DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=message-id:date:from:user-agent:mime-version:to:cc:subject
	:references:in-reply-to:content-type:content-transfer-encoding;
	b=G9/hn5wSukuxCFABoabep7zpwrQFYRWuufh5VO8a2avfwiRbweBugQqL/g4VV2pFQv
	hKQpEiZLQ3pgg398Vxr01GP7+6iOsjvvgAY7rQllB08K1hJ6+36YL3aQVTZter0uxzmz
	IhX9t/WcezxIO8BIQerSYjX73kDmbsIYcaGao=
Message-ID: <4BC88364.9050404@gmail.com>
Date: Fri, 16 Apr 2010 17:33:56 +0200
From: Yeb Havinga <yebhavinga@gmail.com>
User-Agent: Thunderbird 2.0.0.24 (Windows/20100228)
MIME-Version: 1.0
To: Tom Lane <tgl@sss.pgh.pa.us>
CC: =?UTF-8?B?0JLRltGC0LDQu9GW0Lkg0KLQuNC80YfQuNGI0LjQvQ==?=
	<tivv00@gmail.com>, pgsql-performance@postgresql.org
Subject: Re: Planner not using column limit specified for one column
	for another column equal to first
References: <m2y331e40661004160102tff445f9eg165a04efe3b409ed@mail.gmail.com>
	<8534.1271427587@sss.pgh.pa.us>
In-Reply-To: <8534.1271427587@sss.pgh.pa.us>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit

Tom Lane wrote:
> =?KOI8-U?B?96bUwcymyiD0yc3eydvJzg==?= <tivv00@gmail.com> writes:
>   
>> I've thought and someone in this list've told me that this should be done
>> automatically.
>>     
>
> No, that's not true.  We do make deductions about transitive equalities,
> ie, given WHERE a=b AND b=c the planner will infer a=c and use that if
> it's helpful.  We don't make deductions about inequalities such as a>c.
> In theory there's enough information available to do so, but overall
> trying to do that would probably waste more cycles than it would save.
> You'd need a lot of expensive new planner infrastructure, and in the
> vast majority of queries it wouldn't produce anything very helpful.
>   
New expensive planner infrastructure to support from a>b and b>c infer 
a>c, yes.

But I wonder if something like Leibniz's principle of identity holds for 
members of the same equivalence class, e.g. like if x,y are both members 
of the same EC, then for every predicate P, P(x) iff P(y). Probably not 
for every predicate (like varno = 2 or attname='x'), but for the query 
evaluation, the object denoted by the variables are the same, since that 
is the standard meaning of the = operator. I cannot think of any 
standard (btree) operator where 'Leibniz' would fail in this case.

regards,
Yeb Havinga