Обсуждение: unexpected RULE behavior

Hi,

I did the following:

CREATE TABLE data (id SERIAL, title VARCHAR);
CREATE TABLE data_copy(id INT4, title VARCHAR);
CREATE RULE make_copy AS ON INSERT TO data DO INSERT INTO data_copy
(id,title) VALUES (NEW.id, NEW.title);
INSERT INTO data (title) VALUES ('test');

database=# SELECT * FROM data;
 id | title
----+-------
  1 | test
(1 Zeile)

database=# SELECT * FROM data_copy;
 id | title
----+-------
  2 | test
(1 Zeile)

and wondered about the result in the table 'data_copy'. I expect the row
with the same values but 'id' is different although only 'data.id' is a
serial but not 'data_copy.id'.
My database is PostgreSQL 8.2.4 and runs on Windows Server 2003.

On Mon, 2008-07-07 at 12:16 +0200, Mathias Palm wrote:
> CREATE TABLE data (id SERIAL, title VARCHAR);
> CREATE TABLE data_copy(id INT4, title VARCHAR);
> CREATE RULE make_copy AS ON INSERT TO data DO INSERT INTO data_copy
> (id,title) VALUES (NEW.id, NEW.title);
> INSERT INTO data (title) VALUES ('test');
>
> database=# SELECT * FROM data;
>  id | title
> ----+-------
>   1 | test
> (1 Zeile)
>
> database=# SELECT * FROM data_copy;
>  id | title
> ----+-------
>   2 | test
> (1 Zeile)
>

This is a feature, not a bug. To get the kind of behavior you want, you
need to either use a trigger, or have the rule call a function that
inserts the same values into both tables.

SERIAL implicitly calls nextval() on a sequence to get the default value
to insert. The rule rewrites the query into two queries, and each query
calls nextval(), and each call of nextval() returns a different value.

One way to think about it is that a rule doesn't save any values away in
variables. Functions do save the arguments as variables on the stack,
and those variables can be used multiple times, so that's why calling a
function works.

Regards,
    Jeff Davis