Date: Wed, 4 Jun 2025 15:17:10 -0700
From: Noah Misch <noah@leadboat.com>
To: Bruce Momjian <bruce@momjian.us>, Tom Lane <tgl@sss.pgh.pa.us>
Cc: "David G. Johnston" <david.g.johnston@gmail.com>,
	PostgreSQL-development <pgsql-hackers@lists.postgresql.org>
Subject: Re: PG 18 release notes draft committed
Message-ID: <20250604221710.2c.nmisch@google.com>
References: <aBQxop9YxQDad8-W@momjian.us>
 <20250603172123.5f.nmisch@google.com>
 <aECwXlB15awOXivC@momjian.us>
 <CAKFQuwYhd6-fSM-inS9RnB5WydVpzj471-YNgU7CiVZ7WWi3ow@mail.gmail.com>
 <aEDAE0590fqvU0Ud@momjian.us>
 <1071973.1749075038@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1071973.1749075038@sss.pgh.pa.us>
User-Agent: Mutt/2.2.12 (2023-09-09)
Archived-At: 
 <https://www.postgresql.org/message-id/20250604221710.2c.nmisch%40google.com>
Precedence: bulk

On Wed, Jun 04, 2025 at 06:10:38PM -0400, Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > I think we have to keep the non-SECURITY-DEFINER designation to keep the
> > text accurate,

I don't see it that way, because there's currently no experiment the user can
perform to distinguish between the following alternatives:

1. We switch to the user that fired the trigger, then call the SECURITY
   DEFINER function.  As always, the first step of executing a SECURITY
   DEFINER function is to switch to its owner.

2. We know it's a SECURITY DEFINER function, so we don't switch.  We just call
   the SECURITY DEFINER function.  As always, the first step of executing a
   SECURITY DEFINER function is to switch to its owner.

The actual implementation is (1), for what it's worth.  The src/backend part
of the commit didn't special-case SECURITY DEFINER.

> but you are right it is part of the function, not the
> > trigger:
> 
> > 	Execute deferred constraint triggers attached to
> > 	non-SECURITY-DEFINER functions as the role that was active at
> > 	the time the trigger was fired
> 
> > 	Previously such triggers were run as the role that was active at
> > 	commit/execution time.

> It's still inaccurate -- to my mind, a "deferred" trigger is one that
> runs later than the end of the triggering statement.  I think you
> should use "after trigger".  Also, "fired" is a fairly confusing
> choice of word here; I think most people would take that as meaning
> the act of running the trigger.  How about
> 
> 	Execute AFTER triggers as the role that was active at the
> 	moment the trigger event was queued
> 
>  	Previously such triggers were run as the role that is active
> 	when it is time to execute the trigger (e.g., at COMMIT).

I like that.