Select also zero-values while performing SELECT COUNT with GROUP BY

Question

I'm in the middle of creating stats module. One of my jobs is to create plot with sums of registered users for past week.
The problem is that I work on a service that is being used by only few of our customers, so days when noone registers happen quite often.

Table description (simplified) is:

TABLE users:
============
id serial NOT NULL,
date_created timestamp with timezone NOT NULL DEFAULT NOW()

to get sums for last 7 days, I use

SELECT
    COUNT(id) as ile,
    DATE_TRUNC('day', date_created) as day
FROM
    users
WHERE
    date_created >= now() - INTERVAL '7 days'
GROUP BY 2
ORDER BY 2

Everything works fine, Postgres returns for example:

ile | day
===========================
2   | "2013-09-23 00:00:00"
1   | "2013-09-25 00:00:00"

EXCEPT it ommits days with 0 count. Of course I have to parse results of this query (in this case: PHP + Symfony) so I could just iterate through desired date range and fill the blanks, but I wonder if there's simple way to ask PostgreSQL to do it for me, so it returns something like this:

ile | day
===========================
2   | "2013-09-23 00:00:00"
0   | "2013-09-24 00:00:00"
1   | "2013-09-25 00:00:00"
0   | "2013-09-26 00:00:00"
0   | "2013-09-27 00:00:00"
0   | "2013-09-28 00:00:00"
0   | "2013-09-29 00:00:00"

Answer 1

Can't test at the moment, but you can use generate_series() function:

select
    count(u.id) as ile,
    d.day
from generate_series('2013-09-23'::date, '2013-09-29'::date, '1d') as d(day)
    left outer join users as u on u.date_created >= d.day and u.date_created < d.day + '1d'::interval
    -- or left outer join users as u on date_trunc('day', u.date_created) = d.day
    -- but I think that >= and < will be faster if you have index on `date_created`
group by d.day