Thread: BUG #17515: Change in behaviour of a composite type within a domain

The following bug has been logged on the website:

Bug reference:      17515
Logged by:          Dan Kubb
Email address:      dan.kubb@gmail.com
PostgreSQL version: 14.3
Operating system:   All
Description:

I noticed a regression in 13.7 and 14.3 in the behaviour of composite types
within a domain. Here is a simplified reproduction:

BEGIN;

-- Create composite type
CREATE TYPE my_composite_type AS (
  col1 boolean,
  col2 boolean
);

-- Create domain wrapping the composite type
CREATE DOMAIN my_domain AS my_composite_type NOT NULL;

-- Create function to test composite type
CREATE FUNCTION test_my_composite_type()
RETURNS my_composite_type
LANGUAGE sql
AS $$
  SELECT (true, true)
$$;

-- Create function to test domain
CREATE FUNCTION test_my_domain()
RETURNS my_domain
LANGUAGE sql
AS $$
  SELECT (true, true)
$$;

-- Works on 13.6/14.2 and 13.7/14.3
SELECT test_my_composite_type FROM test_my_composite_type();

-- Works on 13.6/14.2, but pg 13.7/14.3 this returns the error:
-- ERROR:  type my_domain is not composite
SELECT test_my_domain FROM test_my_domain();

ROLLBACK;

-- 

Dan Kubb

PG Bug reporting form <noreply@postgresql.org> writes:
> I noticed a regression in 13.7 and 14.3 in the behaviour of composite types
> within a domain. Here is a simplified reproduction:

Argh, looks like I broke this in ec62cb0aa.  Will fix, thanks for
the report!

            regards, tom lane