Re: Weird quirk with pg_dump of complex types

On Thu, 2009-02-26 at 15:52 -0800, Josh Berkus wrote:
> Jeff,
>
> > Functions are similar, actually. The argument list needs to specify
> > schema paths as well, if it's not in some expected place (I think it
> > does so for all schemas other than pg_catalog).
>
> Except that they don't appear to do so.

Here is the case I'm talking about:

postgres=# create schema a;
CREATE SCHEMA
postgres=# create type a.int4 as (i pg_catalog.int4);
CREATE TYPE
postgres=# create function f1(x a.int4, y pg_catalog.int4) returns
pg_catalog.int4 language sql as $$ select 1; $$;
CREATE FUNCTION

-- pg_dump output:

SET search_path = public, pg_catalog;

...

CREATE FUNCTION f1(x a.int4, y integer) RETURNS integer
    LANGUAGE sql
    AS $$ select 1; $$;


So, there are some special cases somewhere so that the pg_dump output
isn't littered with unreadable "pg_catalog.int4" everywhere.

In the general case though, for any object that refers to multiple other
objects, I don't see any way around explicit schema qualification. I
suppose it could be smart and say "foo_type is unique in my search path,
so I don't need to schema-qualify it".

Have you considered working from the "custom" format rather than text?
I'm not sure whether it solves your problem, but I think it provides the
most information.

Regards,
    Jeff Davis