[geos-devel] 3.1.0rc1

[Paul Ramsey]

Well, I'm seeing the same thing, with a bit more information:

aggtest=# select geometrytype(st_union(the_geom)) from tm_world;
NOTICE:  TopologyException: found non-noded intersection between
21.8917 -26.6689, 21.9536 -26.6644 and 21.9536 -26.6644, 21.8917
-26.6689 21.953 -26.6645
server closed the connection unexpectedly
	This probably means the server terminated abnormally
	before or while processing the request.
The connection to the server was lost. Attempting reset: Failed.
!> \q

The crash is probably the result of bad null/exception handling on my
part, but the core problem is that one of the union pairs is failing.
Basically ST_Union(geom1, geom2) still isn't 100% robust, so if you
find (or create) a pairing that doesn't work, you're SOL.

There's a hack approach to this which would probably work "better",
but still not 100%, and that would be to catch the exception when it
occurs in GEOS, and put the bad pair off to the side everything else
is done. Then try to union(finalresult, bad1) and union(finalresult,
bad2) in the hopes that the specific problem of union(bad1, bad2)
doesn't come up in the union with the finalresult.

I'm going to wrap this data into a MULTIPOLYGON for Hartmut to run,
which should allow him to extract the bad pairing, which we can then
give to Martin and see if JTS also dislikes it.

P.

On Wed, Jan 28, 2009 at 11:54 AM, Dane Springmeyer <blake at hailmail.net> wrote:
> Hi Paul and Hartmut,
>
> Sean wisely mentioned I should be careful in testing shapefiles with invalid
> polygons, which I definitely forgot to think about since I was reading in
> the shapefile directly.
>
> Nevertheless with both this original shapefile [1] (that does have invalid
> polygons) and a cleaned up version (with the buffer(0) trick) that I have
> posted here [2], the below script results in the "should never be reached"
> assertion.
>
> Perhaps the cleaned up shapefile [2] is still invalid for passing directly
> to the cascaded_union,  just let me know how to test further.
>
> Cheers,
>
> Dane
>
>
> [1] http://thematicmapping.org/downloads/TM_WORLD_BORDERS-0.3.zip
> [2] http://dbsgeo.com/tmp/test_world_shapefile.zip
>
>
> ### python ###
>
> from shapely.ops import cascaded_union
> from osgeo import ogr
> from shapely.wkb import loads
>
> ds = ogr.Open('tm_world_buffer_zero.shp')
> state = ds.GetLayer(0)
> states = []
> while 1:
>   f = state.GetNextFeature()
>   if f is None: break
>   g = f.geometry()
>   states.append(loads(g.ExportToWkb()))
>
> print len(states) # 246
>
> u = cascaded_union(states)
> [....snip...]
> cxx type MultiPolygon
> cxx type Polygon
> cxx type Polygon
> cxx type Polygon
> cxx type Polygon
> cxx type MultiPolygon
> cxx type MultiPolygon
> cxx type MultiPolygon
> cxx type MultiPolygon
> Assertion failed: (!"should never be reached"), function itemsTree, file
> AbstractSTRtree.cpp, line 358.
> Abort trap
>
>
>
>
>
>
>
> On Jan 28, 2009, at 10:33 AM, Paul Ramsey wrote:
>
>> Dane, can you provide the shapefile? I will see if this reproduces in
>> PostGIS and generate a test file for Hartmut if it does.
>>
>> P.
>>
>> On Tue, Jan 27, 2009 at 9:40 PM, Dane Springmeyer <blake at hailmail.net>
>> wrote:
>>>
>>> Hi again,
>>>
>>> I just did a bit further testing of the new cascaded union goodness via
>>> Shapely. See the linked email below:
>>> http://lists.gispython.org/pipermail/community/2009-January/001960.html
>>>
>>> Seems that I reached where a point that 'should never be reached' ?
>>>
>>> http://trac.osgeo.org/geos/browser/trunk/source/index/strtree/AbstractSTRtree.cpp#L358
>>>
>>> I am new to looking at the geos code, so any pointers to help me test
>>> more
>>> would be great.
>>>
>>> Thanks!
>>>
>>> Dane
>>>
>>>
>>> On Jan 27, 2009, at 9:02 PM, Dane Springmeyer wrote:
>>>
>>>> Paul,
>>>>
>>>> I tested on mac 10.5 and ubuntu 8.10 and all configure;make and make
>>>> installs went perfectly.
>>>>
>>>> Just a few compiler warnings that I've pasted here:
>>>> http://dpaste.com/hold/113736/
>>>>
>>>> Dane
>>>>
>>>> On Jan 27, 2009, at 10:35 AM, Paul Ramsey wrote:
>>>>
>>>>> This is a pre-warning to PSC members to download and test the first
>>>>> 3.1 rc tarball:
>>>>>
>>>>> http://download.osgeo.org/geos/geos-3.1.0rc1.tar.bz2
>>>>>
>>>>> If I don't hear any howls of dismay over the next 24 hours, I will
>>>>> make a wider announcement for testing.
>>>>>
>>>>> Paul
>>>>> _______________________________________________
>>>>> geos-devel mailing list
>>>>> geos-devel at lists.osgeo.org
>>>>> http://lists.osgeo.org/mailman/listinfo/geos-devel
>>>>
>>>> _______________________________________________
>>>> geos-devel mailing list
>>>> geos-devel at lists.osgeo.org
>>>> http://lists.osgeo.org/mailman/listinfo/geos-devel
>>>
>>> _______________________________________________
>>> geos-devel mailing list
>>> geos-devel at lists.osgeo.org
>>> http://lists.osgeo.org/mailman/listinfo/geos-devel
>>>
>> _______________________________________________
>> geos-devel mailing list
>> geos-devel at lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/geos-devel
>
> _______________________________________________
> geos-devel mailing list
> geos-devel at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/geos-devel
>