MS SQL Store Procedure to Merge Multiple Rows into Single Row based on Variable Table and Column Names - sql

Question

I'm working with MS SQL Server 2008. I'm trying to create a stored procedure to Merge (perhaps) several rows of data (answers) into a single row on target table(s). This uses a 'table_name' field and 'column_name' field from the answers table. The data looks like something like this:
answers table
--------------
id int
table_name varchar
column_name varchar
answer_value varchar
So, the target table (insert/update) would come from the 'table_name'. Each row from the anwsers would fill one column on the target table.
table_name_1 table
--------------
id int
column_name_1 varchar
column_name_2 varchar
column_name_3 varchar
etc...
Note, there can be many target tables (variable from answers table: table_name_1, table_name_2, table_name_3, etc.) that insert into many columns (column_name_1...2...3) on each target table.
I thought about using a WHILE statement to loop through the answers table. This could build a variable which would be the insert/update statement(s) for the target tables. Then executing those statements somehow. I also noticed Merge looks like it might help with this problem (select/update/insert), but my MS SQL Stored Procedure experience is very little. Could someone suggestion a strategy or solution to this problem?
Note 6/23/2014: I'm considering using a single Merge statement, but I'm not sure it is possible.

Answer 1

I'm probably missing something, but the basic idea to solve the problem is to use meta-programming, like a dynamic pivot.
In this particular case there is another layer to make the solution more difficult: the result need to be in different execution instead of beeing grouped.
The backbone for a possible solution is
DECLARE #cols AS NVARCHAR(MAX)
DECLARE #query AS NVARCHAR(MAX)
--using a cursor on SELECT DISTINCT table_name FROM answers iterate:
--*Cursor Begin Here*
--mock variable for the first value of the cursor
DECLARE #table AS NVARCHAR(MAX) = 't1'
-- Column list
SELECT #cols = STUFF((SELECT distinct
',' + QUOTENAME(column_name)
FROM answers with (nolock)
WHERE table_name = #table
FOR XML PATH(''), TYPE
).value('.', 'NVARCHAR(MAX)')
, 1, 1, '')
--Query definition
SET #query = '
SELECT ' + #cols + '
INTO ' + #table + '
FROM (SELECT column_name, answer_value
FROM answers
WHERE table_name = ''' + #table + ''') b
PIVOT (MAX(answer_value) FOR column_name IN (' + #cols + ' )) p '
--select #query
EXEC sp_executesql #query
--select to verify the execution
--SELECT * FROM t1
--*Cursor End Here*
SQLFiddle Demo
The cursor definition is omitted, because I'm not sure if it'll work on SQLFiddle
In addition to the template for a Dynamic Pivot the columns list is filtered by the new table name, and in the query definition there is a SELECT ... INTO instead of a SELECT.
This script does not account for table already in the database, if that's a possibility the query can be divided in two:
SET #query = '
SELECT TOP 0 ' + #cols + '
INTO ' + #table + '
FROM (SELECT column_name, answer_value
FROM answers
WHERE table_name = ''' + #table + ''') b
PIVOT (MAX(answer_value) FOR column_name IN (' + #cols + ' )) p '
to create the table without data, if needed, and
SET #query = '
INSERT INTO ' + #table + '(' + #cols + ')'
SELECT ' + #cols + '
FROM (SELECT column_name, answer_value
FROM answers
WHERE table_name = ''' + #table + ''') b
PIVOT (MAX(answer_value) FOR column_name IN (' + #cols + ' )) p '
or a MERGE to insert/update the values in the table.
Another possibility will be to DROP and recreate every table.

Answer 2

Approach I took to this complex problem:
Create several temporary tables to work with your data
Select and populate the temporary tables with the data
Use dynamic pivoting to pivot the rows into one row
Use a CURSOR with WHILE loop for multiple table entries
SET #query with the dynamically built MERGE statement
EXECUTE(#query)
Drop temporary tables