Re: Handling of build-time enabled GUC settings

Peter Eisentraut wrote:
> I've noticed some inconsistencies in the handling of build-time enabled GUC
> setttings. For example, a non-SSL enabled server will happily accept ssl =
> on and simply ignore it. On the other hand, several syslog related
> parameters are not accepted at all when syslog is not enabled.
>
> I propose that in the interest of greatest flexibility, configuration-file
> sharing etc. we do it as follows: All parameters are accepted all the time.
> (The exception will be the few odd debugging parameters which are marked
> GUC_NOT_IN_SAMPLE; we can handle those any way we want to.) Those that
> cannot take effect because of build-time configuration have to use a
> check/assign hook to prevent settings that cannot be honored. This would
> mean that ssl = on has to be prevented. As another example, syslog_facility
> and syslog_ident are accepted with all values but log_destination will not
> accept syslog if that is not available. (The latter is the current
> behavior.)

Agreed. We should also mark the parameters that require a server
restart clearer somehow, perhaps '[restart]' in postgresql.conf.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073