Обсуждение: BUG #16661: Changing columns after the rule is created leads to an error when the RETURNING is used

The following bug has been logged on the website:

Bug reference:      16661
Logged by:          Konstantin Dyachenko
Email address:      drakonard@gmail.com
PostgreSQL version: 12.4
Operating system:   Debian
Description:

The SQL:

CREATE TABLE rule_bug_table (
        id serial NOT NULL PRIMARY KEY,
        archived boolean  NOT NULL
);

CREATE OR REPLACE RULE rule_bug_table__archive_instead_of_delete__rule
    AS ON DELETE TO rule_bug_table
    DO INSTEAD
    UPDATE rule_bug_table
    SET archived = true
    WHERE rule_bug_table."id" = OLD."id"
    RETURNING OLD.*;
    
ALTER TABLE rule_bug_table ADD COLUMN select_allowed boolean NOT NULL
DEFAULT (false);

DELETE FROM rule_bug_table WHERE id = 1 RETURNING *;

gives an error "could not find replacement targetlist entry for attno 3".
Meaning rule cannot find a column with number 3 (select_allowed column).

In other words if after rule was created any column is changed usage of
RETURNING leads to an error.

The workaround is to recreated rule after the table's column was altered.

PG Bug reporting form <noreply@postgresql.org> writes:
> The SQL:

> CREATE TABLE rule_bug_table (
>         id serial NOT NULL PRIMARY KEY,
>         archived boolean  NOT NULL
> );

> CREATE OR REPLACE RULE rule_bug_table__archive_instead_of_delete__rule
>     AS ON DELETE TO rule_bug_table
>     DO INSTEAD
>     UPDATE rule_bug_table
>     SET archived = true
>     WHERE rule_bug_table."id" = OLD."id"
>     RETURNING OLD.*;
    
> ALTER TABLE rule_bug_table ADD COLUMN select_allowed boolean NOT NULL
> DEFAULT (false);

> DELETE FROM rule_bug_table WHERE id = 1 RETURNING *;

> gives an error "could not find replacement targetlist entry for attno 3".

Hmm.  The error message definitely needs to be more user-friendly,
but I don't think this is actually a bug.  The rule's RETURNING clause
expands to this during parsing:

  RETURNING old.id, old.archived

as you can see with "\d+ rule_bug_table".  Then when you try to do
"RETURNING *" in the DELETE, that's asking for a column not available
from the rewritten query.  So you *should* get an error, just not one
that looks like it's an internal failure.

The root issue here is that "*" is expanded at parse time for stored
rules and views.  Unfortunately, that behavior is specifically demanded
by the SQL spec (at least for the view case), so I doubt we can change it.

            regards, tom lane