Group by week, display empty weeks - with where clause - mysql

Question

We are attempting to count how many unique telephone numbers called a particular number each day, with 0 (or NULL) for days with no calls. To simplify the data schema, our table contains four fields:
| id | fromz | toz | date |
fromz: inbound call number
toz: number called
date: yyyy-mm-dd
When all we need to know is how many unique numbers called by each day - and do not care what number was called - it is simple to include no-calls days in results.
We JOIN to another table containing only sequential dates: calendar
| id | date |
date:yyyy-mm-dd
SELECT p.fromz, count(unique(p.fromz)), p.date FROM phones p
RIGHT OUTER JOIN calendar c ON
p.date = c.date
GROUP BY c.date
If no p.fromz on c.date, the result for that p.date is "0" (or NULL)
Problem arises when we begin to sort by numbers called:
SELECT p.fromz, count(unique(p.fromz)), p.date FROM phones p
RIGHT OUTER JOIN calendar ON
p.date = c.date
WHERE p.toz = "#myNumber"
GROUP BY c.date
Because there are no WHERE toz = #myNumber on some days we only get results on days when (WHERE) there were calls to #myNumber.
Any suggestions?

Answer 1

I've modified some of your field names slightly. I'm not sure if this 100% meets your use-case, but if it doesn't, you should be able to figure it out from here. Basically, move your current WHERE condition into the ON clause.
SELECT p.fromz, count(distinct p.fromz), p.callDate, c.theDate FROM phones p
RIGHT OUTER JOIN calendar c ON
p.callDate = c.theDate AND p.toz = 2125555555
GROUP BY c.theDate
SQLFiddle
Full Disclosure
This is not standards-compliant. For a compliant query, take a look at Ollie's answer and simply move the p.toz part from the WHERE to the ON to get the results you want.

Answer 2

You are very close.
You're using a nasty MySQL nonstandard extension to GROUP BY and it is driving you crazy, as it has driven many others crazy before you.
http://dev.mysql.com/doc/refman/5.5/en/group-by-extensions.html
You need to follow the rules for standard GROUP BY. These rules specify that each column mentioned in the SELECT clause must be either (1) also mentioned in the GROUP BY clause or (2) an aggregate function.
There are two problems in your query. One is that you mention p.date instead of c.date in your SELECT clause. That p.date will be null if there aren't any items on the date in question in your phones table. The second item is that you're mentioning p.fromz twice.
I think your basic aggregate query should look like this http://sqlfiddle.com/#!2/0ef4e/3/0:
SELECT p.toz,
COUNT(DISTINCT p.fromz) AS unique_calling_number_count,
c.date
FROM phones AS p
RIGHT OUTER JOIN calendar AS c ON p.date = c.date
GROUP BY p.toz, c.date
ORDER BY c.date, p.toz
That will summarize call-origination numbers by day and call-destination numbers.
Then, when you need to filter your phones records according to some kind of criterion, do that in a subquery as follows.
SELECT p.toz,
COUNT(DISTINCT p.fromz) AS unique_calling_number_count,
c.date
FROM (
SELECT * /* filtering subquery */
FROM phones
WHERE toz = "#myNumber"
) AS p
RIGHT OUTER JOIN calendar AS c ON p.date = c.date
WHERE p.toz = "#myNumber"
GROUP BY p.toz, c.date
ORDER BY c.date, p.toz
Here's an example of this working. http://sqlfiddle.com/#!2/0ef4e/2/0 Thanks to #PatrickQ for noticing the WHERE at the wrong level of the query.