Find number of rows identical one some, but different on another column - sql

Question

Say I have the following table:
CREATE TABLE data (
PROJECT_ID VARCHAR,
TASK_ID VARCHAR,
REF_ID VARCHAR,
REF_VALUE VARCHAR
);
I want to identify rows where
PROJECT_ID, REF_ID, REF_VALUE are the same
but TASK_ID are different.
The desired output is a list of TASK_ID_1, TASK_ID_2 and COUNT(*) of such conflicts. So, for example,
DATA
+------------+---------+--------+-----------+
| PROJECT_ID | TASK_ID | REF_ID | REF_VALUE |
+------------+---------+--------+-----------+
| 1 | 1 | 1 | 1 |
| 1 | 1 | 1 | 2 |
| 1 | 2 | 1 | 1 |
| 1 | 2 | 1 | 2 |
+------------+---------+--------+-----------+
OUTPUT
+-----------+-----------+----------+
| TASK_ID_1 | TASK_ID_2 | COUNT(*) |
+-----------+-----------+----------+
| 1 | 2 | 2 |
| 2 | 1 | 2 |
+-----------+-----------+----------+
would mean that there are two entries with TASK_ID == 1 and two entries with TASK_ID == 2 that share the same values for the other three columns. The inherent symmetry in the output is fine.
How would I go about finding this information? I've tried joining the table onto itself and grouping, but this turned up more results for a single task than the table had rows altogether, so it's clearly wrong.
The database used is PostgreSQL, though a solution that applies to most common SQL systems would be preferable.

Answer 1

You want a self join and aggregation:
select d1.task_id as task_id_1, d2.task_id as task_id_2, count(*)
from data d1 join
data d2
on d1.project_id = d2.project_id and
d1.ref_id = d2.ref_id and
d1.ref_value = d2.ref_value and
d1.task_id <> d2.task_id
group by d1.task_id, d2.task_id;
Notes:
Add the condition d1.task_id < d2.task_id if you want each pair to occur only once in the result set.
This does not handle NULL values, although that is easy enough to handle. Use is not distinct from instead of =.
You can also simplify this a bit with the using clause:
select d1.task_id as task_id_1, d2.task_id as task_id_2, count(*)
from data d1 join
data d2
using (project_id, ref_id, ref_value)
where d1.task_id <> d2.task_id
group by d1.task_id, d2.task_id;
You can get an idea of how many rows might be returned by using:
select d.project_id, d.ref_id, d.ref_value, count(distinct d.task_id), count(*)
from data d
group by d.project_id, d.ref_id, d.ref_value;

Answer 2

This is how I understand your question. This assume there are only two task for the same combination.
SQL DEMO
SELECT "PROJECT_ID", "REF_ID", "REF_VALUE",
MIN("TASK_ID") as TASK_ID_1,
MAX("TASK_ID") as TASK_ID_2,
COUNT(*) as cnt
FROM Table1
GROUP BY "PROJECT_ID", "REF_ID", "REF_VALUE"
HAVING MIN("TASK_ID") != MAX("TASK_ID")
-- COUNT(*) > 1 also should work
OUTPUT
I add more column to make clear what are the same elements:
| PROJECT_ID | REF_ID | REF_VALUE | task_id_1 | task_id_2 | cnt |
|------------|--------|-----------|-----------|-----------|-----|
| 1 | 1 | 2 | 1 | 2 | 2 |
| 1 | 1 | 1 | 1 | 2 | 2 |