Re: Alias collision in `refresh materialized view concurrently`

On Wed, Jun 02, 2021 at 12:30:55PM +1200, Thomas Munro wrote:
> I wondered if we could find a way to make identifiers that regular
> queries are prohibited from using, at least by documentation. You
> could take advantage of the various constraints on unquoted
> identifiers in the standard (for example, something involving $), but
> it does seem a shame to remove the ability for users to put absolutely
> anything except NUL in quoted identifiers. I do wonder if at least
> using something like _$mv would be slightly more principled than
> pg_mv_1234, since nothing says pg_XXX is reserved (except in some very
> specific places like schema names), and the number on the end seems a
> bit cargo-cultish.

Yeah, using an underscore at the beginning of the name would have the
advantage to mark the relation as an internal thing.

+ "(SELECT %s.tid FROM %s %s "
+ "WHERE %s.tid IS NOT NULL "
+ "AND %s.%s IS NULL)",
Anyway, I have another problem with the patch: readability. It
becomes really hard for one to guess to which object or alias portions
of the internal queries refer to, especially with a total of five
temporary names lying around. I think that you should drop the
business with make_temptable_name_n(), and just append those extra
underscores and uses of MyProcPid directly in the query string. The
surroundings of quote_qualified_identifier() require two extra printf
calls, but that does not sound bad to me compared to the long-term
maintenance of those queries.
--
Michael