To: pgsql-hackers@postgresql.org
Subject: TODO Items
From: Greg Stark <gsstark@mit.edu>
Organization: The Emacs Conspiracy; member since 1992
Date: 03 Aug 2004 01:06:40 -0400
In-Reply-To: <200408030208.i73280J26911@candle.pha.pa.us>
Message-ID: <874qnkolnj.fsf@stark.xeocode.com>
Lines: 28
User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.3
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


Bruce Momjian <pgman@candle.pha.pa.us> writes:

> TODO item?

On that note several prior conversations I had here ended with WIBNI
conclusions that really ought to be TODO items, in my humble opinion. Two come
to mind off the top of my head resulting in:

. "SELECT * FROM x JOIN y USING (b) WHERE a=?" could use an index on y(a,b)
  since for a constant value of "a" the index traversal would be effectively
  equivalent just be "b". This could result in an efficient merge join
  avoiding an unnecessary sort.

. The semantics for row-value expressions is wrong. (a,b) < (x,y) should be
  true if a<x or if a=x and b<y. Currently it expands to a<x and b<y. 

. Fix row-value expression handling to not depend on the operator names and
  instead use btree access method strategy values instead, allowing row-value
  expressions on other operators with <,=,> behaviour (ie btree indexable
  behaviour).

. Allow multi-column indexes to be used to optimize row-value expressions. Ie,
  allow a btree index on a,b to be used to execute an expression like (a,b) <
  (x,y).

-- 
greg