[postgis-devel] RectDistance

[Paul Ramsey]

I have now added a _ST_DistanceRectTreeCached(geom, geom) that does
caching, so for cases with nested loop joins we should see improved
performance. There's still cases (small geom vs small geom) that are
handled better with the existing code. And I recall that Nik found
some large cases that are degenerate too?
Anyways, it seems like the final implementation of Distance() will end
up being a hodgepodge with some magic number rules in the middle
determining which code line to use.
P


On Tue, Feb 20, 2018 at 1:53 PM, Paul Ramsey <pramsey at cleverelephant.ca> wrote:
> OK, it's in there, for those who want to try it out,
> _ST_DistanceRectTree(g1 geometry, g2 geometry)
> There is no caching in this version, so it's building and tossing away
> the trees at every invocation.
> P.
>
> On Tue, Feb 20, 2018 at 1:49 PM, Bborie Park <dustymugs at gmail.com> wrote:
>> +1
>>
>> On Tue, Feb 20, 2018 at 12:14 PM, Regina Obe <lr at pcorp.us> wrote:
>>>
>>> No objections from me.  +1 to just bring it in 2.5.
>>>
>>> I'm too lazy to test another branch, but if in place, I'd be willing to
>>> test it out.
>>>
>>> Thanks,
>>> Regina
>>>
>>> -----Original Message-----
>>> From: postgis-devel [mailto:postgis-devel-bounces at lists.osgeo.org] On
>>> Behalf Of Paul Ramsey
>>> Sent: Tuesday, February 20, 2018 2:26 PM
>>> To: PostGIS Development Discussion <postgis-devel at lists.osgeo.org>
>>> Subject: [postgis-devel] RectDistance
>>>
>>> Hey all,
>>>   I've had a branch running for a while, that Nik has tested a little, for
>>> tree-based distance calculations. I've been doing some testing on it and
>>> finding that when it's good, it's very very good (10x improvements when
>>> dealing with large geometries) and when it's bad, it's not so great (3x
>>> worse than existing naive distance on small-vs-small geometries).
>>>   All this so far without also reaching into the next obvious enhancement,
>>> adding a caching behaviour, like the point-in-polygon/geos code. That will
>>> make everything about 2x faster yet, as about 50% of the time seems to be
>>> spent in building up trees.
>>>   I'd like to merge my branch in, along with a temporary testing function
>>> _ST_DistanceRectTree(geom, geom) so that we can figure out how we exactly
>>> want to put this code to use. Any objections? If nothing else, it will
>>> replace the current dead code lying in lwtree.c.
>>>
>>> P.
>>> _______________________________________________
>>> postgis-devel mailing list
>>> postgis-devel at lists.osgeo.org
>>> https://lists.osgeo.org/mailman/listinfo/postgis-devel
>>>
>>> _______________________________________________
>>> postgis-devel mailing list
>>> postgis-devel at lists.osgeo.org
>>> https://lists.osgeo.org/mailman/listinfo/postgis-devel
>>
>>
>>
>> _______________________________________________
>> postgis-devel mailing list
>> postgis-devel at lists.osgeo.org
>> https://lists.osgeo.org/mailman/listinfo/postgis-devel