On Mon, Mar 10, 2014 at 9:33 PM, Alvaro Herrera
<alvherre@2ndquadrant.com> wrote:
> I haven't touched pg_dump yet, but if this proposed design sits well
> with everyone, my intention is that the dump output will contain the
> pg_register_option_namespace() calls necessary so that a table
> definition will be able to do the SET calls to set the values the
> original table has, and succeed. In other words, restoring a dump will
> preserve the values you had, without a need of having the module loaded
> in the new server. I think this is what was discussed. Robert, do you
> agree?
No, I wasn't imagining anything like pg_register_option_namespace().
My thought was that you'd need to have any relevant modules loaded at
restore time. In essence, patching in a new option via an extension
module would work about like adding one by patching the core code: you
need a server version that supports that option in order to set it.
I don't like the idea of using reloptions to let people attach
arbitrary unvalidated settings to tables. I consider the way things
work with GUCs to be a bug, not a feature, and definitely not
something I want to propagate into every other area of the system
where the underlying storage format happens to allow it.
I also kind of think that what you're going to find if you try to
press forward with the pg_register_option_namespace() idea is that
what you really want is CREATE RELOPTION NAMESPACE, ALTER RELOPTION
NAMESPACE, DROP RELOPTION NAMESPACE. Short of that, you're going to
end up with a bunch of kludges, I suspect. And some kind of real DDL
syntax (with better naming) is OK with me, but as you observed
elsewhere on the thread, now you're looking at a new catalog and a
bunch more complexity.
I kind of think that this is too half-baked for 9.4 and we ought to
punt it to 9.5.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
