Re: GUC flags

On Mon, Jan 31, 2022 at 02:17:41PM +0900, Michael Paquier wrote:
> With all those doc fixes, applied after an extra round of review. So
> this makes us rather covered with the checks on the flags.

Thanks

> Now, what do we do with the rest of check_guc that involve a direct
> lookup at what's on disk. We have the following:
> 1) Check the format of the option lists in guc.c.
> 2) Check the format of postgresql.conf.sample:
> -- Valid options preceded by a '#' character.
> -- Valid options followed by ' =', with at least one space before the
> equal sign.
> 3) Check that options not marked as NOT_IN_SAMPLE are in the sample
> file.
>
> I have never seen 1) as a problem, and pgindent takes care of that at
> some degree. 2) is also mostly cosmetic, and committers are usually
> careful when adding a new GUC. 3) would be the most interesting
> piece, and would cover most cases if we consider that a default
> installation just copies postgresql.conf.sample over, as proposed
> upthread in 0002.
>
> Now, 3) has also the problem that it would fail installcheck as one
> can freely add a developer option in the configuration. We could

I'm not clear on what things are required/prohibited to allow/expect
"installcheck" to pass. It's possible that postgresql.conf doesn't even exist
in the data dir, right ?

It's okay with me if the config_file-reading stuff isn't re-implemented.

--
Justin