On 18.11.2010 14:58, Heikki Linnakangas wrote:
> On 17.11.2010 19:36, Teodor Sigaev wrote:
>>> Hmm, will have to do some benchmarking on that. I'm using the Consistent
>>> function when walking down to check if the downlink needs to be updated,
>>> and assumed that it would be insignificant compared to the cost of
>>> calling Penalty on all the keys on the page.
>> Why consistent?! It's impossible - you don't know right strategy number,
>> index with storage type/over type could do not accept the same type as
>> query. Index over tsvector is an example.
>
> Sorry, I was confused. I'm calling the gistgetadjusted() function, which
> uses the Union function. Ie. I'm doing the same we did before when
> propagating the changes up the tree. I'm just doing it on the way down
> instead.
>
> I ran some quick performance tests on my laptop, and couldn't see any
> measurable difference with the patch. So I think we're good on
> performance. I used the attached scripts, with \timing.
>
> Have you had a chance to look at the patch yet? I'm hesitant to commit
> before you take a look at it, though I still have to proofread it myself
> as well.
Here's an updated version with some minor fixes. I'd appreciate review,
as well as pointers to good test cases for this.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com