Re: WIP: cross column correlation ...

On Feb 23, 2011, at 3:46 PM, Robert Haas wrote:

> On Wed, Feb 23, 2011 at 12:50 AM, Nathan Boley <npboley@gmail.com> wrote:
>>> Personally, I think the first thing we ought to do is add a real, bona
>>> fide planner hint to override the selectivity calculation manually,
>>> maybe something like this:
>>>
>>> WHERE (x < 5 AND y = 1) SELECTIVITY (0.1);
>>
>> If you're going to go that far, why not just collect statistics on
>> that specific predicate?
>>
>> ie,  ANALYZE SELECTIVITY ON tablename (x, y) WHERE (x < 5 AND y = 1);
>>
>> Then it won't fall subject to all of the pitfalls that Tom outlines below.
>>
>> Selectivities are easy to estimate if we know the predicate. They only
>> become hard when they have to work for every possible predicate.
>
> Fair point.
>
> --
> Robert Haas


basically we got the idea of allowing "expressions" in cross column stuff. i think this can be very useful. it would
fixthe problem of a query like that: 
SELECT * FROM table WHERE cos(field) = some_number;

this takes a constant fraction of the table which is usually plain wrong as well (and the error tends to multiply
insidethe plan). 
i am just not sure if i have understood all corner cases of that already.
ultimate goal: get it right for join estimates (this is why a syntax extension is definitely needed - you cannot track
allof them automatically). 
many thanks,
    hans

--
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt, Austria
Web: http://www.postgresql-support.de