DECODE - flipping Table axes

Question

Hi,
To obtain this unique row:
CUSTOMER   SUBSCRIPTION_A   SUBSCRIPTION_B    SUBSCRIPTION_C
J. Smith                       01-JUN-2002       01-JUN-2002from this set of multiple rows:
CUSTOMER   SUBSCRIPTION_TYPE   BEGIN_DATE
J.Smith                    B   01-JUN-2002 
J.Smith                    C   01-JUN-2002I'm using this artefact:
SELECT
customer,
MAX(DECODE(subscription_type, 'A', begin_date)),
MAX(DECODE(subscription_type, 'B', begin_date)),
MAX(DECODE(subscription_type, 'C', begin_date))
FROM customer, subscriptions
WHERE customer.customer = subscriptions.customer
GROUP BY customerThe solely purpose of the MAX function here (one might be puzzled to see it here)is its side effect: flattening the otherwise multiple rows returned set, into a unique, has-it-all row by customer.
This is often described as "flipping table axes" (vertical to horizontal), though the Oracle 8 Reference book covers only half of the way: provides no information as how to obtain this single row.
My questions:
- is there another syntax for the same functional result?
- if so how does it compare in terms of performances?
Thank You!
JRoch

Answer 1

One little change in your SQL above,
When you DECODE the subscription_type(type), IF the type is 'A' then do some thing, else do NOTHING(NULL) to get the expected results. So a small tweak to your code,
SELECT
cust.customer,
MAX(DECODE(sub.type, 'A', sub.begin_date, NULL)),
MAX(DECODE(sub.type, 'B', sub.begin_date, NULL)),
MAX(DECODE(sub.type, 'C', sub.begin_date, NULL))
FROM customer cust, subscriptions sub
WHERE cust.customer = sub.customer
GROUP BY cust.customer;Now, the alternative for this logic would be PL/SQL. If you do not know the number of rows ahead of time, PL/SQL would be the only option in fact. In this query, we are assuming that we can get a maximun of three rows, one for each TYPE. In that case, performance is SQL versus PL/SQL, my personal choice is SQL as it is advised by many Oracle experts.
Also, you could use SUM() versus MAX() if the decode is on the amounts, that should slightly improve the performance. Basic thing here is, we need to use an aggregate function in order to group the rows and flip them into columns.
I will be interested to see any other SQL logic for this spec. Thanks,
SriDHAR