Joining poorly designed SQL tables? - sql-server

Question

I've tried searching for information on joining tables without foriegn keys but it seems the answer is always to create the foreign key. I cannot modify the tables in question to do this and I must report on data that is already in production. The following is a portion of the data in the tables involved in order to exemplify the issue.
Table A
Journal Account Debit Credit Sequence
--------------------------------------------------
87041 150-00 100.00 0.00 16384
87041 150-10 0.00 100.00 32768
87041 150-00 50.0 0.0 49152
87041 210-90 0.0 50.0 65536
Then the second table, tracking additional bits of information, is largely the same but missing the Sequence number that would tie the line items together properly. It has its own Sequence Number that is unrelated.
Table B
Journal Account Label Artist Sequence
--------------------------------------------------
87041 150-00 Label02 Artist12 1
87041 150-10 Label09 Artist03 2
87041 150-00 Label04 Artist01 3
87041 210-90 Label01 Artist05 4
At present the best I can come up with is to join on Journal and Account but that duplicates records. I have gotten close by playing around with grouping and max() on sequence number but the result has been that that not all duplicates are removed for journal entries with a very large number of rows and the first match from the second table is always displayed for lines that have the same account.
Closest - but bad - result
Journal Account Debit Credit Sequence Label Artist
----------------------------------------------------------------------
87041 150-00 100.00 0.00 16384 Label02 Artist12
87041 150-10 0.00 100.00 32768 Label09 Artist03
87041 150-00 50.0 0.0 49152 Label02 Artist12 <-- wrong
87041 210-90 0.0 50.0 65536 Label01 Artist05
How can I join the tables such that duplicates are excluded but also so that the correct Label and Artist are displayed? It sort of feels like I have to produce a query which knows that one of the records from Table B has already been used when the 49152 record from Table A comes looking for a match.
EDIT:
#Justin Crabtree A.Sequence will be the order in which the line items were entered. So a user could have entered the last line in the example first, then the first line, then the third, and finally the second.
#Edper Microsoft SQL Server...hmm, I cannot remote into the client's machine this morning...otherwise I would provide the version.
#Abe Miessler yes, you are correct.
As soon as I can get back into the server I will try your suggestion #pkuderov

Answer 1

Try this
;WITH a AS
(
SELECT Journal,
Account,
Debit,
Credit,
Sequence,
Id = ROW_NUMBER() OVER(PARTITION BY Journal ORDER BY Sequence)
FROM dbo.tablea
)
, b AS
(
SELECT Journal,
Account,
Label,
Artist,
Id = ROW_NUMBER() OVER(PARTITION BY Journal ORDER BY Sequence)
FROM dbo.tableb
)
SELECT a.Journal,
a.Account,
a.Debit,
a.Credit,
a.Sequence,
b.Label,
b.Artist
FROM a
JOIN b ON b.Journal = a.Journal
AND b.Account = a.Account
AND b.Id = a.Id

Answer 2

Hi, that's just an idea:
select
a.Journal, a.Account, a.Debit, a.Credit, a.Sequence, b.Label, b.Artist
from (
select
*,
row_number() over(partition by Journal, Account order by Sequence) as idInGroup
from a
) as a
join (
select
*,
row_number() over(partition by Journal, Account order by Sequence) as idInGroup
from b
) as b on
a.Journal = b.Journal
and a.Account = b.Account
and a.idInGroup = b.idInGroup
Here I assume that orders appeared in Sequence order (in both tables) and that's the base hint for join tables.

Answer 3

If you ordered the 2 table rows by their own sequence numbers, will the rows align in the same order?
If so, this is a possible solution for SQL server:
You can create 2 CTEs, one for each table, with ROW_NUMBER column, and that way, both tables will have a matching row number column that you can use to join. Let me know if you need an example.

Answer 4

If I'm reading your requirements correctly and you want all rows from Table A, but only the first matching row from Table B, your best bet would be to do an OUTER APPLY with a TOP(1). That would look something like this:
select *
from TableA
OUTER APPLY
(select TOP(1) Journal, Account, Label, Artist, Sequence
FROM TableB
WHERE Journal = TableA.Journal AND Account = TableA.Account
ORDER BY Sequence) as B
(Definitely pseudo-code, but that should be somewhat close.)
If it comes down to it, you could use ROW_NUMBER(), partition that by Journal and Account and then match on those Row_Number values for each result set. You'd generate one sub-query/CTE for TableA and another CTE for TableB - each with a RowNumber value that would be essentially a new sequence integer. The first row in TableA would match the first row in TableB, Second row in TableA would match the second in TableB, etc. Of course, you'd run into some issues if there are more rows for Journal/Account in "A" than there are in "B".
A better question might be - "How does your code determine all matches between TableA and TableB if they can't use any data columns to tie them together?"