> I have not yet done anything about this, mainly because I'm unwilling to
> encourage people to use CLUSTER, since it's so far from being ready for
> prime time (see TODO list).
Well imho making the optimizer cluster aware is a step that has to be done
anyway. No need to advertise the cluster feature, but if a user does
already take the chances of the current cluster implementation,
he deserves the fruits, no ?
Actually I would be using the cluster command on a freshly created table
with an index that I know corresponds to insert order. No risc here.
Imho the assumption that the dba guards the cluster state of his tables
is better than assuming random distribution of a clustered table.
Of course a real statistic for all indices as we discussed before would be
better, but assuming perfect clustering state of a clustered index would be
a good step to do in lack of such a statistic.
Andreas