> > attached patch forwards the "guessed" drive-on-left value in a thread
> > safe way to the NOD header by using a ThreadLocal variable. This is a
> > small code change.
> > Anyhow this makes only sense if you think that the guess algorithm is
> > ok.
>
> Shouldn't it be a TRE header? Or TRE header too?
If we use guessing, yes, TRE header too.

>
> According to help for options, guessing is connected with
> "check-roundabouts", which revers directions of roundabouts. This option
> is wrong for tiles, that contain multiple countries with different
> drive-on rules. In that case "check-roundabouts" would damage some good
> roundabouts.
well, "check-roundabouts" also enables some checks in RouteNode
regarding overlapping and forking of roundabouts. I do not yet understand
the meaning of them.

>
> If you make a map for a single country, you don't need guessing. And you
> can't invoke guessing for a map with multiple countries. So this
> guessing feature looks rather redundant.
+1

>
> In my opinion there should be no guessing but a simple count. If a tile
> has more than 50% of roundabouts as drive-on-left, then this tile should
> be marked as drive-on-left. That way most of roundabouts would look
> correctly when navigating.
+1, esp. if you take care to have reasonable tile borders. In my eyes, a value near 50%
should produce a warning.

>
> Perfect solution would be to make tile splitting along country border.
> But if we can't do that, then maybe we could get statistically best
> settings for drive-on flags?
That's a very interesting point. If I got that right, the IMG format doesn't
require a rectangular tile border, so we would need some changes:
1) tile splitter should allow to create such tiles. The bounding polygon
option was a first step into this direction.
2) tile splitter has to pass the information about the used polygon
within each tile. I don't know an "official" way, so maybe this means
to add a way with a tag like splitter:tile-border which has to be filtered
by mkgmap?

Gerd