Danny Shemesh <dany74q@gmail.com> writes:
> 2. Would it be theoretically possible to use the collected statistics if
> the index where clause is also specified in the query itself ?
> or in other words, if the index only contains records where x is not null,
> and the query also filters on x is not null, would the partial distribution
> not be safe to use ?
The hard part is to avoid double-counting the selectivity of the "is not
null" clause if you do that. It's something that would be nice to do
but it's much trickier than it sounds.
regards, tom lane