On Wed, May 25, 2016 at 3:03 PM, Alvaro Herrera <
alvherre@2ndquadrant.com> wrote:
>
> Nikolay Shaplov wrote:
> > В письме от 25 мая 2016 13:25:38 Вы написали:
> > > Teodor Sigaev wrote:
> > > > >This all should me moved behind "access method" abstraction...
> > > >
> > > > +1 relopt_kind should be moved in am, at least. Or removed.
> > >
> > > Hm, but we have tablespace options too, so I'm not sure that using AM as
> > > abstraction level is correct.
> > We will use am for all indexes, and keep all the rest in relopotion.c for
> > non-indexes. May be divided options catalog into sections one section for each
> > kind.
>
> That makes sense. I can review the patch later.
>
> > And as I also would like to use this code for dynamic attoptions later, I
> > would like to remove relopt_kind at all. Because it spoils live in that case.
>
> As I remember, Fabrízio (in CC) had a patch for dynamic reloptions, but
> there was some problem with it and we dumped it; see
>
https://www.postgresql.org/message-id/CAFcNs+rqCq1H5eXW-cvdti6T-xo2STEkhjREx=OdmAkK5tiOOw@mail.gmail.com> for previous discussion.
>
Yeah, and it was forked into other long discussion thread [1] that explain the reasons to patch got rejected.
IMHO we need a way (maybe at SQL level too) to define and manipulate the reloptions, and this way should cover all database objects. It isn't a simple patch because we'll need introduce new catalogs, refactor and rewrite a lot of code... but it should be a better way to do it. Anyway we can start with your approach and grow it in small pieces. I have a lot of ideas about it so I'm glad to discuss it if you want.