Обсуждение: BUG #15246: Does not allow an INOUT parameter to receive values whenits data type is a user-defined data type.

The following bug has been logged on the website:

Bug reference:      15246
Logged by:          Anderson Antunes
Email address:      anderson.ant.oli@gmail.com
PostgreSQL version: 10.0
Operating system:   Windows
Description:

I'm trying to assign a popular values to a custom data type that comes from
an INOUT parameter in the function. I already searched the internet and I
did not find a solution. This type of assignment is common in Oracle. I do
not understand why it was not implemented. I get the following message:
"ERROR:" Parameter XXX "is not variable unknown".

Thank you if you can answer me.

create table my_table1
(
    co_1 integer,
    co_2 character varying(20)
);

create table my_table2
(
    co_1 char,
    co_2 double precision
);

create type my_type as
(
    rc_tb1 my_table1,
    rc_tb2 my_table2
);

create or replace function fc_test
(
    inout p_my_type my_type
)
as
$$
begin
    p_my_type.rc_tbl1.co_1 := 1; -- ERRO:  "p_my_type.rc_tbl1.co_1" is not
variable unknown !!!!
    p_my_type.rc_tbl1.co_2 := 'Teeeeeeeeeest';
    p_my_type.rc_tbl2.co_3 := 'T';
    p_my_type.rc_tbl2.co_4 := 10.56;
end;
$$
language plpgsql;

"David G. Johnston" <david.g.johnston@gmail.com> writes:
> On Mon, Jun 18, 2018 at 3:32 PM, PG Bug reporting form <
> noreply@postgresql.org> wrote:
>> p_my_type.rc_tbl1.co_1 := 1; -- ERRO:  "p_my_type.rc_tbl1.co_1" is
>> not
>> variable unknown !!!!

> Short answer is that you cannot simply assign components of a composite
> type one-at-a-time, you have to build up the full final composite result in
> one expression and assign the result of the expression to the typed
> variable (p_my_type in this instance).​

It's not quite that bad.  IIRC, plpgsql handles only one level of field
assignment, so you could write

    p_my_type.rc_tbl1 := ROW(1, 'Teeeeeeeeeeest');

but not

    p_my_type.rc_tbl1.co_1 := 1;

Improving that --- and also allowing mixed array-element-and-field
assignment, say "p_my_type.rc_tbl1[2].co_1 := 1;" --- has been on the
radar screen for a long time, but nobody has gotten round to it.

I think I might've made it a bit easier as of v11, because it'd no longer
be necessary to implement field assignment in two separate code paths
for "rows" and "records".  But it's still a fair amount of work.

            regards, tom lane