[OSM-talk] Multipolygon fixing effort done

Christoph Hormann osm at imagico.de
Tue Aug 29 10:10:47 UTC 2017

On Tuesday 29 August 2017, Jochen Topf wrote:
> We have completed the 7-months effort to switch away from old-style
> multipolygons and fix a lot of broken (multi)polygons. More about
> this on my blog:
> https://blog.jochentopf.com/2017-08-28-polygon-fixing-effort-conclude

First of all congratulations to what was achieved, i.e. a fairly massive 
reduction in the number of errors.

The sad news however seems to be that given the current circumstances 
the number of errors will likely be back to near the pre-cleanup levels 
in 2-3 years for many types of errors.

From my point of view this is because in contrast to the old style 
multipolygons where

* the problem was fully eliminated in the data
* the most important data user (the standard map) was changed to not 
interpret old style MPs any more after that
* the major editors had already ceased to generate old style MPs long 

the circumstances that lead to the large number of broken multipolygons 
are essentially unchanged.  We certainly got rid of a number of errors 
from bad imports and can hope that future imports will be better in 
that regard but the problem that mappers introduce this kind of error 
in manual mapping and don't realize they are making an error is 

Of the points above both completely eliminating MP geometry errors and 
changing the editors not to upload broken geometries are things that 
are very hard to accomplish.  This leaves the third point and therefore 
my question:

Would the number of visible problems in the map due to dropping broken 
geometries now, after the fixing effort, be low enough so this change 
could be made to the main map to give mappers a better feedback about 
broken geometries?

If the answer to this question is yes (or almost yes) we should not wait 
too long with making this step because once the number of errors has 
risen again this will again become increasingly problematic.

And if this is not considered a practicable step there will of course be 
increasing need for libosmium to be able to fix these errors, at least 
the self intersections. ;-)

Christoph Hormann

More information about the talk mailing list