combining rows in same table sql - mysql

Question

sample_data
id_type,seq_no,acct_name,_acct#,address
12345,67,jiimm,,167 s.40th st
12345,67,jiimm joe the 3rd,,167 s.40th st
12345,67,jiimm
12345,67,,0981_1,po box 1234
12345,80,Lee,,1234 street ave
12345,80,Lee
12345,80,,588_1,109 road st
CODE
SELECT `ID`_type,
seq_no,
MAX(`acct_name`) AS acct_name,
MAX(`acct_#`) AS acct_#,
address
FROM `test_table`
GROUP BY `ID`_type,
seq_no;
I want to merge rows based on by id_type and seq_no. I am using max to merge the rows but I am overwriting any exisiting addresses and acct_names due to the MAX acct#.
my results
id_type,seq_no.,acct_name,_acct#,address
12345,67,jiimm joe the 3rd,0981_1,167 s.40th st
12345,80,Lee,588_1,109 road st
Losing po box 1234 for 67-
Losing 1234 street ave for 80, losing jiimm-
desired results
12345,80,Lee,588_1,109 road st
12345,80,Lee,588_1,1234 street ave
12345,67,jiimm,0981_1,167 s.40th st
12345,67,jiimm,0981_1,po box 1234
12345,67,jiimm joe the 3rd,0981_1,167 s.40th st

Answer 1

This gives you what you're looking for, but read my comment / question underneath your question above. There is an ambiguous "pick one row from many" situation that needs clarification. In that ambiguous situation, you imply that rules call for delivering the minimum non-blank account name, which this code does, but you can see how it requires treating the account name one way, and treating the acct (#) and address a different way. I think you're headed for an application that delivers results based on hard to remember rules. Funky rules like that end up getting reported as defects, even if you publish said processing rules. Hence, you may want to enhance the process upstream that captures this data to deliver more disciplined data.
SQLFIDDLE link - In short the inner query populates missing values, then the outer result set delivers the distinct rows. I tested this with blank values not null. I did make a quick effort to add the code to handle nulls, but I didn't test it using nulls, so I suggest testing it as such if that is what production will use.
select distinct * from (
select d.id_type, d.seq_no
,coalesce( nullif( acct_name, ''), min_acct_name ) as merged_acct_name
,coalesce( nullif( acct, ''), max_acct ) as merged_acct
,coalesce( nullif( address, ''), max_address ) as merged_address
from test_table d
left join ( select id_type, seq_no
,max( acct ) as max_acct
,max( address ) as max_address
from test_table
group by id_type, seq_no
) as max_
on max_.id_type = d.id_type and max_.seq_no = d.seq_no
and ( coalesce( d.acct,'' ) = ''
or coalesce( d.address,'' ) = '' )
left join ( select id_type, seq_no
,min( acct_name ) as min_acct_name
from test_table
where coalesce( acct_name, '' ) <> ''
group by id_type, seq_no
) as min_
on min_.id_type = d.id_type and min_.seq_no = d.seq_no
and coalesce( d.acct_name,'' ) = ''
) as t
order by id_type, seq_no desc, merged_acct_name, merged_acct, merged_address

Answer 2

Try this:
select distinct t1.id_type, t1.seq_no
,coalesce( t1.acct_name, t2.acct_name ) as merged_acct_name
,coalesce( t1.acct, t2.acct ) as merged_acct
,coalesce( t1.address, t2.address ) as merged_address
from test_table t1
left join test_table t2
on t1.id_type = t2.id_type
and t1.seq_no = t2.seq_no
where concat(coalesce( t1.acct_name, t2.acct_name )
,coalesce( t1.acct, t2.acct )
,coalesce( t1.address, t2.address ) ) is not null
order by t1.id_type, t1.seq_no;
Or:
select distinct t1.id_type, t1.seq_no
,coalesce( t1.acct_name, t2.acct_name ) as merged_acct_name
,coalesce( t1.acct, t3.acct ) as merged_acct
,coalesce( t1.address, t4.address ) as merged_address
from test_table t1
left join test_table t2
on t1.id_type = t2.id_type
and t1.seq_no = t2.seq_no
left join test_table t3
on t1.id_type = t3.id_type
and t1.seq_no = t3.seq_no
left join test_table t4
on t1.id_type = t4.id_type
and t1.seq_no = t4.seq_no
where concat(coalesce( t1.acct_name, t2.acct_name )
,coalesce( t1.acct, t3.acct )
,coalesce( t1.address, t4.address ) ) is not null
order by t1.id_type, t1.seq_no;
SQL Fiddle Demo

Answer 3

SELECT
D1.id_type
, D1.seq_no
, IFNULL(D1.acct_name, (SELECT MIN(acct_name) FROM data D WHERE D.id_type = D1.id_type AND D.seq_no = D1.seq_no)) t
, IFNULL(D1.acct_no, (SELECT MAX(acct_no) FROM data D WHERE D.id_type = D1.id_type AND D.seq_no = D1.seq_no)) s
, D1.address
FROM data D1
WHERE D1.address IS NOT NULL
ORDER BY id_type, seq_no DESC, acct_name
;
returns
| ID_TYPE | SEQ_NO | T | S | ADDRESS |
|---------|--------|-------------------|--------|-----------------|
| 12345 | 80 | Lee | 588_1 | 109 road st |
| 12345 | 80 | Lee | 588_1 | 1234 street ave |
| 12345 | 67 | jiimm | 0981_1 | po box 1234 |
| 12345 | 67 | jiimm | 0981_1 | 167 s.40th st |
| 12345 | 67 | jiimm joe the 3rd | 0981_1 | 167 s.40th st |
Which is consistent with your expected output except for the order of the third and fourth row. However, for larger amounts of data MAX and MIN will become more and more likely to be of limited help.
SQL Fiddle