To: Alvaro Herrera <alvherre@commandprompt.com>
cc: Ow Mun Heng <Ow.Mun.Heng@wdc.com>, Mario Weilguni <mweilguni@sime.com>,
	Pgsql performance <pgsql-performance@postgresql.org>
Subject: Re: Altering a column type - Most efficient way 
In-reply-to: <20080711134423.GA4110@alvh.no-ip.org> 
References: <1215676380.15319.2.camel@neuromancer.home.net>
	<4875C9FA.7020101@sime.com>
	<1215682481.15319.14.camel@neuromancer.home.net>
	<20080710135756.GD3757@alvh.no-ip.org>
	<1215742620.853.10.camel@neuromancer.home.net>
	<20080711134423.GA4110@alvh.no-ip.org>
Comments: In-reply-to Alvaro Herrera <alvherre@commandprompt.com>
	message dated "Fri, 11 Jul 2008 09:44:23 -0400"
Date: Fri, 11 Jul 2008 10:27:34 -0400
Message-ID: <3462.1215786454@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Alvaro Herrera <alvherre@commandprompt.com> writes:
> Ow Mun Heng wrote:
>> If it were this simple a change, I'm not certain why (I believe) PG is
>> checking each and every row to see if it will fit into the new column
>> definition/type. 

> Because the code that does the ALTER TYPE is very generic, and it
> doesn't (yet) have an optimization that tells it to skip the check and
> the possible table rewrite in the cases where it's obviously not needed
> (like this one).

Awhile back I looked into teaching ALTER TYPE that it needn't rewrite
if the type conversion expression parses out as just a Var with
RelabelType, but it seemed that that wouldn't cover very much of the
use-cases where a human thinks that it's "obvious" that no rewrite
is needed.  You'd really need to build in hard-wired knowledge about
the behavior of specific coercion functions, which seems entirely
unappealing.

			regards, tom lane