Re: BUG #3696: FK integrity check bypassed using rules.

On 10/25/07, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Yes, a poorly designed rule can invalidate all kinds of expectations
> about behavior. This isn't a bug in my humble opinion.

Yes, this was my first impression.

I was just surprised because of this: the script

CREATE TABLE data (
id INTEGER PRIMARY KEY
);

CREATE TABLE ref (
id INTEGER PRIMARY KEY,
ref_id INTEGER NULL REFERENCES data(id) ON DELETE CASCADE
);

CREATE RULE ref_delete_rule
AS ON DELETE TO ref
DO INSTEAD NOTHING;

INSERT INTO data VALUES (1);
INSERT INTO ref (id, ref_id) VALUES(2, 1);

DELETE FROM data;

gives for the DELETE statement:

ERROR: referential integrity query on "data" from constraint
"ref_ref_id_fkey" on "ref" gave unexpected result
HINT: This is most likely due to a rule having rewritten the query.

But if I change the rule by adding a "WHERE True" to it:

CREATE RULE ref_delete_rule
AS ON DELETE TO ref WHERE True
DO INSTEAD NOTHING;

The integrity is violated. In this sense, this could reveal a
bug (unless postgresql states clearly that it uses a best effort
algorithm when dealing with rewrite rules that can potentially
rewrites auto-generated statements)

I agree, this is not a big deal.