To: Svenne Krap <svenne@krap.dk>
Cc: pgsql-performance@postgresql.org
Subject: Re: Perfomance of views 
In-reply-to: <4360B375.8000303@krap.dk> 
References: <435FA002.4060509@krap.dk> <43609EF4.8060907@archonet.com>
	<4360B375.8000303@krap.dk>
Comments: In-reply-to Svenne Krap <svenne@krap.dk>
	message dated "Thu, 27 Oct 2005 13:01:09 +0200"
Date: Thu, 27 Oct 2005 10:38:02 -0400
Message-ID: <18899.1130423882@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Svenne Krap <svenne@krap.dk> writes:
> The ways I have considered was :
> 1) write a big query in hand (not preferred as it gets hard to manage)
> 2) write layers of views (still not prefered as I still have to remember 
> to put on the right conditions everywhere)
> 3) write layers of sql-functions (returning the right sets of rows from 
> the underlying tables) - which I prefer from a development angel .. it 
> gets very clean and I cant forget a parameter anywhere.

#1 and #2 should behave pretty similarly, assuming that the "one big
query" would have been structured the same way as the nest of views is.
#3 unfortunately will pretty much suck, because there's no chance for
cross-level optimization.

There's been some discussion of inline-expanding SQL functions that
return sets when they are called in FROM, which would make a SQL
function that contains just a SELECT effectively equivalent to a view
as far as the planner's powers of optimization go.  No one's tried to
make it happen yet though.

			regards, tom lane