How to use a subselect in a LEFT JOIN ON clause? - sql

Question

I have a table t with
ORD_DATE
ORD_ID
ORD_REF
ORD_TYPE1
ORD_TYPE2
PRODNUM
PRODQUAL
PRICE
2020-09-01
101
101
ORDER
ORDER
456
F
555
2020-09-02
102
101
CONF
ORDER
456
F
555
2020-11-30
103
102
ORDER
ORDER
123
K
444
2020-12-01
104
102
CONF
ORDER
123
K
444
2020-12-01
105
103
ORDER
ORDER
123
K
444
2020-12-01
106
104
ORDER
ORDER
123
K
333
2020-12-02
107
104
CONF
ORDER
123
K
333
2020-12-08
108
104
CONF
RETURN
123
K
-333
2020-12-01
109
105
ORDER
ORDER
123
F
222
2020-12-02
110
105
CONF
ORDER
123
F
222
and a table s with:
ORD_DATE
PROD_NUMBER
PROD_QUAL
2020-12-01-00.00.00.000000
123
K
2020-12-01-00.00.00.000000
123
L
In table t are all sales per day.
A sale has 2 stages: first the order is generated when the customer buys something
("ORDER"/"ORDER"). Then it gets confirmed which is at the next day or within the next days normally ("CONF"/"ORDER"). If a customer sends the product back it's a return ("CONF"/"RETURN").
In table s are the products that are "second hand".
if a product is in that table it means all sales from table t with
ORDER_TYPE_1 = "ORDER"
AND ORDER_TYPE_2 = "ORDER"
AND t.ORD_DATE >= s.ORD_DATE
AND t.PROD_NUMBER = s.PROD_NUMBER
AND t.PROD_QUAL = s.PROD_QUAL
count as "second hand".
I need the sum of all "second hand" sales that are confirmed from the year 2021 and month 12. But only rows with CONF/ORDER or CONF/RETURN should be in the calculation. I have CAL_YEAR and CAL_MONTH in table t for that (omitted for less clutter).
From table t only ORDER_REF 105 matches that and the sum would be 0 because only these 2 rows matter:
| 2020-12-02 | 107 | 104 | CONF | ORDER | 123 | K | 333
| 2020-12-08 | 108 | 104 | CONF | RETURN | 123 | K | -333
My code so far:
SELECT SUM(PRICE)
FROM t
--
LEFT JOIN s
ON t.PRODNUM = s.PRODNUM
AND t.PRODQUAL = s.PRODQUAL
AND (SELECT ORD_DATE FROM t WHERE ORDER_TYPE_1 = 'ORDER' AND ORDER_TYPE_2 = 'ORDER') >= s.ORD_DATE
--
WHERE CAL_YEAR = 2021
AND CAL_MONTH = 12
AND ORDER_TYPE_1 = 'CONF'
AND ORDER_TYPE_2 IN ('ORDER', 'RETURN')
--
GROUP BY PRICE
;
SQL-Error: "single-row subquery returns more than one row
My problem is limiting the LEFT JOIN to ORDER/ORDER (so that ORDER_REF 105 is in) but only use CONF/ORDER and CONF/RETURN for the sum (so that ORDER_REF 102 is out).
Anyone can help?

Answer 1

The simplest way I can think of would be to do a self-join, where you join a second copy of table t aliased t2 to use for the CONF/ORDER and CONF/RETURN rows, while you use t for the ORDER/ORDER rows.
SELECT SUM(t2.PRICE)
FROM t
--
INNER JOIN t t2
ON t2.ORD_REF = t.ORD_REF
AND t2.ORDER_TYPE_1 = 'CONF'
AND t2.ORDER_TYPE_2 IN ('ORDER', 'RETURN')
--
LEFT JOIN s
ON t.PRODNUM = s.PRODNUM
AND t.PRODQUAL = s.PRODQUAL
AND t.ORD_DATE >= s.ORD_DATE
--
WHERE t.CAL_YEAR = 2021
AND t.CAL_MONTH = 12
AND t.ORDER_TYPE_1 = 'ORDER'
AND t.ORDER_TYPE_2 = 'ORDER'
;
If you need it to be more efficient, you could use analytic/window functions to pull the summed price from the CONF rows into the ORDER/ORDER row as a new column. This way it will only query table t once instead of twice.
SELECT SUM(t2.order_price_sum)
FROM (select t.*,
sum(case when ORDER_TYPE_1 = 'CONF'
AND ORDER_TYPE_2 IN ('ORDER', 'RETURN')
then t.price
else 0 end) over (partition by ord_ref) as order_price_sum
from t) t2
--
LEFT JOIN s
ON t2.PRODNUM = s.PRODNUM
AND t2.PRODQUAL = s.PRODQUAL
AND t2.ord_date >= s.ORD_DATE
--
WHERE CAL_YEAR = 2021
AND CAL_MONTH = 12
AND ORDER_TYPE_1 = 'ORDER'
AND ORDER_TYPE_2 = 'ORDER'
;