SQL group by under some conditions - sql

Question

I have a big table with tons of duplicated rows (among those columns that I care about). Let me start with the following example:
|field1 | field2| field3| field4| field5|
| aa | 1 | NULL | 1 | 0 |
| aaa | 1 | NULL | 1 | 1 |
| aaa | 1 | NULL | 1 | 2 |
| a | 2 | 0 | 1 | 3 |
| a | 2 | 0 | NULL | 4 |
| a | 2 | NULL | 2 | 5 |
| b | 3 | NULL | 2 | 6 |
| b2 | 3 | NULL | NULL | 7 |
| c | 4 | NULL | NULL | 8 |
I am interested in an effiecient query to get the following table:
|field1 | field2| field3| field4|
| aaa | 1 | NULL | 1 |
| a | 2 | 0 | 1 |
| b | 3 | NULL | 2 |
| c | 4 | NULL | NULL |
Basically, it follows the following rules:
for each value of field2, there should be one and exactly one row present
among all the rows with the same value of field2 select the row that satisfy the following in order:
select the row that field4 is not Null (if possible)
among those that have a non Null value for the field4 select the row that has has a non Null value for field 3
among those that have a non Null value for the field4 and 3, select the row that has the longest string value for field 1
among those that satisfy all above, select only one row (does not matter what is the value of field5).
I could do it with bunch of joins, but it becomes very slow. Any better suggestions?
EDIT
The field2 values may not be in an specific order. I just put 1,2,3,4 in the example but this is not generally true in my case. I did not change it directly on the table since one of the suggested solutions are actually considering sequential value for field2, so I kept if for future readers that maybe interested in that.

Answer 1

This type of prioritization is challenging. I think the simplest method in MySQL uses variables:
select t.*
from (select t.*,
(#rn := if(#f2 = field2, #rn + 1,
if(#f2 := field2, 1, 1)
)
) as seqnum
from t cross join
(select #rn := 0, #field2 := '') params
order by field2,
(field4 is not null) desc,
(field3 is not null) desc,
length(field1) desc
) t
where seqnum = 1;
I'm not 100% sure I have the conditions right (the third seems to conflict with the first two). But whatever the prioritization, the idea is the same: use order by to get the rows in the right order and use variables to get the first one.
EDIT:
In SQL Server -- or any other reasonable database -- you do this with row_number():
select t.*
from (select t.*,
row_number() over (partition by field2
order by (case when field4 is not null then 0 else 1 end),
(case when field3 is not null then 0 else 1 end),
len(field1)
) as seqnum
from t
) t
where seqnum = 1;