[postgis-devel] (Bug?) Strange Behavior with ST_SymDifference (line, surface)

[Sandro Santilli]

On Fri, Sep 27, 2013 at 04:15:10PM +0200, Rémi Cura wrote:

> Output for line vs surface should be multipolygon, or poylgon with holes (
> flat hole : the parts of the line inside the polygon), or multipolygon with
> holes (flat holes : same).

Your multipolygon would be invalid (two elements of a multipolygon cannot
have a linear intersection), but a GEOMETRYCOLLECTION containing two
polygons could substitute it.

On the other hand the "flat hole" is really impossible to represent with
a valid geometry, I'm afraid, as it'd need a self-intersecting boundary.

> Output for point vs surface should be polygon with holes (1 point hole :
> point inside or on  boundary of surface), or multipolygons with holes (1
> point holes : same).

Same thing, would be an invalid ring.

> I think there would be no problem to store this geometry in postgis ().

No, no problem to store it, but it'd wouldn't pass ST_IsValid.

> Then output of SymDiff would be a SymDiff and not an union when points or
> lines are involved with surfaces (Which makes output false, and anything
> using it possibly false also).

Note that any overlay operation would throw an exception when given
a geometry that doesn't pass ST_IsValid, so you couldn't check.

> At the very least SymDiff should not add points to surface boundary where
> it crossed line (what is the sense of it??).

That's indeed implementation specific, due to noding operation still
being performed as the first step with every overlay.

> Surely enough some output wouldn't be "valid", depending on your criteria
> (quite sure QGis wouldn't like it). Yet it would be possible to warn about
> invalid (for postgis) output and possibly deal with it.
> 
> Shall we switch to the dev list?

Yep, more appropriate, let's continue there.

--strk;

> Thanks for your time,
> 
> Rémi-C
> 
> 
> 
> 2013/9/27 Sandro Santilli <strk at keybit.net>
> 
> > Your ascii art doesn't render correctly with fixed font,
> > unfortunately. Anyway I think I understood the issue.
> >
> > I belive the problem is with mix of dimensions.
> > A line has 1 dimension only, by definition.
> > If we had to make "space" to subtract the points that
> > compose that line, how much space would those points take ?
> > Points have even less dimension: 0...
> >
> > It's like subtracting points from an area and expecting
> > to see holes in it. Does it sound ?
> >
> > --strk;
> >
> >
> > On Fri, Sep 27, 2013 at 02:52:42PM +0200, Rémi Cura wrote:
> > > hey,
> > > I'm experiencing a strange behavior with ST_SymDifference
> > > :<http://postgis.net/docs/manual-2.0/ST_SymDifference.html>
> > >
> > > When doing symdiff of a line and a surface, I get the same surfaces with
> > > added points on surfaces boundary (where line crossed surface).
> > >
> > > Example :
> > >   ___________
> > > *├---------┤*
> > >   -------------------
> > >
> > > gives
> > >   ___________
> > > *|             |***
> > >   -------------------
> > >
> > > I read the SQL MM definition
> > > (here<http://jtc1sc32.org/doc/N1101-1150/32N1107-WD13249-3--spatial.pdf
> > ‎>,
> > > page 67-68-69), and I found no trace of behavior with line against
> > > surfaces.
> > >
> > > Mathematically , if we take for definition the union without the
> > > intersection, the result is false (no point on the line should be in the
> > > result, yet they are).
> > >
> > > I was expecting as output one polygon each side of the line (ie 2
> > polygons).
> > >
> > > I would appreciate any thoughts about this.
> > > Cheers,
> > >
> > > Rémi-C
> > > PS : I know I can use ST_Split or other tricks for the same result in
> > this
> > > simple case, but not so easily for trickier case.
> > _______________________________________________
> > postgis-users mailing list
> > postgis-users at lists.osgeo.org
> > http://lists.osgeo.org/cgi-bin/mailman/listinfo/postgis-users

> _______________________________________________
> postgis-users mailing list
> postgis-users at lists.osgeo.org
> http://lists.osgeo.org/cgi-bin/mailman/listinfo/postgis-users


-- 

 http://www.cartodb.com - Map, analyze and build applications with your data

                                       ~~ http://strk.keybit.net