[PROPOSAL] Move all am-related reloption code into src/backend/access/[am-name] and get rid of relopt_kind

While working on patch for attribute options for indexes (see
http://www.postgresql.org/message-id/5213596.TqFRiqmCTe@nataraj-amd64 )
I found out that code for reloptions is not flexible at all.

All definitions of attoptons are stored in central catalog in
src/backend/access/common/reloptions.c
It is done this way for both heap and tuple relations, and also for all index
access methods that goes with source code.

Most of the code of the indexes is now hidden behind
"access method" abstraction, but not the reloption code. If you grep through
src/backend/access/common/reloptions.c, you will find RELOPT_KIND_GIN,
RELOPT_KIND_BTREE, RELOPT_KIND_GIST, RELOPT_KIND_SPGIST, RELOPT_KIND_BRIN...

This all should me moved behind "access method" abstraction...

Custom indexes, like postgresql/contrib/bloom can add own reloptions, and
relopt_kind. But the number of relopt_kinfd available is short (it would be
enough for reloptions, but if we add attoptions, we will meet shortage soon).

So my proposial is the following: Each access method has it's own catalog of
options (reloptions, and later attoptions) and when it want to call any
function from src/backend/access/common/reloptions.c it uses a reference to
that catalog instead of relopt_kind.

In the patch that is attached to this message, there is an idea of how it can
be done.

In that patch I've left relopt_kind, and added refence to options catalog,
and functions from reloptions.c uses that one that is defined. I've implemented
"catalog reference" version for bloom index, and all the rest still work as
they were.

In final patch there should be no relopt_kind at all, all descriptions of
reloptions for indexes should go to src/backend/access/[am-name], reloptions
for heap, toast, views and so on, should stay in
src/backend/access/common/reloptions.c but should be stored as separate option
catalog for each type. I still not sure what to do  with
RELOPT_KIND_HEAP | RELOPT_KIND_TOAST options. But one or another solutions can
be found...

--
Nikolay Shaplov
Postgres Professional: http://www.postgrespro.com
Russian Postgres Company