On 12/15/2014 08:32 PM, Alexander Korotkov wrote:
> On Mon, Dec 15, 2014 at 7:47 PM, Heikki Linnakangas <hlinnakangas@vmware.com
>> wrote:
>>
>> On 12/15/2014 05:22 PM, Alexander Korotkov wrote:
>>
>>> On Mon, Dec 15, 2014 at 6:20 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>
>>>>
>>>> Alexander Korotkov <aekorotkov@gmail.com> writes:
>>>>
>>>>> On Mon, Dec 15, 2014 at 4:12 PM, Heikki Linnakangas <
>>>>>
>>>> hlinnakangas@vmware.com
>>>>
>>>>> wrote:
>>>>>> Right. I also looked at it briefly, but I wasn't sure if we really want
>>>>>> it. AFAICT, no-one has actually asked for that operator, it was written
>>>>>> only to be an example of an operator that would benefit from the
>>>>>>
>>>>> knn-gist
>>>>
>>>>> with recheck patch.
>>>>>>
>>>>>
>>>> Lack of recheck is major limitation of KNN-GiST now. People are not
>>>>>
>>>> asking
>>>>
>>>>> for that because they don't know what is needed to implement exact KNN
>>>>>
>>>> for
>>>>
>>>>> PostGIS. Now they have to invent kluges like this:
>>>>> [ query using ORDER BY ST_Distance ]
>>>>>
>>>>
>>>> It's not apparent to me that the proposed operator is a replacement for
>>>> ST_Distance. The underlying data in an example like this won't be either
>>>> points or polygons, it'll be PostGIS datatypes.
>>>>
>>>> In short, I believe that PostGIS could use what you're talking about,
>>>> but I agree with Heikki's objection that nobody has asked for this
>>>> particular operator.
>>>>
>>>
>>> "polygon <-> point" is for sure not ST_Distance replacement. I was giving
>>> this argument about KNN-GiST with recheck itself. "polygon <-> point" is
>>> needed just as in-core example of KNN-GiST with recheck.
>>>
>>
>> Right. I don't think point <-> polygon is too useful by itself, but we
>> need an example in core that could make use KNN-GiST recheck patch. We
>> can't write a regression test for it otherwise, for starters.
>>
>> Actually, we probably could've used the circle <-> polygon for that just
>> as well...
>
> Did you mean searching for circles or polygons in the last sentence?
circle as the key, on an index on a polygon column. A circle with radius
0 is equivalent to a point, after all.
I'm looking at the knn-gist with recheck patch now, and I see that it
actually adds two more operators: polygon <-> point (commutator of what
I just committed) and circle <-> point. We might want to split those off
into a separate patch too.
- Heikki
