Re: Sequence Access Method WIP

Hi,

sending new version that is updated along the lines of what we discussed 
at FOSDEM, which means:

- back to single bytea amdata column (no custom columns)

- the dump/restore interfaces were changed to produce/accept array of 
key/value pairs

- psql shows the output of dump interface in verbose mode (this makes it 
easier to inspect the state)

- last_value and is_called are always present columns even if the 
sequence am does not care for them (this helps with backwards 
compatibility with various admin tools)

- pg_dump uses new interface for dumping only for non-local sequences so 
that dumps of schemas which are not using custom seqams are restorable 
to older postgres

- the default_sequenceam was changed to serial_sequenceam and only 
applies to serial/bigserial columns - I would personally prefer to have 
the old default but default for serial is better than nothing

- added description of pg_seqam catalog to docs

- gapless_seq still uses table internally to support snapshots for pg_dump

- pg_dump support for dumping the sequence ams


It would be nice to also have something along the lines of chapter 55.2. 
(Index Access Method Functions), I plan to send that as additional patch 
in few days.

-- 
  Petr Jelinek                  http://www.2ndQuadrant.com/
  PostgreSQL Development, 24x7 Support, Training & Services