Thread: BUG #17827: Rule on insert into table doesn't work when excepting select from the table itself

The following bug has been logged on the website:

Bug reference:      17827
Logged by:          Hanna  Yanchurevich
Email address:      hyanchurevich@spotware.com
PostgreSQL version: 13.2
Operating system:   Centos 7
Description:

To reproduce this bug you may use following script:

create table tbl (id int);
create table rule_stat (msg text, id int);
create rule rule_tbl as on insert to tbl do insert into rule_stat
values('Rule triggered for ',new.id);

insert into tbl 
select 1
except
select id from tbl;

table rule_stat; -- no rows

And using WHERE true clause for target table rule not working too:

insert into tbl 
select 2
except
select id from tbl where true; 

table rule_stat; -- no rows

But using WHERE false clause for target table rule starts working:

insert into tbl 
select 1
except
select id from tbl where false; 

table rule_stat; -- new row appears

PG Bug reporting form <noreply@postgresql.org> writes:
> create table tbl (id int);
> create table rule_stat (msg text, id int);
> create rule rule_tbl as on insert to tbl do insert into rule_stat
> values('Rule triggered for ',new.id);

> insert into tbl 
> select 1
> except
> select id from tbl;

> table rule_stat; -- no rows

This is not a bug.  The DO ALSO command executes after the original
INSERT command, and what it executes looks basically like

insert into rule_stat
select 'Rule triggered for ', id from
(select 1
 except
 select id from tbl);

But at this point we've already completed the original INSERT,
so now there is a row with id 1 in "tbl", and thus the EXCEPT
produces nothing.

While there are use-cases for this sort of behavior, most people
find that propagating data to another table is better done with
an AFTER trigger.  It's far easier to understand what will happen.

            regards, tom lane

Hanna Yanchurevich <hyanchurevich@spotware.com> writes:
> According to this information I can cause another kind of error:
> create table tbl (id serial primary key, msg text);
> create table rule_stat (msg text, id int references tbl(id));
> create rule rule_tbl as on insert to tbl do insert into rule_stat values('Last
> inserted id was ',new.id);
> insert into tbl (msg)
> select 'I`m an insert';
> SQL Error [23503]: ERROR: insert or update on table "rule_stat" violates
> foreign key constraint "rule_stat_id_fkey"
>   Detail: Key (id)=(2) is not present in table "tbl".

Yup.  It's pretty obvious what's going on if you look at EXPLAIN:

explain verbose insert into tbl (msg)
select 'I`m an insert';
                               QUERY PLAN
------------------------------------------------------------------------
 Insert on public.tbl  (cost=0.00..0.01 rows=0 width=0)
   ->  Result  (cost=0.00..0.01 rows=1 width=36)
         Output: nextval('tbl_id_seq'::regclass), 'I`m an insert'::text

 Insert on public.rule_stat  (cost=0.00..0.01 rows=0 width=0)
   ->  Result  (cost=0.00..0.01 rows=1 width=36)
         Output: 'Last
 inserted id was '::text, nextval('tbl_id_seq'::regclass)
(8 rows)

> Such behaviour is a bit confusing. Because by using new.* I expect to get a
> recently inserted row, but not the result of some query running the second
> time (which causes implicit incrementing of id serial).

If that's the mental model you want to work with, use a trigger.
When you work with rules, you are working with macros, and they
have the same sort of multiple-evaluation hazards as macros in,
say, C.

            regards, tom lane