From: Tom Lane <tgl@sss.pgh.pa.us>
To: poliv78@gmail.com
cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #18676: Execute function while selecting from table with
 partial index using this function.
In-reply-to: <18676-383b0d2b6dc65bf6@postgresql.org>
References: <18676-383b0d2b6dc65bf6@postgresql.org>
Comments: In-reply-to PG Bug reporting form <noreply@postgresql.org>
	message dated "Mon, 28 Oct 2024 14:58:34 -0000"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <3972733.1730130166.1@sss.pgh.pa.us>
Date: Mon, 28 Oct 2024 11:42:46 -0400
Message-ID: <3972734.1730130166@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/3972734.1730130166%40sss.pgh.pa.us>
Precedence: bulk

PG Bug reporting form <noreply@postgresql.org> writes:
> Please run the script below. It creates sample table with partial index
> using IMMUTABLE function. 
> I think when table is changing - index in changing and thus function used in
> WHERE cluse can be run.
> But it is not supposed to be run on just simple SELECT from table without
> even condition to use this index. 

[ shrug... ]  There is no bug here.  The planner is entitled to
investigate the properties of indexes attached to a table it's
trying to plan for.  Sure, this index is not relevant to this
query --- but how's the planner going to know that without
examining the index?  And part of that examination involves
reducing any partial-index predicate to standard form, which
includes constant-folding.  Since your function is marked
IMMUTABLE, that means it should be evaluated and reduced to
a constant.

Part of the contract for marking a function IMMUTABLE is that
it has no interesting side-effects.  This example breaks that
in two ways: the RAISE NOTICE is a side-effect, and the
possibility of a permissions failure is another one.

			regards, tom lane