Is pg_attribute bloat from temp table creation still a potential issue in modern versions of PostgreSQL?

I am not an expert in PostgreSQL, however I recently ran across some PostgreSQL code during a code review which creates and drops temporary tables in a manner that seems consistent with the typical way we do that in SQL Server. i.e. in SQL Server we'd do:

DROP TABLE IF EXISTS #temp_data;
CREATE TABLE #temp_data
(
    i int NOT NULL
);

The code review showed this code:

BEGIN;
DROP TABLE IF EXISTS temp_data;
COMMIT;

BEGIN;
CREATE TEMP TABLE IF NOT EXISTS temp_data(
    i int NOT NULL
);
COMMIT;

[This answer on Stack Overflow](https://stackoverflow.com/a/55580279/1595565) claims you shouldn't drop and recreate temp tables frequently because of pg_attribute bloat. ChatGPT (gah!) has this to say about pg_attribute bloat: > In PostgreSQL, creating temporary tables frequently can cause bloat in pg_attribute, as each new temporary table adds metadata that persists in the system catalogs even after the table is dropped. To avoid excessive bloat, consider these best practices: > > 1. Use ON COMMIT DELETE ROWS Instead of Dropping Tables: > >

> CREATE TEMP TABLE temp_data (
>    i int NOT NULL
> ) ON COMMIT DELETE ROWS;
>

> > 2. Use pg_temp Schema for Session-Level Temporary Tables > >

> CREATE TEMP TABLE pg_temp.temp_data (
>     id SERIAL PRIMARY KEY,
>     value TEXT
> );
>

Since I am very doubtful about the veracity of any claims made by any large language model, which approach should I choose? Indeed, is either approach even valid?