Converting updated column values to a table as rows - sql

Question

ID State Name Department City
1 O George Sales Phoenix
1 N George Sales Denver
2 O Michael Order Process San diego
2 N Michael Marketing San jose
I got a situation that I need to convert the above tables values to the following format.(Consider the top row is column names)
ID Column OldValue New Value
1 Department Phoenix Denver
2 Department Order Process Marketing
2 City San diego San jose
I.e : I need to capture the changed column values for a table from its old and new records and record them in a different table.But the problem is we have many tables like that and the column names and no of columns are different for each table.
If anyone come with a solution that would be greatly appreciated..!
Thank you in advance.

Answer 1

Is this what you want?
ID Column OldValue New Value
1 City Phoenix Denver
2 Department Order Process Marketing
2 City San Diego San jose
Here is the dynamic code:
DECLARE #sqlStm varchar(max);
DECLARE #sqlSelect varchar(max);
DECLARE #tablename varchar(200);
SET #tablename = 'testtable';
-- Assume table has ID column and State column.
SET #sqlSelect = ''
SET #sqlStm = 'WITH old AS
(
SELECT *
FROM '+#tablename+'
WHERE State=''O''
), new AS
(
SELECT *
FROM '+#tablename+'
WHERE State=''N''
)';
DECLARE #aCol varchar(128)
DECLARE curCols CURSOR FOR
SELECT column_name
FROM information_schema.columns
WHERE table_name = #tablename
AND UPPER(column_name) NOT IN ('ID','STATE')
OPEN curCols
FETCH curCols INTO #aCol
WHILE (##FETCH_STATUS = 0)
BEGIN
SET #sqlStm = #sqlStm +
', changed'+#aCol+' AS
(
SELECT n.ID, '''+#aCol+''' AS [Column], o.['+#aCol+'] AS oldValue, n.['+#aCol+'] AS newValue
FROM new n
JOIN old o ON n.ID = o.ID AND n.['+#aCol+'] != o.['+#aCol+']
)'
IF LEN(#sqlSelect) > 0 SET #sqlSelect = #sqlSelect + ' UNION ALL '
SET #sqlSelect = #sqlSelect + '
SELECT * FROM changed'+#aCol
FETCH curCols INTO #aCol
END
CLOSE curCols
DEALLOCATE curCols
SET #sqlSelect = #sqlSelect + '
ORDER BY id, [Column]'
PRINT #sqlStm+#sqlSelect
EXEC (#sqlStm+#sqlSelect)
Which in my test output the following:
WITH old AS
(
SELECT *
FROM testtable
WHERE State='O'
), new AS
(
SELECT *
FROM testtable
WHERE State='N'
), changedName AS
(
SELECT n.ID, 'Name' AS [Column], o.[Name] AS oldValue, n.[Name] AS newValue
FROM new n
JOIN old o ON n.ID = o.ID AND n.[Name] != o.[Name]
), changedDepartment AS
(
SELECT n.ID, 'Department' AS [Column], o.[Department] AS oldValue, n.[Department] AS newValue
FROM new n
JOIN old o ON n.ID = o.ID AND n.[Department] != o.[Department]
), changedCity AS
(
SELECT n.ID, 'City' AS [Column], o.[City] AS oldValue, n.[City] AS newValue
FROM new n
JOIN old o ON n.ID = o.ID AND n.[City] != o.[City]
)
SELECT * FROM changedName UNION ALL
SELECT * FROM changedDepartment UNION ALL
SELECT * FROM changedCity
ORDER BY id, [Column]
Original answer below:
I would do it like this -- because I think it is clearer than other ways which might be faster:
with old as
(
Select ID, Name,Department,City
From table1
Where State='O'
), new as
(
Select ID, Name,Department,City
From table1
Where State='N'
), oldDepartment as
(
Select ID, 'Department' as Column, o.Department as oldValue, n.Department as newValue
From new
join old on new.ID = old.ID and new.Department != old.Department
), oldCity as
(
Select ID, 'City' as Column, o.City as oldValue, n.City as newValue
From new
join old on new.ID = old.ID and new.City != old.City
)
select * from oldDepartment
union all
select * from oldCity
Depending on many things (size of tables and indexes etc) it might actually be faster than using pivots or cases or grouping. It really depends on your data. If this is a one-off run I'd just go for the easiest to grok.

Answer 2

The cleanest approach is probably to unpivot the data and then use aggregation. This does require custom coding for each table, which you might be able to generalize by using some form a dynamic SQL.
For your particular example, here is an illustration of what to do:
select id, col,
max(case when OldNew = 'Old' then value end) as OldValue,
max(case when OldNew = 'New' then value end) as NewValue
from ((select ID, OldNew, 'Name' as col, Name as value
from t
) union all
(select ID, OldNew, 'Department' as col, Department as value
from t
) union all
(select ID, OldNew, 'City' as col, City as value
from t
)
) unpvt
group by id, col
having max(value) <> min(value) and max(value) is not null;
This is for illustration purposes. The unpivot can be done more efficiently than using union all, particularly when there are many scans. Here is a more efficient version, although the exact syntax depends on the database:
select id, col,
max(case when OldNew = 'Old' then value end) as OldValue,
max(case when OldNew = 'New' then value end) as NewValue
from (select ID, OldNew, cols.col,
(case when cols.col = 'Name' then Name
when cols.col = 'Department' then Department
when cols.col = 'City' then City
end) as value
from t cross join
(select 'Name' as col union all select 'Department' union all select 'City') cols
) unpvt
group by id, col
having max(value) <> min(value) and max(value) is not null;
This is more efficient because it will typically only scan your table once, rather than once for each column as in the union all version.
In either version, there is an implicit assumption that all the columns have the same character type. This is implicit in the format you are converting to, where all the values are in a single column.