Postgres 13.5 out parameter argument with explicit cast fails with argument is not writable

On Mon, Feb 14, 2022 at 3:50 PM Delaney, Ed <Ed.Delaney@ellucian.com> wrote:

I do not have pg 13.6 installed yet as we just upgraded all of our systems
from 11 to 13.5 but I suspect it will be the same.

declare

lv_somestring character varying (4000);

begin

call foo(lv_somestring::character varying); -- note cast

While no explicit cast is actually required, we have a very large amount
of generated code that uses explicit type casting this way.

I'm unsure about the regression but in fact the explicit cast is simply
incorrect conceptually. You are supposed to be supplying a variable to the
call. What you are supplying is a constant. That constant has the same
value as the variable but is not itself a variable and even has a different
type. If you write: SELECT lv_somestring::varchar(1) you will get back the
first character of the max 4,000 character value presently in the variable
lv_somestring. But you will not be changing the value stored in
lv_somestring. Extend that concept to CALL and you can see why the error
you are being shown in v13 is correct.

David J.

"Delaney, Ed" <Ed.Delaney@ellucian.com> writes:

create or replace procedure bar ()
language plpgsql
as $procedure$
declare
lv_somestring character varying (4000);
begin
call foo(lv_somestring::character varying); -- note cast
raise notice 'lv_somestring: %', lv_somestring;
end;
$procedure$;

I think you're out of luck on that. Releases before last November
ignored that cast entirely, thinking it a no-op. Current releases
do not ignore it, recognizing that in fact it has to be understood
as casting to varchar-of-unspecified-length. But then the argument
isn't a bare variable anymore.

We debated about whether to back-patch such a behavioral change,
and maybe we shouldn't have. But there are scenarios in which
the old behavior makes it just impossible to do things, eg you
may not be able to get a UNION to produce the desired type.
On balance we felt this was a bug fix.

regards, tom lane

Thanks for the timely replies.
I've found that either removing the length specification or the cast addresses the issue.
So this all makes sense to me.

I could not find a reference to this change in the release notes, but since it is a breaking change, perhaps it should be noted?
regards
Ed

From: Tom Lane <tgl@sss.pgh.pa.us>
Date: Monday, February 14, 2022 at 6:28 PM
To: Delaney, Ed <Ed.Delaney@ellucian.com>
Cc: pgsql-bugs@lists.postgresql.org <pgsql-bugs@lists.postgresql.org>
Subject: [EXT]: Re: Postgres 13.5 out parameter argument with explicit cast fails with argument is not writable
"Delaney, Ed" <Ed.Delaney@ellucian.com> writes:

create or replace procedure bar ()
language plpgsql
as $procedure$
declare
lv_somestring character varying (4000);
begin
call foo(lv_somestring::character varying); -- note cast
raise notice 'lv_somestring: %', lv_somestring;
end;
$procedure$;

I think you're out of luck on that. Releases before last November
ignored that cast entirely, thinking it a no-op. Current releases
do not ignore it, recognizing that in fact it has to be understood
as casting to varchar-of-unspecified-length. But then the argument
isn't a bare variable anymore.

We debated about whether to back-patch such a behavioral change,
and maybe we shouldn't have. But there are scenarios in which
the old behavior makes it just impossible to do things, eg you
may not be able to get a UNION to produce the desired type.
On balance we felt this was a bug fix.

regards, tom lane

"Delaney, Ed" <Ed.Delaney@ellucian.com> writes:

I could not find a reference to this change in the release notes, but since it is a breaking change, perhaps it should be noted?

The 2021-11-11 release notes all say

* Don't discard a cast to the same type with unspecified type modifier (Tom Lane)

For example, if column f1 is of type numeric(18,3), the parser used to
simply discard a cast like f1::numeric, on the grounds that it would
have no run-time effect. That's true, but the exposed type of the
expression should still be considered to be plain numeric, not
numeric(18,3). This is important for correctly resolving the type of
larger constructs, such as recursive UNIONs.

regards, tom lane