From: Tom Lane <tgl@sss.pgh.pa.us>
> "tsunakawa.takay@fujitsu.com" <tsunakawa.takay@fujitsu.com> writes:
> > Finally, I think I've understood what you meant. Yes, the current code seems
> to be wrong.
>
> I'm fairly skeptical of this claim, because that code has stood for a
> long time. Can you provide an example (not involving hasModifyingCTE)
> in which it's wrong?
Hmm, I don't think of an example. I wonder if attaching WITH before INSERT SELECT and putting WITH between INSERT and
SELECTproduce the same results. Maybe that's why the regression test succeeds with the patch.
To confirm, the question is that when we have the following rule in place and the client issues the query:
[rule]
CREATE RULE myrule AS
ON {INSERT | UPDATE | DELETE} TO orig_table
DO INSTEAD
INSERT INTO some_table SELECT ...;
[original query]
WITH t AS (
SELECT and/or NOTIFY
)
{INSERT INTO | UPDATE | DELETE FROM} orig_table ...;
which of the following two queries do we expect?
[generated query 1]
WITH t AS (
SELECT and/or NOTIFY
)
INSERT INTO some_table SELECT ...;
[generated query 2]
INSERT INTO some_table
WITH t AS (
SELECT and/or NOTIFY
)
SELECT ...;
Although both may produce the same results, I naturally expected query 1, because WITH was originally attached before
thetop-level query, and (2) the top-level query has been replaced with a rule action, so it's natural that the WITH is
attachedbefore the rule action. Super-abbreviated description is:
x -> y (rule)
WITH t x (original query)
WITH t y (generated query 1)
one-part-of-y WITH t another-part-of-y (generated query 2)
As we said, we agree to fail the query if it's the above generated query 2 and WITH contains a data-modyfing CTE, if we
cannotbe confident to accept the change to the WITH position. Which do you think we want to choose?
Regards
Takayuki Tsunakawa