Can I apply a WHERE clause if COUNT condition is met? - oracle

Question

I'm having a hard time trying to add a WHERE clause that filters null values from a column ONLY when there are other rows that return data for that same column.
If all rows have null values for that column, keep them all.
If any row has data for that column, remove the rows with null values and just keep the rows with data.
I'm working on an Oracle database.
In my SELECT statement, I'm currently using a LEFT JOIN to pull data from table B even if the values for column B.info are null.
The actual query goes as follows:
SELECT A.id as A_ID, A.name as A_NAME,
B.id as B_ID, B.name as B_NAME, B.info as B_INFO
FROM A
LEFT JOIN B ON B.id = A.id_B
WHERE A.filename = 'file1.txt'
I have 2 possible scenarios in the bussiness I'm working on:
For a given "filename", the query returns some rows with the B.info column with null values and some others with the B.info column filled with data. I want the query to return only the rows with B.info != null.
Scenario 1 - Actual output:
+-------+--------+------+--------+-----------+
| A_ID | A_NAME | B_ID | B_NAME | B_INFO |
+-------+--------+------+--------+-----------+
| 1 | John | null | null | null |
+-------+--------+------+--------+-----------+
| 2 | John | 3 | Julia | Age is 35 |
+-------+--------+------+--------+-----------+
| 3 | John | null | null | null |
+-------+--------+------+--------+-----------+
Scenario 1 - Desired output:
+-------+--------+------+--------+-----------+
| A_ID | A_NAME | B_ID | B_NAME | B_INFO |
+-------+--------+------+--------+-----------+
| 2 | John | 3 | Julia | Age is 35 |
+-------+--------+------+--------+-----------+
For a given "filename", the query returns all of the rows with the B.info column with null values.
I want the query to keep returning those rows.
Scenario 2 - Actual output = desired output:
+-------+--------+------+--------+--------+
| A_ID | A_NAME | B_ID | B_NAME | B_INFO |
+-------+--------+------+--------+--------+
| 1 | Mark | null | null | null |
+-------+--------+------+--------+--------+
| 2 | Mark | null | null | null |
+-------+--------+------+--------+--------+
| 3 | Mark | null | null | null |
+-------+--------+------+--------+--------+
I tried adding the condition B.info is not null in the where clause but, although it returns the desired output for the scenario 1, the output for the scenario 2 returns no rows:
SELECT A.id as A_ID, A.name as A_NAME,
B.id as B_ID, B.name as B_NAME, B.info as B_INFO
FROM A
LEFT JOIN B ON B.id = A.id_B
WHERE A.filename = 'file1.txt'
AND B.info is not null
Scenario 1 - Output
+-------+--------+------+--------+-----------+
| A_ID | A_NAME | B_ID | B_NAME | B_INFO |
+-------+--------+------+--------+-----------+
| 2 | John | 3 | Julia | Age is 35 |
+-------+--------+------+--------+-----------+
Scenario 2 - Output
+-------+--------+------+--------+--------+
| A_ID | A_NAME | B_ID | B_NAME | B_INFO |
+-------+--------+------+--------+--------+
+-------+--------+------+--------+--------+
I also tried adding a CASE in the WHERE clause but it throws an error (ORA-00934: group function is not allowed here)
SELECT A.id as A_ID, A.name as A_NAME,
B.id as B_ID, B.name as B_NAME, B.info as B_INFO
FROM A
LEFT JOIN B ON B.id = A.id_B
WHERE A.filename = 'file1.txt'
AND B.info = CASE WHEN count(B.info) > 0 THEN null
ELSE B.info
END
I'm sorry I can't use the real example for confidentiality issues. I hope my example is clear enough. I would appreciate any help!

Answer 1

Count all rows and nullable rows. Use analytic count, because you need details. Then show only rows containing data or null rows if both counts are equal:
select id, a_name, b_name, info
from (
select a.id, b.id b_id, a.name a_name, b.name b_name, b.info,
count(case when b.id is null then 1 end) over (partition by a.filename) c1,
count(1) over (partition by a.filename) c2
from a left join b on a.id = b.id )
where b_id is not null or c1 = c2
demo

Answer 2

You can consider this problem a ranking problem: You want to show the best rows only, with non-null rows being considered "better" than null rows.
Ranking can be achieved with an appropriate ORDER BY clause. As of Oracle 12c:
select
a.id as a_id, a.name as a_name,
b.id as b_id, b.name as b_name, b.info as b_info
from a
left join b on b.id = a.id_b
where a.filename = 'file1.txt'
order by case when b.id is null then 2 else 1 end
fetch first rows with ties;
In older versions:
select a_id, a_name, b_id, b_name, b_info
from
(
select
a.id as a_id, a.name as a_name,
b.id as b_id, b.name as b_name, b.info as b_info,
rank() over (order by case when b.id is null then 2 else 1 end) as rnk
from a
left join b on b.id = a.id_b
where a.filename = 'file1.txt'
)
where rnk = 1;