> few possibilities.
>
> 1. The routing arcs have a length field, when the arc length is less
> than 4.9m (in our implementation) this is stored as 0. Perhaps zero is
> bad and using 1 instead would work.
>
> 2. Perhaps the nodes must have different Garmin coordinates. This would
> imply a minimum separation of 2.3 (or less depending on direction and
> latitude).
>
> 3. I believe that boundary nodes between two maps are considered
> connected if they are up to 2 (or 3, I forget) garmin units apart.
> Perhaps this is true in general and routing will jump to unconnected
> roads if nodes on nearby roads are too close.

A lot of maybes :-(
I have one more: I think Garmin also changes the algorithmn with new firmware
(one version broke routing completely on my Oregon 450t, don't remember exactly
which one it was)

>
> (I thought of that last in relation to removeShortArcs removing complete
> ways - how did no one notice? Well perhaps because very small gaps can
> be jumped).

The small way is removed, but the gpx device will probably still route
you through it (if my last description of the algo is correct)
I am more concerned that it may remove a way with highway=steps
and connect the nodes of two big street directly instead.

Gerd