SQL Unpivot / Coalesce multiple columns into one column based on values - sql

Question

I am trying to unpivot / coalesce multiple columns into one value, based on values in the target columns.
Given the following sample data:
CREATE TABLE SourceData (
Id INT
,Column1Text VARCHAR(10)
,Column1Value INT
,Column2Text VARCHAR(10)
,Column2Value INT
,Column3Text VARCHAR(10)
,Column3Value INT
)
INSERT INTO SourceData
SELECT 1, NULL, NULL, NULL, NULL, NULL, NULL UNION
SELECT 2, 'Text', 1, NULL, NULL, NULL, NULL UNION
SELECT 3, 'Text', 2, 'Text 2', 1, NULL, NULL UNION
SELECT 4, NULL, NULL, NULL, 1, NULL, NULL
I am trying to produce the following result set:
Id ColumnText
----------- ----------
1 NULL
2 Text
3 Text 2
4 NULL
Where ColumnXText column values become one "ColumnText" value per row, based on the following criteria:
If all ColumnX columns are NULL, then ColumnText = NULL
If a ColumnXValue value is "1" and the ColumnXText IS NULL, then
ColumnText = NULL
If a ColumnXValue value is "1" and the ColumnXText IS NOT NULL, then
ColumnText = ColumnXText.
There are no records with more than one ColumnXValue of "1".
What I'd tried is in this SQL Fiddle # http://sqlfiddle.com/#!6/f2e18/2
I'd tried (shown in SQL fiddle):
Unpivioting with CROSS / OUTER APPLY. I fell down on this approach because I was not able to get WHERE conditions to produce the expected results.
I'd also tried using UNPIVOT, but had no luck.
I was thinking of a brute-force approach that did not seem to be correct. The real source table has 44MM rows. I do not control the schema of the source table.
Please let me know if there's a simpler approach than a brute-force tangle of CASE WHENs. Thank you.

Answer 1

I don't think there is much mileage in trying to be too clever with this
SELECT
Id,
CASE
WHEN COLUMN1VALUE = 1 THEN COLUMN1TEXT
WHEN COLUMN2VALUE = 1 THEN COLUMN2TEXT
WHEN COLUMN3VALUE = 1 THEN COLUMN3TEXT
End as ColumnText
From
Sourcedata
I did have them in 321 order, but considered that the right answer might be hit sooner if the checking is done in 123 order instead (fewer checks, if there are 44million rows, might be significant)

Answer 2

Considering you have 44 million rows, you really don't want to experiment to much to join table on itself with apply or something like that. You need just go through it once, and that's best with simple CASE, what you call "brute-force" approach:
SELECT
Id
, CASE WHEN Column1Value = 1 THEN Column1Text
WHEN Column2Value = 1 THEN Column2Text
WHEN Column3Value = 1 THEN Column3Text
END AS ColumnText
FROM SourceData
But, if you really want to get fancy and write something without case, you could use UNION to merge different columns into one, and then join on it:
wITH CTE_Union AS
(
SELECT Id, Column1Text AS ColumnText, Column1Value AS ColumnValue
FROM SourceData
UNION ALL
SELECT Id, Column2Text, Column2Value FROM SourceData
UNION ALL
SELECT Id, Column3Text, Column3Value FROM SourceData
)
SELECT s.Id, u.ColumnText
FROM SourceData s
LEFT JOIN CTE_Union u ON s.Id = u.id and u.ColumnValue = 1
But I guarantee first approach will outperform this by a margin of 4 to 1

Answer 3

If you do not want to use a case expression, then you can use another outer apply() on a common table expression (or subquery/derived table) of your original unpivot with outer apply():
;with cte as (
select s.Id, oa.ColumnText, oa.ColumnValue
from sourcedata s
outer apply (values
(s.Column1Text, s.Column1Value)
, (s.Column2Text, s.Column2Value)
, (s.Column3Text, s.Column3Value)
) oa (ColumnText, ColumnValue)
)
select s.Id, x.ColumnText
from sourcedata s
outer apply (
select top 1 cte.ColumnText
from cte
where cte.Id = s.Id
and cte.ColumnValue = 1
) x
rextester demo: http://rextester.com/TMBR41346
returns:
+----+------------+
| Id | ColumnText |
+----+------------+
| 1 | NULL |
| 2 | Text |
| 3 | Text 2 |
| 4 | NULL |
+----+------------+

Answer 4

This will give you the first non-null text value in order. It seems as if this is what you are trying to accomplish.
select ID, Coalesce(Column3Text,Column2Text,Column1Text) ColumnText
from SourceData