We have a database table TBLTEST. Below is part of its data:
The data is ordered by date. We are trying to group rows by the first five columns, convert dates in each group into an interval, and record the ending date of the last record as the infinite date 99991230. Below is the desired result:
SELECT
code1,
code2,
code3,
rate,
value,
min(MONTH) start_dt,
CASE
WHEN ROW_NUMBER() OVER(PARTITION BY code1, code2, code3 ORDER BYmax(MONTH) DESC) = 1 THEN 99991230
ELSE max(MONTH)
END end_dt
FROM
(
SELECT
t.*,
ROW_NUMBER() OVER(PARTITION BY code1, code2, code3 ORDER BY MONTH) rn1,
ROW_NUMBER() OVER(PARTITION BY code1, code2, code3, rate, value ORDERBY MONTH) rn2
FROM
TBLTEST t
) t
GROUP BY
code1,
code2,
code3,
rate,
value,
rn1 - rn2
ORDER BY
start_dt
It is rather simple to perform the task in the natural way of thinking. We compare neighboring values between rows on the first five columns, and put the current one and the previous row in the same group when values are same, or create a new group if they are different until the last record is compared. As SQL set is unordered, we need to first invent two columns of indexes manually in an extremely complicated way and then perform grouping according to the relationship between the two columns of indexes. You need to be really smart to come up with the solution.
Yet it is easy to write the code using the open-source esProc SPL:
SPL supports ordered sets directly, making it easy to perform grouping when a neighboring value is different.
Top comments (0)