如何为db2 Sql中的每一列汇总没有空值的记录？

Question

I have a table as below-

id	name	col1	col2	col3
1	aaa	1	null	null
1	aaa	null	1	null
1	aaa	null	null	1
1	aaa	null	null	2
1	aaa	null	1	null
2	bbb	null	null	null
2	bbb	1	null	null

i want the output as below-

id	name	col1	col2	col3
1	aaa	1	1	1
1	aaa	null	1	2
2	bbb	1	null	null

i tried doing this:

select id ,name, max(col1),max(col2),max(col3) 
from table group by id, name;

but the output is this, which is not as expected

id	name	col1	col2	col3
1	aaa	1	1	2

can u pls help me achieve the desired output. many thanks in advance.

Answer 1

You may try this as is.

Firstly, the rows in each group specified by (id, name) columns are enumerated. Note, that if you don't use the order by clause in the row_number function (as in the example), such an enumeration may be done in some non-repeatable and non-predictable way. So, if you want a repeatable result, you should have a way to order the rows explicitly. I've added an additional row, and you may use it uncommenting the order by clause to get the exact result as you specified in the question.

Secondly, the rows in each group are processed with a Recursive Common Table Expression statement. A new surrogate column grp (additional group number inside each group) is introduced. The first row of each group (having rn_=1) gets all the original values and grp=1. The grp value of each next row is increased by 1, if the the value of c1, c2 or c3 in this next row is not null and a value in a previous row of the corresponding column is non null as well.
If the group number is not increased, a non-null value from a previous row is copied instead the current value for each cx.
If it is increased, the current row values are retained.

Finally, we use this surrogate group number in the group by clause.
dbfiddle link .

with mytab (id, name, c1, c2, c3, ord) as
(
values
  (1, 'aaa', 1, null::int, null::int, 10)
, (1, 'aaa', null, 1, null, 20)
, (1, 'aaa', null, null, 1, 30)
, (1, 'aaa', null, null, 2, 40)
, (1, 'aaa', null, 1, null, 50)
, (2, 'bbb', null, null, null, 1)
, (2, 'bbb', 1, null, null, 2)
)
-- Rows enumeration inside all groups
, mytab2 as 
(
select row_number () over (partition by id, name
--order by ord
) rn_, t.*
from mytab t
)
--
, a (id, name, rn_, grp, c1, c2, c3) as
(
select id, name, rn_, 1 as grp, c1, c2, c3 
from mytab2
where rn_ = 1
  union all
select t.id, t.name, t.rn_
, a.grp 
+ case 
    when a.c1 is not null and t.c1 is not null
      or a.c2 is not null and t.c2 is not null
      or a.c3 is not null and t.c3 is not null
    then 1
    else 0
  end as grp
, case 
    when a.c1 is not null and t.c1 is not null
      or a.c2 is not null and t.c2 is not null
      or a.c3 is not null and t.c3 is not null
    then t.c1
    else coalesce (a.c1, t.c1)
  end as c1
, case 
    when a.c1 is not null and t.c1 is not null
      or a.c2 is not null and t.c2 is not null
      or a.c3 is not null and t.c3 is not null
    then t.c2
    else coalesce (a.c2, t.c2)
  end as c2
, case 
    when a.c1 is not null and t.c1 is not null
      or a.c2 is not null and t.c2 is not null
      or a.c3 is not null and t.c3 is not null
    then t.c3
    else coalesce (a.c3, t.c3)
  end as c3
from a, mytab2 t
where t.id = a.id and t.name = a.name and t.rn_ = a.rn_ + 1
)
select id, name
, max (c1) as c1
, max (c2) as c2
, max (c3) as c3
from a
group by id, name, grp