To: Michael Fuhr <mike@fuhr.org>
Cc: chester c young <chestercyoung@yahoo.com>,
	pgsql-sql@postgresql.org
Subject: Re: broken join optimization? (8.0) 
In-reply-to: <20051026024211.GA98628@winnie.fuhr.org> 
References: <20051025235612.20767.qmail@web54301.mail.yahoo.com>
	<20051026024211.GA98628@winnie.fuhr.org>
Comments: In-reply-to Michael Fuhr <mike@fuhr.org>
	message dated "Tue, 25 Oct 2005 20:42:11 -0600"
Date: Tue, 25 Oct 2005 23:53:44 -0400
Message-ID: <1416.1130298824@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Michael Fuhr <mike@fuhr.org> writes:
> On Tue, Oct 25, 2005 at 04:56:11PM -0700, chester c young wrote:
>> in php (for example) it's frequently nice to get the structure of a
>> table without any data,

> Have you considered "SELECT * FROM mytable LIMIT 0"?

Indeed.

> I see the same behavior in the latest 8.1beta code.  Maybe one of
> the developers will comment on whether optimizing that is a simple
> change, a difficult change, not worth changing because few people
> find a use for it, or a behavior that can't be changed because of
> something we're not considering.

Not worth changing --- why should we expend cycles (even if it only
takes a few, which isn't clear to me offhand) on every join query, to
detect what's simply a brain-dead way of finding out table structure?
I can't think of any realistic scenarios for a constant-false join
clause.

The relevant bit of code is in initsplan.c:

    /*
     * If the clause is variable-free, we force it to be evaluated at its
     * original syntactic level.  Note that this should not happen for
     * top-level clauses, because query_planner() special-cases them.  But it
     * will happen for variable-free JOIN/ON clauses.  We don't have to be
     * real smart about such a case, we just have to be correct.
     */
    if (bms_is_empty(relids))
        relids = qualscope;

Possibly you could get the planner to generate a gating Result node for
such a case, the way it does for constant-false top level WHERE clauses,
but I really doubt it's worth any extra cycles at all to make this
happen.  The proposed example is quite unconvincing ... why would anyone
want to depend on the existence of a "dual" table rather than LIMIT 0?

			regards, tom lane