Обсуждение: pg_catalog.pg_proc procedure has correct proargnames array, but proargtypes is empty

On Sat, 16 Jul 2022 at 12:14, Walker Philips <wphilips53@gmail.com> wrote:
> The following function's metadata is not correctly reported in the pg_proc table (see picture below).

Seems quite unlikely this such a fundamental thing could be broken.

Can you share which version of PostgreSQL you see this on?

I see the proargtypes populated properly on current master.

David

Walker Philips <wphilips53@gmail.com> writes:
> The following function's metadata is not correctly reported in the pg_proc
> table (see picture below).

Looks fine to me:

regression=# CREATE OR REPLACE FUNCTION get_period_type_adjustment(period_type_id
integer)
 RETURNS numeric
...
regression=# \x
Expanded display is on.
regression=# select * from pg_proc where proname = 'get_period_type_adjustment';
...
pronargs        | 1
pronargdefaults | 0
prorettype      | 1700
proargtypes     | 23
proallargtypes  |
proargmodes     |
proargnames     | {period_type_id}
proargdefaults  |
...

One IN integer argument named period_type_id, result type numeric.

Note the comments in

https://www.postgresql.org/docs/devel/catalog-pg-proc.html

that proallargtypes etc. are generally set to NULL if they wouldn't
carry any additional data.  This is a choice we made a long time
ago to prioritize storage space over simplicity of implementation.
I don't know whether we'd make the same choice in a green field
today ... but we're not going to change it now, because the ensuing
confusion would be pretty bad.

            regards, tom lane

David Rowley <dgrowleyml@gmail.com> writes:
> Can you share which version of PostgreSQL you see this on?

Hmm ... looking more closely, it's definitely not stock PG,
because proargtypes is oidvector which would not include any
curly braces in the output.

Hard to tell what's going on exactly; seems like it could be
corrupt catalog data or faulty code modifications.  Or maybe
there's something wrong with whatever the viewing tool is?

            regards, tom lane

Walker Philips <wphilips53@gmail.com> writes:
> I'm using DBeaver as my IDE. The column does appear to be oidvector. Other
> functions seem to come through ok weirdly enough. It potentially might be a
> visual glitch with DBeaver based on the following screenshot. Still pretty
> odd. I can submit to DBeaver if that might be the underlying issue.

I'd try looking at the row in psql.  If it looks normal there,
then it's a DBeaver problem.

            regards, tom lane