Trying to merge rows into one row with certain conditions - sql

Question

Given 2 or more rows that are selected to merge, one of them is identified as being the template row. The other rows should merge their data into any null value columns that the template has.
Example data:
Id Name Address City State Active Email Date
1 Acme1 NULL NULL NULL NULL blah#yada.com 3/1/2011
2 Acme1 1234 Abc Rd Springfield OR 0 blah#gmail.com 1/12/2012
3 Acme2 NULL NULL NULL 1 blah#yahoo.com 4/19/2012
Say that a user has chosen row with Id 1 as the template row, and rows with Ids 2 and 3 are to be merged into row 1 and then deleted. Any null value columns in row Id 1 should be filled with (if one exists) the most recent (see Date column) non-null value, and non-null values already present in row Id 1 are to be left as is. The result of this query on the above data should be exactly this:
Id Name Address City State Active Email Date
1 Acme1 1234 Abc Road Springfield OR 1 blah#yada.com 3/1/2011
Notice that the Active value is 1, and not 0 because row Id 3 had the most recent date.
P.S. Also, is there any way possible to do this without explicitly defining/knowing beforehand what all the column names are? The actual table I'm working with has a ton of columns, with new ones being added all the time. Is there a way to look up all the column names in the table, and then use that subquery or temptable to do the job?

Answer 1

You might do it by ordering rows first by template flag, then by date desc. Template row should always be the last one. Each row is assigned a number in that order. Using max() we are finding fist occupied cell (in descending order of numbers). Then we select columns from rows matching those maximums.
; with rows as (
select test.*,
-- Template row must be last - how do you decide which one is template row?
-- In this case template row is the one with id = 1
row_number() over (order by case when id = 1 then 1 else 0 end,
date) rn
from test
-- Your list of rows to merge goes here
-- where id in ( ... )
),
-- Finding first occupied row per column
positions as (
select
max (case when Name is not null then rn else 0 end) NamePosition,
max (case when Address is not null then rn else 0 end) AddressPosition,
max (case when City is not null then rn else 0 end) CityPosition,
max (case when State is not null then rn else 0 end) StatePosition,
max (case when Active is not null then rn else 0 end) ActivePosition,
max (case when Email is not null then rn else 0 end) EmailPosition,
max (case when Date is not null then rn else 0 end) DatePosition
from rows
)
-- Finally join this columns in one row
select
(select Name from rows cross join Positions where rn = NamePosition) name,
(select Address from rows cross join Positions where rn = AddressPosition) Address,
(select City from rows cross join Positions where rn = CityPosition) City,
(select State from rows cross join Positions where rn = StatePosition) State,
(select Active from rows cross join Positions where rn = ActivePosition) Active,
(select Email from rows cross join Positions where rn = EmailPosition) Email,
(select Date from rows cross join Positions where rn = DatePosition) Date
from test
-- Any id will suffice, or even DISTINCT
where id = 1
You might check it at Sql Fiddle.
EDIT:
Cross joins in last section might actually be inner joins on rows.rn = xxxPosition. It works this way, but change to inner join would be an improvement.

Answer 2

It's not so complicated.
At first..
DECLARE #templateID INT = 1
..so you can remember which row is treated as template..
Now find latest NOT NULL values (exclude template row). The easiest way is to use TOP 1 subqueries for each column:
SELECT
(SELECT TOP 1 Name FROM DataTab WHERE Name IS NOT NULL AND NOT ID = #templateID ORDER BY Date DESC) AS LatestName,
(SELECT TOP 1 Address FROM DataTab WHERE Address IS NOT NULL AND NOT ID = #templateID ORDER BY Date DESC) AS AddressName
-- add more columns here
Wrap above into CTE (Common Table Expression) so you have nice input for your UDPATE..
WITH Latest_CTE (CTE_LatestName, CTE_AddressName) -- add more columns here; I like CTE prefix to distinguish source columns from target columns..
AS
-- Define the CTE query.
(
SELECT
(SELECT TOP 1 Name FROM DataTab WHERE Name IS NOT NULL AND NOT ID = #templateID ORDER BY Date DESC) AS LatestName,
(SELECT TOP 1 Address FROM DataTab WHERE Address IS NOT NULL AND NOT ID = #templateID ORDER BY Date DESC) AS AddressName
-- add more columns here
)
UPDATE
<update statement here (below)>
Now, do smart UPDATE of your template row using ISNULL - it will act as conditional update - update only if target column is null
WITH
<common expression statement here (above)>
UPDATE DataTab
SET
Name = ISNULL(Name, CTE_LatestName), -- if Name is null then set Name to CTE_LatestName else keep Name as Name
Address = ISNULL(Address, CTE_LatestAddress)
-- add more columns here..
WHERE ID = #templateID
And the last task is delete rows other then template row..
DELETE FROM DataTab WHERE NOT ID = #templateID
Clear?

Answer 3

For dynamic columns, you need to write a solution using dynamic SQL.
You can query sys.columns and sys.tables to get the list of columns you need, then you want to loop backwards once for each null column finding the first non-null row for that column and updating your output row for that column. Once you get to 0 in the loop you have a complete row which you can then display to the user.

Answer 4

I should pay attention to posting dates. In any case, here's a solution using dynamic SQL to build out an update statement. It should give you something to build from, anyway.
There's some extra code in there to validate the results along the way, but I tried to comment in a way that made that non-vital code apparent.
CREATE TABLE
dbo.Dummy
(
[ID] int ,
[Name] varchar(30),
[Address] varchar(40) null,
[City] varchar(30) NULL,
[State] varchar(2) NULL,
[Active] tinyint NULL,
[Email] varchar(30) NULL,
[Date] date NULL
);
--
INSERT dbo.Dummy
VALUES
(
1, 'Acme1', NULL, NULL, NULL, NULL, 'blah#yada.com', '3/1/2011'
)
,
(
2, 'Acme1', '1234 Abc Rd', 'Springfield', 'OR', 0, 'blah#gmail.com', '1/12/2012'
)
,
(
3, 'Acme2', NULL, NULL, NULL, 1, 'blah#yahoo.com', '4/19/2012'
);
DECLARE
#TableName nvarchar(128) = 'Dummy',
#TemplateID int = 1,
#SetStmtList nvarchar(max) = '',
#LoopCounter int = 0,
#ColumnCount int = 0,
#SQL nvarchar(max) = ''
;
--
--Create a table to hold the column names
DECLARE
#ColumnList table
(
ColumnID tinyint IDENTITY,
ColumnName nvarchar(128)
);
--
--Get the column names
INSERT #ColumnList
(
ColumnName
)
SELECT
c.name
FROM
sys.columns AS c
JOIN
sys.tables AS t
ON
t.object_id = c.object_id
WHERE
t.name = #TableName;
--
--Create loop boundaries to build out the SQL statement
SELECT
#ColumnCount = MAX( l.ColumnID ),
#LoopCounter = MIN (l.ColumnID )
FROM
#ColumnList AS l;
--
--Loop over the column names
WHILE #LoopCounter <= #ColumnCount
BEGIN
--Dynamically construct SET statements for each column except ID (See the WHERE clause)
SELECT
#SetStmtList = #SetStmtList + ',' + l.ColumnName + ' =COALESCE(' + l.ColumnName + ', (SELECT TOP 1 ' + l.ColumnName + ' FROM ' + #TableName + ' WHERE ' + l.ColumnName + ' IS NOT NULL AND ID <> ' + CAST(#TemplateID AS NVARCHAR(MAX )) + ' ORDER BY Date DESC)) '
FROM
#ColumnList AS l
WHERE
l.ColumnID = #LoopCounter
AND
l.ColumnName <> 'ID';
--
SELECT
#LoopCounter = #LoopCounter + 1;
--
END;
--TESTING - Validate the initial table values
SELECT * FROM dbo.Dummy ;
--
--Get rid of the leading common in the SetStmtList
SET #SetStmtList = SUBSTRING( #SetStmtList, 2, LEN( #SetStmtList ) - 1 );
--Build out the rest of the UPDATE statement
SET #SQL = 'UPDATE ' + #TableName + ' SET ' + #SetStmtList + ' WHERE ID = ' + CAST(#TemplateID AS NVARCHAR(MAX ))
--Then execute the update
EXEC sys.sp_executesql
#SQL;
--
--TESTING - Validate the updated table values
SELECT * FROM dbo.Dummy ;
--
--Build out the DELETE statement
SET #SQL = 'DELETE FROM ' + #TableName + ' WHERE ID <> ' + CAST(#TemplateID AS NVARCHAR(MAX ))
--Execute the DELETE
EXEC sys.sp_executesql
#SQL;
--
--TESTING - Validate the final table values
SELECT * FROM dbo.Dummy;
--
DROP TABLE dbo.Dummy;