To: "Grace C. Unson" <gracec@ntsp.nec.co.jp>
Cc: "PgSQL SQL" <pgsql-sql@postgresql.org>
Subject: Re: Index not recognized
References: <3FD0A87B.4010803@scmbb.ulb.ac.be>
	<001101c3bbb6$8b0db840$ec64a8c0@GREECE>
In-Reply-To: <001101c3bbb6$8b0db840$ec64a8c0@GREECE>
From: Greg Stark <gsstark@mit.edu>
Organization: The Emacs Conspiracy; member since 1992
Date: 06 Dec 2003 19:36:40 -0500
Message-ID: <87ptf11n1j.fsf@stark.dyndns.tv>
Lines: 42
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.3
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


"Grace C. Unson" <gracec@ntsp.nec.co.jp> writes:

> Why is it that my index for text[] data type is not recognized by the
> Planner?
> 
> I did these steps:
> 
> 1. create function textarr(text[]) returns text language sql as 'select
> $1[1]' strict immutable
> 2. create index org_idx on EmpData (textarr(org));

This index will only be used if you use the expression textarr(org) in your
query. You would probably have some success if you did:

 select * from empdata where textarr(org) = 'math' 

> 3. vacuum full
> 4. explain analyze select name from EmpData where org *= 'math';

Is this *= operator from the contrib/array directory? It's not an indexable
operator at all using standard btree indexes.

The GiST indexing does make indexable operators that can do things like *= but
that's a whole other ball of wax.

What are you really trying to do?

> Result:
> =========
> Seq Scan on EmpData (cost=0.00..3193.20 rows=102 width=488)
> (actual time=3.71.35..371.35 rows=0 loops=1)
> 
> Filter: (org[0]='math'::text)

Well that's awfully odd. I don't know how that expression came out of the
query you gave. You'll have to give a lot more information about how you're
defining *= and why you think it's related to the function you used to define
the index.

-- 
greg