BUG #16520: Deleting from non-existent column in CTE removes all rows
The following bug has been logged on the website:
Bug reference: 16520
Logged by: Jesse Lieberg
Email address: jesse007@ymail.com
PostgreSQL version: 12.3
Operating system: debian:buster-slim
Description:
Using the `postgres:12` docker image and given the following:
```
DROP TABLE IF EXISTS test;
CREATE TABLE test (
id SERIAL,
username varchar(32)
);
INSERT INTO test (username)
VALUES ('Jesse'), ('Jesse'), ('Scott'), ('Scott'), ('John');
```
This will throw an error that the column does not exist:
```
WITH to_delete AS (
SELECT MIN(id), username
FROM test
GROUP BY username
HAVING count(*) > 1
)
SELECT id
FROM to_delete;
```
However, the this will not return an error and instead deletes all rows:
```
WITH to_delete AS (
SELECT MIN(id), username
FROM test
GROUP BY username
HAVING count(*) > 1
)
DELETE FROM test
WHERE id IN (
SELECT id
FROM to_delete
);
```
More information: https://stackoverflow.com/q/62661721/3903479
PG Bug reporting form <noreply@postgresql.org> writes:
This will throw an error that the column does not exist:
```
WITH to_delete AS (
SELECT MIN(id), username
FROM test
GROUP BY username
HAVING count(*) > 1
)
SELECT id
FROM to_delete;
```
Sure, because the columns exposed by to_delete are named "min" and
"username", not "id".
However, the this will not return an error and instead deletes all rows:
```
WITH to_delete AS (
SELECT MIN(id), username
FROM test
GROUP BY username
HAVING count(*) > 1
)
DELETE FROM test
WHERE id IN (
SELECT id
FROM to_delete
);
```
You've been bit by the standard SQL newbie trap that sub-selects
allow outer references. That IN clause devolves to constant true
(at least for non-null id values, and with to_delete known not
empty), because it's just "id = id".
regards, tom lane
On Tuesday, June 30, 2020, PG Bug reporting form <noreply@postgresql.org>
wrote:
The following bug has been logged on the website:
Bug reference: 16520
Logged by: Jesse Lieberg
Email address: jesse007@ymail.com
PostgreSQL version: 12.3
Operating system: debian:buster-slim
Description:However, the this will not return an error and instead deletes all rows:
```
WITH to_delete AS (
SELECT MIN(id), username
FROM test
GROUP BY username
HAVING count(*) > 1
)
DELETE FROM test
WHERE id IN (
SELECT id
FROM to_delete
);
Yes, because the column id does exist - you just omitted the table
reference which ends up making the subquery query equivalent to: “select
test.id from to_delete” which is mandatory, and generally useful, sql
syntax.
David J.