To: pgsql-sql@postgresql.org
From: Harald Fuchs <hari.fuchs@gmail.com>
Subject: Re: Window function trouble
Date: Fri, 11 Dec 2009 16:40:13 +0100
Lines: 57
Message-ID: <pu7hstiko2.fsf@srv.protecting.net>
References: <pud42mahnr.fsf@srv.protecting.net>
	<pu7hsuafcq.fsf@srv.protecting.net>
	<25983.1260468559@sss.pgh.pa.us>
	<puy6la8rw7.fsf@srv.protecting.net>
	<28855.1260486487@sss.pgh.pa.us>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.2 (gnu/linux)
Cancel-Lock: sha1:dBSL41SjVA8y11HPaDY8J5MEjg0=

In article <28855.1260486487@sss.pgh.pa.us>,
Tom Lane <tgl@sss.pgh.pa.us> writes:

> Harald Fuchs <hari.fuchs@gmail.com> writes:
>> That being said, I still think that PostgreSQL could do better - how
>> about naming expression columns so that they are distinct from column
>> names?

> Even though the rules we use are pretty arbitrary, I'm hesitant to make
> changes in them; it seems more likely to break existing applications
> than to do anyone any good.

Well, it would be far less confusing.  Here's a self-contained example:

  CREATE TABLE t1 (
    id serial NOT NULL,
    adr text NOT NULL,
    usr text NOT NULL,
    PRIMARY KEY (id)
  );

  COPY t1 (adr, usr) FROM stdin DELIMITER '|';
  a1|u1_1
  a2|u2_1
  a2|u2_2
  a2|u2_2
  a3|u3_1
  a3|u3_2
  a3|u3_2
  a3|u3_3
  a3|u3_3
  a3|u3_3
  \.

  SELECT CASE lag(adr) OVER (ORDER BY adr, usr)
         WHEN adr THEN NULL
         ELSE adr
         END,
         usr, count(*)
  FROM t1
  GROUP BY adr, usr
  ORDER BY adr, usr;

The result set is:

   adr | usr  | count 
  -----+------+-------
   a1  | u1_1 |     1
   a2  | u2_1 |     1
   a3  | u3_1 |     1
       | u2_2 |     2
       | u3_2 |     2
       | u3_3 |     3

This means IMHO that "GROUP BY adr, usr" groups by the column named
"adr", whereas "ORDER BY adr, usr" orders by the unnamed CASE
expression which happens to "hide" the column name without warning.