[mkgmap-dev] Sea generation

Thu Nov 4 22:04:24 GMT 2010

> On Thu, Nov 04, 2010 at 08:06:18PM +0100, Lambertus wrote:
>> So I'm looking forward to any improvements that work without
>> hand-picking tile borders.
>
> You are cutting a planet extract yourself, aren't you? My problems are
> different, because I am relying on an area extract that is not
> rectangular and lacks information of the bounding polygon. You could be
> suffering from bad data. You know, I have been told that the sea areas
> on the Mapnik rendition are based on some fixed data and the
> natural=coastline lines are ignored. It is relatively easy to keep a
> small area free from mkgmap-reported issues. Usually there are fewer
> than 5 errors when I do a test build, and after fixing the issues, the
> next day I typically get an error-free map. For any bigger area, I guess
> you can never get a mkgmap-complaint-free map, not even if you had the
> resources to fix all the reported errors by the time the next day's
> extract is cloned off.

Some loud thinking:
Would it be helpful if we copy the Mapnik behaviour of well defined 
coastlines? One could define one separate file that contains all 
coastline data (from europe, from the world?). This file could be 
maintained and improved in a better way than it is done in the OSM data. 
natural=coastline would be ignored/removed from the OSM input.
As a second step the separate coastline file would build a basis and the 
data from the OSM could be compared to this. Differences in the OSM 
files are only accepted if they don't invert the sea polygons and if 
they don't build new gaps.

Question:
Has anyone extracted coastline data for the whole world/europe/asia/...? 
How many data is this?

>
> I have been toying with the idea of hand-picking tile borders for
> Estonia as well, but I don't know if it is worth my trouble. I'd guess
> that the country extract (as opposed to an Europe or planet extract) is
> most useful for islands. Finland is an island in some sense: you can't
> cross the Russian border without a passport and visa, the land borders
> to Sweden and Norway in the north are very far away from most of the
> population, and you will have plenty of time to switch maps when
> crossing the sea borders. Sweden, Norway and Estonia are less of
> islands, because they have easily crossable land borders. The map would
> have to lump several countries together in order to be practical, and I
> don't want to start downloading and cutting the Europe extract.
>
> When it comes to incomplete multipolygons that are severed by the tile
> borders, I guess that mkgmap should just ignore incomplete polygons that
> are totally outside the bounding box. Maybe this has been implemented
> already, because I am seeing very few multipolygon warnings nowadays. I
> should check if there is anything wrong about the two multipolygons that
> mkgmap now complains about near the tile borders.

Yeah. The mp code throws all polygons away that don't touch the bounding 
box. Additionally it closes polygons automatically if their endpoints 
lie outside the bounding box.
I haven't checked the mp warnings for long but I think most of the 
warnings are "real" warnings now. With one exception:
The conversion from float to int coordinates sometimes causes some 
polygons to overlap which don't overlap in the original OSM data.

WanMil

>
> 	Marko
> _______________________________________________
> mkgmap-dev mailing list
> mkgmap-dev at lists.mkgmap.org.uk
> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
>