Re: Performance on inserts

> I have no good way to guess how much bit-rot has occurred in all that
> unexercised code ... but it'd be interesting to try to get it going
> again.

Yes, it is a *great* feature, since it directly addresses the problems
associates with one of the most common non-random data distributions
(the index can be constructed to completely ignore those most common
values, and hence be smaller, less often updated, and holding only those
values which might actually be used in an index scan). If we don't get
to outer joins, this would be a good second choice for 7.1 ;)
                    - Thomas