[OSM-dev] Using NULL values in mapnik / postgres DB

[Jon Burgess]

On Tue, 2007-02-20 at 23:20 +0000, Artem Pavlenko wrote:
> > I'll have to take a look. Would linemerge fix up the ways looping back
> > on themselves that you reported before?
> That's the plan
> > What node function are you 
> > thinking of and what would it do?
> >
> 
> Here are some examples of invalid ways (geometrically) :
> 
> 1.  http://artem.dev.openstreetmap.org/files/notsimple_way.png  
>    osm2pgsql actually outputs three LINESTRINGS from this way but one is 
> self-tangent (see example 3)
> 
> 2.  http://artem.dev.openstreetmap.org/files/notsimple_way-2.png  - 
> self-intersecting way.
> 
> 3. http://artem.dev.openstreetmap.org/files/notsimple_way-3.png - 
> self-tangent. Only start and end points can match!!!
> 
> 
> 'Noding' is an operation that given collection of LineStrings will produce a 
> new one where none of LineStrings are crossing each other  ( in example [2] 
> it will insert extra node at intesection).
> 
> 'LineMerge' will join segments into simple geometries (the current segment 
> join algorithm produces cases like [2])
> 
> I can code all this in C++  and see if it works and then we can try to merge 
> it into osm2pgsql?  Let me know!

Handling 1 & 3 should be fairly easy to add to the WKT function. It
already locates segments which attach to a given node, it should be
possible to make it split the way if it finds more that 1 matching
segment.

Handling intersecting lines (example 2) is more tricky. I could work out
the math for detecting this if I thought about it for a bit but my old
vector math is a little rusty at the moment. If someone has a link to
some neat algorithm then it would be handy :-)

If you wanted to port osm2pgsql to C++ it should be fairly straight
forward. Some of the code could be simplified by making use of the more
advanced libraries and data constructs.

	Jon