Subject: Re: Using count on a join, group by required?
From: Peter Eisentraut <peter_e@gmx.net>
To: Jose Ildefonso Camargo Tolosa <ildefonso.camargo@gmail.com>
Cc: pgsql-sql@postgresql.org
In-Reply-To: <AANLkTinYNWi1AQft188MBWj4n6OBfSQX1DdAO2Jf8wQu@mail.gmail.com>
References: <AANLkTinYNWi1AQft188MBWj4n6OBfSQX1DdAO2Jf8wQu@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
Date: Wed, 11 Aug 2010 15:39:28 +0300
Message-ID: <1281530368.8340.4.camel@fsopti579.F-Secure.com>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

On tis, 2010-08-10 at 22:21 -0430, Jose Ildefonso Camargo Tolosa wrote:
> And it works, it gives me something like:
> 
>  product_id |         name         |     code      | manufacturer_id |
>     manufacturer_name      | num_serials
> ------------+----------------------+---------------+-----------------+----------------------------+-------------
>          17 | THE product          | 1235711131719 |              19 |
> THE product's manufacturer |           5
>           6 | Car Battery 500A 12V | 7591512021575 |               8 |
> Acumuladores Duncan, C.A.  |          11
>           1 | Test product 1       | 123456789012  |               1 |
> Test Manufacturer          |           6
> 
> Which is correct, and exactly what I wanted.
> 
> So far, so good.  The thing is: the group by clause, I had to add it
> because the parser forced me to, because it complained like this:
> 
> ERROR:  column "manufacturer.name" must appear in the GROUP BY clause
> or be used in an aggregate function
> 
> and I had to include *all* the requested columns on the group by
> clause, can anybody tell me why? or at least point to some doc that
> help me understanding this?

This is fixed in PostgreSQL 9.1devel (*); there you only need to put the
primary key into the GROUP BY clause.  Earlier versions didn't know that
that was enough to ensure a deterministic result.

(*) -- It will probably be a bit over a year before that is released.