From: Tom Lane <tgl@sss.pgh.pa.us>
To: jinhui.lai@qq.com
cc: pgsql-bugs@lists.postgresql.org
Subject: Re: BUG #19059: PostgreSQL fails to evaluate the cheaper expression
 first, leading to 45X performance degradation
In-reply-to: <19059-c751eed385314309@postgresql.org>
References: <19059-c751eed385314309@postgresql.org>
Comments: In-reply-to PG Bug reporting form <noreply@postgresql.org>
	message dated "Sat, 20 Sep 2025 16:08:11 -0000"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2510591.1758403162.1@sss.pgh.pa.us>
Date: Sat, 20 Sep 2025 17:19:22 -0400
Message-ID: <2510592.1758403162@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/2510592.1758403162%40sss.pgh.pa.us>
Precedence: bulk

PG Bug reporting form <noreply@postgresql.org> writes:
> However, the optimizer fails to reorder the expressions in the WHERE clause
> for the second query. You can observe this from the second row in the plan:
> "Filter: (EXISTS(SubPlan 1) OR (c0 > 0))"
> A more optimal strategy would be for PG to use its cost model to reorder
> expressions, prioritizing the evaluation of less expensive operations first.

We do do that at the top AND level (cf. order_qual_clauses())
but we have not bothered for OR clauses.

The whole exercise is pretty questionable really, considering how weak
our cost model for expressions is.  We could easily end up pessimizing
a clause that the user had put into carefully-selected order.
While we've not gotten a huge amount of push-back about the top-level
re-ordering, there's been some complaints.  So I'm not eager to
propagate the idea further down.

			regards, tom lane