Query to determine cumulative changes to records - sql

Question

Given the following table containing the example rows, I’m looking for a query to give me the aggregate results of changes made to the same record. All changes are made against a base record in another table (results table), so the contents of the results table are not cumulative.
Base Records (from which all changes are made)
Edited Columns highlighted
I’m looking for a query that would give me the cumulative changes (in order by date). This would be the resulting rows:
Any help appreciated!
UPDATE---------------
Let me offer some clarification. The records being edited exist in one table, let's call that [dbo].[Base]. When a person updates a record from [dbo].[Base], his updates go into [dbo].[Updates]. Therefore, a person is always editing from the base table.
At some point, let's say once a day, we need to calculate the sum of changes with the following rule:
For any given record, determine the latest change for each column and take the latest change. If no change was made to a column, take the value from [dbo].[Base]. So, one way of looking at the [dbo].[Updates] table would be to see only the changed columns.
Please let's not discuss the merits of this approach, I realize it's strange. I just need to figure out how to determine the final state of each record.
Thanks!

Answer 1

This is dirty, but you can give this a shot (test here: https://rextester.com/MKSBU15593)
I use a CTE to do an initial CROSS JOIN of the Base and Update tables and then a second to filter it to only the rows where the IDs match. From there I use FIRST_VALUE() for each column, partitioned by the ID value and ordered by a CASE expression (if the Base column value matches the Update column value then 1 else 0) and the Datemodified column to get the most recent version of the each column.
It spits out
CREATE TABLE Base
(
ID INT
,FNAME VARCHAR(100)
,LNAME VARCHAR(100)
,ADDRESS VARCHAR(100)
,RATING INT
,[TYPE] VARCHAR(5)
,SUBTYPE VARCHAR(5)
);
INSERT INTO dbo.Base
VALUES
( 100,'John','Doe','123 First',3,'Emp','W2'),
( 200,'Jane','Smith','Wacker Dr.',2,'Emp','W2');
CREATE TABLE Updates
(
ID INT
,DATEMODIFIED DATE
,FNAME VARCHAR(100)
,LNAME VARCHAR(100)
,ADDRESS VARCHAR(100)
,RATING INT
,[TYPE] VARCHAR(5)
,SUBTYPE VARCHAR(5)
);
INSERT INTO dbo.Updates
VALUES
( 100,'1/15/2019','John','Doe','123 First St.',3,'Emp','W2'),
( 200,'1/15/2019','Jane','Smyth','Wacker Dr.',2,'Emp','W2'),
( 100,'1/17/2019','Johnny','Doe','123 First',3,'Emp','W2'),
( 200,'1/19/2019','Jane','Smith','2 Wacker Dr.',2,'Emp','W2'),
( 100,'1/20/2019','Jon','Doe','123 First',3,'Cont','W2');
WITH merged AS
(
SELECT b.ID AS IDOrigin
,'1/1/1900' AS DATEMODIFIEDOrigin
,b.FNAME AS FNAMEOrigin
,b.LNAME AS LNAMEOrigin
,b.ADDRESS AS ADDRESSOrigin
,b.RATING AS RATINGOrigin
,b.[TYPE] AS TYPEOrigin
,b.SUBTYPE AS SUBTYPEOrigin
,u.*
FROM base b
CROSS JOIN
dbo.Updates u
), filtered AS
(
SELECT *
FROM merged
WHERE IDOrigin = ID
)
SELECT distinct
ID
,FNAME = FIRST_VALUE(FNAME) OVER (PARTITION BY ID ORDER BY CASE WHEN FNAME = FNAMEOrigin THEN 1 ELSE 0 end, datemodified desc)
,LNAME = FIRST_VALUE(LNAME) OVER (PARTITION BY ID ORDER BY CASE WHEN LNAME = LNAMEOrigin THEN 1 ELSE 0 end, datemodified desc)
,ADDRESS = FIRST_VALUE(ADDRESS) OVER (PARTITION BY ID ORDER BY CASE WHEN ADDRESS = ADDRESSOrigin THEN 1 ELSE 0 end, datemodified desc)
,RATING = FIRST_VALUE(RATING) OVER (PARTITION BY ID ORDER BY CASE WHEN RATING = RATINGOrigin THEN 1 ELSE 0 end, datemodified desc)
,[TYPE] = FIRST_VALUE([TYPE]) OVER (PARTITION BY ID ORDER BY CASE WHEN [TYPE] = TYPEOrigin THEN 1 ELSE 0 end, datemodified desc)
,SUBTYPE = FIRST_VALUE(SUBTYPE) OVER (PARTITION BY ID ORDER BY CASE WHEN SUBTYPE = SUBTYPEOrigin THEN 1 ELSE 0 end, datemodified desc)
FROM filtered

Answer 2

Don't you just want the last record?
select e.*
from edited e
where e.datemodified = (select max(e2.datemodified)
from edited e2
where e2.id = e.id
);