Date: Sun, 17 Aug 2003 19:39:38 +0400
From: "Alexander M. Pravking" <fduch@antar.bryansk.ru>
To: Richard Huxton <dev@archonet.com>
Cc: pgsql-sql@postgresql.org
Subject: Re: Expression transformation curiosity
Message-ID: <20030817153938.GA437@dyatel.antar.bryansk.ru>
References: <200308171132.58073.dev@archonet.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=koi8-r
Content-Disposition: inline
In-Reply-To: <200308171132.58073.dev@archonet.com>
User-Agent: Mutt/1.4.1i

On Sun, Aug 17, 2003 at 11:32:58AM +0100, Richard Huxton wrote:
> I just had a look at EXPLAIN ANALYSE output for an SQL function I'm trying to 
> write. The WHERE clause wasn't too complex, but the explain output took up a 
> dozen lines.
> 
> Boiling the problem down, I've looked at a clause of the form:
>   a OR (b AND c)
> which PG converts to:
>   (a OR b) AND (a OR c)
> 
> Now these two are equivalent, but it would take me forever to demonstrate that 
> with the full query. I'm happy the planner is going to get it right, but I'm 
> confused as to why the transformation occurs.
> 
> Is it an artefact of displaying the EXPLAIN, or is it actually processed that 
> way? You could see how testing "a" twice could be expensive in some 
> situations.

Looks like it actually works this way.
I had the same problem several weeks ago on 7.3.3 with 4 such OR's.
The final filter became monsterous, and the query was very slow.

I've simply rewritten the query using UNION, and it became much faster.


-- 
Fduch M. Pravking