Re: Continue work on changes to recovery.conf API

Hi,

On 2018-09-28 16:36:35 -0400, Tom Lane wrote:
> Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com> writes:
> > I think this was the major point of contention. I reread the old
> > thread, and it's still not clear why we need to change this. _type and
> > _value look like an EAV system to me. GUC variables should be
> > verifiable independent of another variable.
>
> No, they MUST be independently verifiable. The interactions between
> the check_xxx functions in this patch are utterly unsafe. We've
> learned that lesson before.

I'm not sure those concerns apply quite the same way here - we can move
the interdependent verification to the the point where they're used
first rather than relying on guc.c infrastructure. We already have
plenty of checks interdependent that way, without it causing many
problems. UI wise that's not too bad, if they're things that cannot be
changed arbitrarily at runtime.

Greetings,

Andres Freund