Hi,
I found a few issues in r2644, mainly with the link-pois-to-ways option, but also
with short arcs that created when splitting loops.

As discussed before, this option is used to handle nodes with barrier=* or traffic_calming=*
which are so-called CoordPOI.

1) the removeShortArcsByMergingNodes() method may replace

a) a CoordPOI with a normal node.

This typically just leads to some roads where the POI doesn't influence routing.
b) a normal node or a CoordPOI with a(nother) CoordPOI.
In the worst case this might lead to a high speed road having a small part
tagged with access=no, foot=yes, eg. see
http://www.openstreetmap.org/browse/way/37535681

It is likely that the last node with highway=crossing is replaced by the middle node which
has barrier=cycle_barrier

This happens quite often because nodes with barrier=* are typically close
to a node that connects roads.

2) If a CoordPOI replaces a normal node, the highwayCount might be < 2 so that
 later this node is not to connect the roads, leaving some roads with
bollards unconnected.
Sample:
http://www.openstreetmap.org/browse/way/4087549
It is likely that this road gets unconnected, and the
mkgmap:set_unconnected_type feature will not help here
because the road is "disconnected" after the check.

I did not find a simple fix for all issues, sometimes the CoordPOI
has to be ignored, but a lot of routing errors (esp. for pedestrians)
disappear with the attached patch.

The compiled binary is here:
http://files.mkgmap.org.uk/download/133/mkgmap.jar

You should enable logging for the StyledConverter to see new
messages regarding CoordPOI.

Gerd