On 09/11/20 13:00, Daniele Varrazzo wrote:
> On Mon, 9 Nov 2020 at 06:57, Federico Di Gregorio <fog@dndg.it> wrote:
[snip]
>> IMHO, oid is a bad idea
>> because it has a very specific semantic and the error messages generated
>> by PostgreSQL will be more confusing.
>
> I'm not sure I understand this. At the moment, the oids are something
> that don't really surface to the end-users, who are not required to
> use them explicitly and shouldn't be seen in the error messages. For
> instance the query above might results in a call:
>
> >>> from psycopg3.oids import builtins
> >>> builtins["numeric"].oid
> 1700
>
> >>> res = conn.pgconn.exec_params(b"select '[]'::jsonb -> $1",
> [b"1"], [1700])
> >>> res.status
> <ExecStatus.FATAL_ERROR: 7>
>
> >>> print(res.error_message.decode("utf8"))
> ERROR: operator does not exist: jsonb -> numeric
> LINE 1: select '[]'::jsonb -> $1
> ^
> HINT: No operator matches the given name and argument types. You
> might need to add explicit type casts.
>
> So the oid is only used internally, in the mapping python type ->
> exec_params() types array, the 1700 shouldn't surface anywhere.
>
> Maybe I'm misunderstanding your concern: can you tell me better?
My fault. I misread and though you wanted to use OID as *the* type to
pass to PostggreSQL for numbers.
federico