At 2020-10-28 18:29:30 -0700, andres@anarazel.de wrote:
>
> I think my gut feeling about this still is that it's not worth
> implementing something that doesn't work in postgresql.conf. The
> likelihood of ending up with something that makes it hard to to
> eventually implement proper postgresql.conf seems high.
OK, thanks for explaining. That seems a reasonable concern.
I can't think of a reasonable way to accommodate the variety of possible
modifications to settings in postgresql.conf without introducing some
kind of functional syntax:
shared_preload_libraries =
list('newval', $shared_preload_libraries, 'otherval')
I rather doubt my ability to achieve anything resembling consensus on a
feature like that, even if it were restricted solely to list operations
on a few settings to begin with. I am also concerned that such a feature
would make it substantially harder to understand where and how the value
of a particular setting is derived (even if I do find some way to record
multiple sources in pg_settings—a problem that was brought up in earlier
discussions).
I'm certainly not in favour of introducing multiple operators to express
the various list operations, like += for prepend and =+ for append. (By
the standard of assembling such things from spare parts, the right thing
to do would be to add M4 support to postgresql.conf, but I'm not much of
a fan of that idea either.)
Therefore, for lack of any plausible way to proceed, I do not plan to
work on this feature any more.
-- Abhijit