still be breaking backwards compatibility, though --- for
> instance, pg_dump dumps made with --use-set-session-authorization
> would stop working.  I think that a proposal to change this has
> very little chance of succeeding.

Can we have an extension to say PERMANENTLY?

This is the SQL equivalent of `setreuid()`, essentially.  Except that
unix has a way to do it permanently.  It's a _very_ useful thing to
have.

Nico
--