Resent-From: POSTGRES mailing list <postman@postgres.Berkeley.EDU>
Resent-Message-Id: <199403240241.SAA19092@nobozo.CS.Berkeley.EDU>
Sender: owner-postman@postgres.Berkeley.EDU
Message-Id: <199403240241.SAA02022@faerie.CS.Berkeley.EDU>
From: aoki@postgres.Berkeley.EDU (Paul M. Aoki)
To: cs400514@comp.hkp.hk (Lee Siu Hoi (BACS4 Class A))
Cc: postgres@postgres.Berkeley.EDU
Subject: Re: Rtree index on points 
Reply-To: aoki@postgres.Berkeley.EDU (Paul M. Aoki)
In-reply-to: Your message of Mon, 7 Mar 94 19:24:36 HKT 
	     <9403071124.AA02985@csun02.COMP.HKP.HK> 
Date: Wed, 23 Mar 94 18:41:10 -0800
Resent-To: postgres-dist@postgres.Berkeley.EDU
Resent-Date: Wed, 23 Mar 94 18:41:54 -0800
Resent-XMts: smtp

cs400514@comp.hkp.hk (Lee Siu Hoi (BACS4 Class A)) writes:
> There are 3 classes of rtree operators : box_ops, poly_ops and bigbox_ops.
> Can I use them to index on point, path, line attributes ?

the guys who put together the sequoia 2000 benchmark wrote a
set of routines to convert points into polygons and then tried
to build a "functional index" using this function and poly_ops.
this didn't work because there were some bugs in the rtree code 
having to do with variable length types (that have been fixed in 
4.2).

functional indices are (minimally) documented in the 4.2 reference 
manual under define index.  basically, the index search keys are 
computed using the result of a function of the real key, but the
final result of an index probe is the actual key.

it might be possible to do something similar by writing functions
to map paths and lines into bounding boxes/polygons but i haven't
tried it or even thought about it very hard.
--
  Paul M. Aoki  |  CS Div., Dept. of EECS, UCB  |  aoki@postgres.Berkeley.EDU
                |  Berkeley, CA 94720           |  ...!uunet!ucbvax!aoki