Re: pg_depend query

"Sim Zacks" <sim(at)nospam(dot)com> writes:
> I am trying to write a query on the pg_depend table and it is confusing the
> hell out of me. Hopefully someone could help.
> I would like to know which views depend on another view.

The dependencies are actually from the view's _RETURN rule to the
other view. So for example:

regression=# create view v1 as select * from int4_tbl;
CREATE VIEW
regression=# create view v2 as select * from v1;
CREATE VIEW

The only dependency v2 has explicitly is one on its namespace:

regression=# select classid::regclass,objid,objsubid,
regression-# refclassid::regclass,refobjid,refobjsubid,deptype from pg_depend
regression-# where classid = 'pg_class'::regclass and objid = 'v2'::regclass;
classid | objid | objsubid | refclassid | refobjid | refobjsubid | deptype

----------+--------+----------+--------------+----------+-------------+---------

pg_class | 855926 | 0 | pg_namespace | 2200 | 0 | n
(1 row)

regression=# select nspname from pg_namespace where oid = 2200;
nspname
---------
public
(1 row)

But if we look for what *depends on* v2, we find:

egression=# select classid::regclass,objid,objsubid,
regression-# refclassid::regclass,refobjid,refobjsubid,deptype from pg_depend
regression-# where refclassid = 'pg_class'::regclass and refobjid = 'v2'::regclass;
classid | objid | objsubid | refclassid | refobjid | refobjsubid | deptype
------------+--------+----------+------------+----------+-------------+---------
pg_rewrite | 855928 | 0 | pg_class | 855926 | 0 | n
pg_rewrite | 855928 | 0 | pg_class | 855926 | 0 | i
pg_type | 855927 | 0 | pg_class | 855926 | 0 | i
(3 rows)

regression=# select rulename from pg_rewrite where oid = 855928;
rulename
----------
_RETURN
(1 row)

The second of these is an implicit dependency from the view's _RETURN rule
to the view. If we now look for everything the _RETURN rule depends on:

regression=# select classid::regclass,objid,objsubid,
regression-# refclassid::regclass,refobjid,refobjsubid,deptype from pg_depend
regression-# where classid = 'pg_rewrite'::regclass and objid = 855928;
classid | objid | objsubid | refclassid | refobjid | refobjsubid | deptype
------------+--------+----------+------------+----------+-------------+---------
pg_rewrite | 855928 | 0 | pg_class | 855926 | 0 | n
pg_rewrite | 855928 | 0 | pg_class | 855923 | 1 | n
pg_rewrite | 855928 | 0 | pg_class | 855926 | 0 | i
(3 rows)

or more legibly

regression=# select classid::regclass,objid,objsubid,
regression-# refclassid::regclass,refobjid::regclass,refobjsubid,deptype from pg_depend
regression-# where classid = 'pg_rewrite'::regclass and objid = 855928;
classid | objid | objsubid | refclassid | refobjid | refobjsubid | deptype
------------+--------+----------+------------+----------+-------------+---------
pg_rewrite | 855928 | 0 | pg_class | v2 | 0 | n
pg_rewrite | 855928 | 0 | pg_class | v1 | 1 | n
pg_rewrite | 855928 | 0 | pg_class | v2 | 0 | i
(3 rows)

The second of these rows shows that the _RETURN rule depends on v1's
first column, and it's that dependency that you are looking for.

You can probably collapse this into one query with some sub-selects,
but I'll leave that exercise to the reader.

The important point here is that an implicit dependency is sort of a
two-way street, because it indicates that the dependent object (here the
_RETURN rule) is an integral part of the referenced object (here the v2
view), and therefore a DELETE CASCADE will cascade in *both* directions.
Dropping v2 should obviously drop its _RETURN rule (that's the normal
forward direction of cascaded drops). But dropping something that the
_RETURN rule depends on forces dropping both the _RETURN rule and its
owning view (reverse direction). You can see this happening if you look
at DROP's notices:

regression=# drop view v1;
NOTICE: rule _RETURN on view v2 depends on view v1
NOTICE: view v2 depends on rule _RETURN on view v2
ERROR: cannot drop view v1 because other objects depend on it
HINT: Use DROP ... CASCADE to drop the dependent objects too.
regression=#

The cascade went through the _RETURN rule, not directly to v2.

regards, tom lane