Re: WIP: cross column correlation ...

On Wed, Feb 23, 2011 at 10:30 PM, Bruce Momjian <bruce@momjian.us> wrote:
> Robert Haas wrote:
>> If you want to take the above as in any way an exhaustive survey of
>> the landscape (which it isn't), C seems like a standout, maybe
>> augmented by the making the planner able to notice that A1 = x1 AND A2
>> = x2 is equivalent to (A1,A2) = (x1, x2) so you don't have to rewrite
>> queries as much.
>>
>> I don't really know how to handle the join selectivity problem.  I am
>> not convinced that there is a better solution to that than decorating
>> the query.  After all the join selectivity depends not only on the
>> join clause itself, but also on what you've filtered out of each table
>> in the meantime.
>
> Thinking some more, I think another downside to the "decorate the query"
> idea is that many queries use constants that are supplied only at
> runtime, so there would be no way to hard-code a selectivity value into
> a query when you don't know the value.  Could a selectivity function
> handle that?

Beats me.  What do you have in mind?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company