Left join on 3 tables - can't get it right - mysql

Question

I'm trying to do a left join over 3 tables with a where clause, and I can't get it to work.
My tables are:
sale:
sale_id (int)
terminal_id (int)
zread_id (int)
... more fields about total amounts, dates times etc
sale_payment:
payment_id (int)
sale_id (int)
payment_type (enum: 'P','W','A')
method_id (int)
payment_amount (Decimal)
sale_writeoff_method:
method_id (int)
description (varchar)
display_order (int)
Sales can be finalised (paid) in 3 different ways (hence the Enum) a physical payment - Cash, Cheque etc, a "Write Off" payment - where stock gets used up at cost price (i.e. Wastage, Giveaways etc) or Accounts - Customer credit etc
At the end of a sales period (end of day) the user does a Z-Read which collects all the transactions that occur and generates a report where the user then has to balance the cash in the drawer etc. When I'm creating the initial Z-Read object, I can collect the information I need by using the query:
SELECT m.method_id, m.description, SUM(s.sale_total) as z_total, COUNT(s.sale_id) as total_sales
FROM sale_writeoff_method m
LEFT JOIN sale_payment p ON m.method_id = p.method_id
LEFT JOIN sale s ON s.sale_id = p.sale_id
WHERE s.zread_id IS NULL
AND (p.payment_type = 'W' OR p.payment_type IS NULL)
AND (s.terminal_id = ? OR s.terminal_id IS NULL)
GROUP BY p.payment_type, m.method_id
ORDER BY m.display_order;
Once everything is balanced and finalised, all of the collected sales for all types in the sales table are flagged with the zread_id that results from inserting this object.
Now my issue is, when I need to re-create the z-read object in the future, for re-printing reports for example, I can't get all the method_id's and descriptions to show - the query I've been using is:
SELECT m.method_id, m.description, SUM(s.sale_total) as z_total, COUNT(s.sale_id) as total_sales
FROM sale_writeoff_method m
LEFT JOIN sale_payment p ON m.method_id = p.method_id
LEFT JOIN sale s ON s.sale_id = p.sale_id
WHERE s.zread_id = 1
AND (p.payment_type = 'W' OR p.payment_type IS NULL)
GROUP BY p.payment_type, m.method_id
ORDER BY m.display_order;
But it only displays methods that had sales attached for that Z-Read period. I can't use WHERE s.zread_id IS NULL because that will include all the sales that haven't been finalised yet.
Any suggestions would be greatly appreciated!

Answer 1

The problem is that a left join returns nulls for the joined column values where no matching row is found, but you are checking those column values in the where clause, but the where predicates are is executed after all the rows are joined, so they'll never match and your outer joins are sunk.
ie, this example query:
select *
from table1 t1
left join table2 t2 on t2.fk = t1.id
where t2.col1 = 'x'
will never return any rows that don't have corresponding rows in table2, because col1 will be null, and comparisons with null are always false, except for col1 is null.
To fix this, you need to move the tests into the ON clause, so the comparison happens while the join is being made, like this:
select *
from table1 t1
left join table2 t2 on t2.fk = t1.id and t2.col1 = 'x'
Now the left join will still return rows while matching on the key, and applying extra predicates to further refine the match.
In your case, you are doing an outer (ie left) join to sale_payment p, but testing p.payment_type = 'W' in the where clause, which won't work.
Here's the fixed query, with the tests on the left-joined tables made in the ON clause:
SELECT
m.method_id,
m.description,
SUM(s.sale_total) as z_total,
COUNT(s.sale_id) as total_sales
FROM sale_writeoff_method m
LEFT JOIN sale_payment p ON m.method_id = p.method_id AND p.payment_type = 'W'
LEFT JOIN sale s ON s.sale_id = p.sale_id AND s.terminal_id = ?
GROUP BY m.method_id, m.description
ORDER BY m.display_order;
Note that I also removed the group by p.payment_type, because you haven't selected that column, and I added group by m.description, because you have selected that.
You may need to fine tune the query, but hopefully this will be pretty close