Marko wrote:
> If I remember previous discussions correctly, splitter does not split
> lines exactly at the tile border, but instead it includes all nodes
> that are not too much outside the tile.

That is also what I understood. Splitter includes in each tile, the
nodes that fall within the tile, and also the nodes in an overlap region
around the tile. The width of the overlap can be set with --overlap.

> An almost-straight long way can be drawn with few nodes. You could
> help the situation by adding a point to the line close to the
> splitting border.

I would expect it to cause a problem, if a particular way does not have
a node in the overlap region. The difficulty with your fix is that it
would only work for me (and only if I use the same areas.list each
time); it will not work for other users. Also, it will not work for me
in the long term, because data is continuously added to OSM, so the
tiles need to be made smaller as time goes on.

If the overlap region is always bigger than the largest internode spacing between two nodes in a way, then this shouldn't be a problem. Decide a distance, say, 20km, and then set the overlap to be that distance. Finding and fixing ways that violate the rule can be done by a bot.

As a workaround, until the mkgmap problem is diagnosed, increase the overlap. Check out the crosby_integration splitter branch. It can correctly handle large overlaps.

Scott