Thread: rules help
I am trying to create a view with rules to make data entry, etc. more
convenient. The view must be a union of two tables and the rules must
manipulate the underlying tables. Everything is fine except for one
thing I need help on.
One table must maintain a unique list of keywords that can be
referenced by >1 row in the second table. Inserts into the view need
to insert into the keyword list if necessary, but not if not
necessary. If I simply create the normal insert rule, some inserts to
the view fail because the keyword is not unique. That would be fine
if either the insert into the keyword table could be conditional on
the lack of the keyword or the entire set of rules could proceed even
if the insert into the keyword table fails.
Apparently, putting a where condition on a rule doesn't work (see
error message below).
Any ideas on how to accomplish this?
Thanks for your help.
Cheers,
Brook
===========================================================================
-- tables
drop sequence rule_table_1_id_seq;
drop table rule_table_1;
create table rule_table_1
(id serial,name text,
unique (name)
);
drop sequence rule_table_2_id_seq;
drop table rule_table_2;
create table rule_table_2
(id serial,table_1_id int4 references rule_table_1 (id),address text
);
-- view
drop view rule_view;
create view rule_view as select a.name, b.address from rule_table_1 a, rule_table_2 b
where a.id = b.table_1_id;
-- rules
create rule rule_view_insert_1a as on insert to rule_viewwhere not exists (select id from rule_table_1 where name =
new.name)do insteadinsert into rule_table_1 (name) values (new.name);
create rule rule_view_insert_2 as on insert to rule_view do insteadinsert into rule_table_2 (table_1_id, address)
selectid, new.address from rule_table_1 where name = new.name;
-- insert
insert into rule_view (name, address) values ('Tom', 'New York');
ERROR: ExecEvalExpr: unknown expression type 108
insert into rule_view (name, address) values ('Sue', 'Boston');
ERROR: ExecEvalExpr: unknown expression type 108
insert into rule_view (name, address) values ('Bill', 'Chicago');
ERROR: ExecEvalExpr: unknown expression type 108
insert into rule_view (name, address) values ('Tom', 'Boston');
ERROR: ExecEvalExpr: unknown expression type 108
select * from rule_view;
select * from rule_table_1;
select * from rule_table_2;
Brook Milligan wrote:
>
> I am trying to create a view with rules to make data entry, etc. more
> convenient. The view must be a union of two tables and the rules must
> manipulate the underlying tables. Everything is fine except for one
> thing I need help on.
>
> [...]
>
> Apparently, putting a where condition on a rule doesn't work (see
> error message below).
Looks strange - must dive into soon. But anyway, in your case
the order in which the rule actions take place is important.
So the entire approach to setup two separate rules is wrong,
because it is not guaranteed in which order the rules will
get applied later! If the rule inserting into table 2 get's
applied first, the entry in table 1 might not be there and
thus, the insert will become a noop since the join
t1.id+new.address would be empty.
This is a case where multi action rules are required. And in
that case, the qualification must be part of the actions, not
the entire rule because you want the second action to execute
allways.
I got it working, but another very strange thing surfaced and
absolutely have no clue where that was coming from.
>
> ===========================================================================
> -- tables
> create table rule_table_1
> (
> id serial,
> name text,
> unique (name)
> );
>
> create table rule_table_2
> (
> id serial,
> table_1_id int4 references rule_table_1 (id),
> address text
> );
>
> -- view
>
> drop view rule_view;
> create view rule_view as select a.name, b.address from rule_table_1 a, rule_table_2 b
> where a.id = b.table_1_id;
>
> -- rules
create rule rule_view_ins as on insert to rule_view
do instead (
insert into rule_table_1 (name)
select new.name
where not exists
(select a.id from rule_table_1 a where a.name = new.name);
insert into rule_table_2 (table_1_id, address)
select b.id, new.address from rule_table_1 b
where new.name = b.name group by 1;
);
The strangeness here is, that (for me) it should work without
the GROUP BY clause. But then I got multiple entries into
table 2. Exactly the number of rows in table 1, but they all
referenced the correct entry. So with your test data I got 2x
Sue in Boston, 3x Bill in Chicago and 3x Tom in Boston.
The next problem is, that setting up another table with 2
text fields and doing a
insert into rule_view select * from addr_data;
doesn't work at all. Seems the rules aren't triggered and the
data is stored in the view relation.
And the final problem is that after defining the above rule a
select from pg_rules crashes the backend. Oh man - think I've
held my hands for too long off of the rule system :-(
Jan
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#======================================== jwieck@debis.com (Jan Wieck) #
Thanks for the QUICK response, Jan. I guess I didn't realize one
could put multiple actions in a rule.
The strangeness here is, that (for me) it should work without the GROUP BY clause. But then I got
multipleentries into table 2. Exactly the number of rows in table 1, but they all referenced the correct
entry.So with your test data I got 2x Sue in Boston, 3x Bill in Chicago and 3x Tom in Boston.
I don't see this problem (NetBSD 1.3.2/i386, postgresql 6.4.2). See
my script below which does what I really want (with one exception).
The next problem is, that setting up another table with 2 text fields and doing a
insert into rule_view select * from addr_data;
doesn't work at all. Seems the rules aren't triggered and the data is stored in the view relation.
I'll see if I see that, too.
And the final problem is that after defining the above rule a select from pg_rules crashes the backend. Oh
man- think I've held my hands for too long off of the rule system :-(
I see the crash, too.
Now, for the one exception. The delete rule below has one action
commented out, the removal of the original name. If this is
uncommented, it is still not removed. Have I misconstrued that action
or is this something else odd?
Thanks again for your help.
Cheers,
Brook
===========================================================================
drop sequence rule_table_1_id_seq;
drop table rule_table_1;
create table rule_table_1
(id serial,name text,
unique (name)
);
drop sequence rule_table_2_id_seq;
drop table rule_table_2;
create table rule_table_2
(id serial,table_1_id int4 references rule_table_1 (id),address text
);
drop view rule_view;
create view rule_view as select a.name, b.address from rule_table_1 a, rule_table_2 b
where a.id = b.table_1_id;
create rule rule_view_insert as on insert to rule_view
do instead ( insert into rule_table_1 (name) select new.name where not exists
(selecta.id from rule_table_1 a where a.name = new.name); insert into rule_table_2 (table_1_id, address)
select b.id, new.address from rule_table_1 b where new.name = b.name group by 1;
);
create rule rule_view_delete as on delete to rule_view
do instead (delete from rule_table_2 where table_1_id = (select id from rule_table_1 where name = current.name);
-- delete from rule_table_1 where name = current.name;
);
create rule rule_view_update as on update to rule_view
do instead (update rule_table_2 set address = new.address where table_1_id = (select id from rule_table_1 where name =
current.name);updaterule_table_1 set name = new.name where name = current.name;
);
insert into rule_view (name, address) values ('Tom', 'New York');
insert into rule_view (name, address) values ('Sue', 'Boston');
insert into rule_view (name, address) values ('Bill', 'Chicago');
insert into rule_view (name, address) values ('Tom', 'Boston');
select * from rule_view order by name, address;
select * from rule_table_1 order by name;
select * from rule_table_2 order by table_1_id;
update rule_view set address = 'Boston' where name = 'Bill';
select * from rule_view order by name, address;
select * from rule_table_1 order by name;
select * from rule_table_2 order by table_1_id;
delete from rule_view where name = 'Tom';
select * from rule_view order by name, address;
select * from rule_table_1 order by name;
select * from rule_table_2 order by table_1_id;