I have a question on how things are handled in SQL vs PL/SQL for the same code.

If I define a few simple types like this:

CREATE OR REPLACE TYPE PARENT FORCE IS OBJECT
(
    parent_name VARCHAR2(128)
) NOT FINAL;
/

CREATE OR REPLACE TYPE child FORCE UNDER PARENT
(
    child_name VARCHAR2(128)
);
/

And then define a simple function to get the type name (the idea being that I can get the type name of an object that I pass in which could be parent or any number of inherited types):

CREATE OR REPLACE FUNCTION get_type_name(p_obj PARENT) RETURN VARCHAR2 IS
BEGIN
    RETURN sys.anydata.convertobject(obj => p_obj).gettypename;
END get_type_name;
/

If I run it like this:

DECLARE
    v_child child;
BEGIN
    dbms_output.put_line(get_type_name(p_obj => v_child));
END;

I get the type name of '<my_schema>.CHILD'

Which is what I want.

If I run it like this:

SELECT get_type_name(p_obj => child(parent_name => 'parent'
                                   ,child_name  => 'chld')) type_name
FROM   dual;

Or with SQL in a pl/sql context

SELECT get_type_name(p_obj => child(parent_name => 'parent'
                                   ,child_name  => 'chld')) type_name
into v_type_name
FROM   dual;

I get '<my_shchema>.PARENT'.

My question is: Why do the pl/sql and sql engines handle it differently? And can I rely on the pl/sql implementation to do this in future releases of Oracle (i.e. this is expected behavior and not a "bug" that could be fixed later).

Thanks in advance! I've learned so much from this site.

SQL> select sys.anydata.convertobject(parent('x')) from dual;
ERROR:
ORA-22636: collection unpickling failed due to corrupt image