You can find some of this discussion in "Much Ado About COUNT(*)". Related to that discussion, I had written a patch which added visibility information to the indexes.
If you're interested in the patch and/or consulting, contact me offline.
-Jonah
On 1/18/06, Tom Lane <[EMAIL PROTECTED]> wrote:
David Scott <[EMAIL PROTECTED]> writes:
> Is the additional overhead of keeping full tuple visibility
> information inside of the index so odious to the Postgres community as
> to prevent a patch with this solution from being applied back to the
> head?
This has been discussed and rejected before (multiple times). If you
want it considered you'll have to present stronger arguments than have
so far been made. The current consensus is that the probability of a
net performance win is not good enough to justify the large amount of
development effort that would be required.
What sort of problems are you dealing with exactly? There has been
some discussion of changes that would improve certain scenarios. For
instance it might be plausible to do joins using index information and
only go back to the heap for entries that appear to pass the join test.
regards, tom lane
---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster
