Tom Lane wrote:
> I wouldn't mind seeing people be a little more vocal on the hackers list
> about what they plan to be doing, just so that there's not duplication
> of effort.
Stuff I have done in some form that I need to finish up and submit:
- GiST improvements: sane memory management, 10% scan perf. improvement
(not sure if I'll get to WAL and page-level locking for 8.1)
- CREATE TABLE AS overhaul & SQL 2003 compliance
- pl/pgsql dead code checking (only for trivially-dead code)
- default_with_oids=true by default
- use # of CPUs at runtime to adjust spinlock behavior for UP systems
Interested in looking at for 8.1 but no code yet:
- PREPARE planning improvements (at the least, do the planning when we
see the first EXECUTE, as in the fe/be protocol-level prepared statements)
- various planner improvements; haven't really decided what specifically
to do, yet
- logical column ordering, and possibly repacking of physical order of
columns to optimize disk space consumption by reducing alignment/padding
requirements
- O_DIRECT for WAL
- UNIQUE predicate per SQL2003
Investigated, probably not worth pursuing:
- GCC PGO support. At least in GCC 3.4, PGO is sufficiently flaky it
isn't really worth adding support for it. Maybe I'll take another look
when GCC 4.0 is out.
- futexes in PG spinlocks. Didn't solve the CS problem, perf.
improvement possibly (?) not worth the portability headaches.
(Of course, absolutely no guarantees that I actually get around to
implementing any of this stuff, this is just what's on my mind at the
moment...)
-Neil