[OSM-talk] Fixing broken multipolygons

[Sarah Hoffmann]

Hi,

let me add a bit of motivation to this. For osm2pgsql, the software
that process the OSM data for rendering the map on osm.org, we are
currently discussing about changing the algorithm that assembles the
polygons[1]. The new algorithms will be a lot faster but that comes
at the price that it is less tolerant with invalid geometries. A lot
of bad geometries that are currently still drawn some way or another
will be simply dropped. I'm convinced that in the long run this
stricter handling will be good not only for data consumers but also
for mappers, who will see immediately when they made a mistake.
However, in the short run a switch from the old algorithm to the
new one will leave a few bald patches on the map.

There is a comparison map where you can see the changes:

https://osmium.osm2pgsql.paulnorman.ca

There are some notable holes, for example in the woods of
Scandinavia. It would be great if they are gone by the time we
switch the software.

Sarah


[1] https://github.com/openstreetmap/osm2pgsql/pull/684

On Wed, Feb 15, 2017 at 04:20:53PM +0100, Jochen Topf wrote:
> There are a lot of (multi)polygons in OSM that are broken in one way or
> another. And we have to fix them. While some of the broken ones appear
> on the map just fine, some don't appear and some mess up the map. And
> some of those that appear fine on the main OSM map will not show up on
> other maps where different software is used.
> 
> A while ago I set up a web page at http://area.jochentopf.com/ and a
> Github repository at https://github.com/osmlab/fixing-polygons-in-osm
> devoted to that issue that I never announced properly. Go there and read
> up in more detail where the problems are and how we are going to fix
> them.
> 
> Yesterday I launched several Maproulette challenges that allow you to
> easily help with the "cleaning up" effort. Read
> http://area.jochentopf.com/fixing.html for more details on those
> Maproulette challenges. This is a huge effort that will take a long time
> and we really need any help we get. The challenges posted today are only
> the beginning. They only show the about 6,500 ways worldwide tagged as
> buildings (with less than 100 nodes) where the way intersects itself.
> 
> I have decided to start with a simple problem like this, to learn how
> the Maproulette stuff works and how well, you, the community, responds.
> Especially for beginners fixing those building is much easier than
> starting with 10,000 node multipolygons with possibly multiple errors in
> them. (If you want to, you can still do that. All multipolygon errors
> show up in the OSM Inspector areas view at
> http://tools.geofabrik.de/osmi/?view=areas )
> 
> Jochen
> -- 
> Jochen Topf  jochen at remote.org  https://www.jochentopf.com/  +49-351-31778688
> 
> _______________________________________________
> talk mailing list
> talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/talk