Re: pg_get_functiondef forgets about most GUC_LIST_INPUT GUCs

Michael Paquier <michael@paquier.xyz>

From: Michael Paquier <michael@paquier.xyz>
To: Arthur Zakirov <a.zakirov@postgrespro.ru>
Cc: Michael Paquier <michael.paquier@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Postgres hackers <pgsql-hackers@postgresql.org>
Date: 2018-02-21T07:35:50Z
Lists: pgsql-hackers
On Tue, Feb 20, 2018 at 06:46:57PM +0300, Arthur Zakirov wrote:
> Just 2 cents from me. It seems that there is a problem with extensions
> GUCs.
>
> [...]
>
> =# SELECT pg_get_functiondef('func_with_set_params'::regproc);
> ERROR:  unrecognized configuration parameter "plpgsql.extra_errors"

You have an excellent point here, and I did not consider it.
For pg_dump and pg_dumpall, something like what I described in
https://www.postgresql.org/message-id/20180112012440.GA2222@paquier.xyz
to extend pg_settings so as GUC types are visible via SQL could make
sense, now it is really a lot for just being able to quote parameters
correctly.  On top of that, what I suggested previously would not work
reliably except if we have pg_get_functiondef load the library
associated to a given parameter.  Even in this case there could
perfectly be a custom parameter from another plugin and not a parameter
associated to the function language itself.

It seems to me that this brings us back to a best-effort for the backend
and the frontend by maintaining a list of hardcoded parameter names,
which could be refined a maximum by considering lists for in-core
extensions and perhaps the most famous extensions around :(

Thoughts?
--
Michael

Commits

  1. Prevent extensions from creating custom GUCs that are GUC_LIST_QUOTE.

  2. Fix mishandling of quoted-list GUC values in pg_dump and ruleutils.c.