Hello Marko,
I tested with the FI map, but did not get any anti-island warnings. I tested both without and with --generate-sea=polygons. Sorry, I did not run --generate-sea=polygons without your patch. With generate-sea, I found some anti-treasure like the following.
2010/01/26 22:41:52 WARNING (Osm5XmlHandler): Way null (http://www.openstreetmap.org/browse/way/4611686018427399860) has consecutive nodes with the same coordinates (http://www.openstreetmap.org/?mlat=62.22656&mlon=25.82736&zoom=17) but they can't be merged because both are boundary nodes!
This looks like a generated way (bogus browse URL) and a generated node at a tile boundary, lat=62.226562. The generated node is very close to a real coastline node. Can you avoid adding a node when it is "Garmin equal" to an existing node? There are many lakes (tagged as natural=coastline) at my tile boundaries.
I don't know what's happening there. This generated node you refer to, where's it coming from? That message comes from the short-arc removal code so the node was either in the original OSM data or it has been introduced by the sea generation or possibly by the "soft clipping" code. Does it go away if you don't generate sea?
2010/01/26 22:41:47 WARNING (Osm5XmlHandler): Non-closed coastline segment does not hit bounding box: 28953387 (60,51214/27,91898) 28953375 (60,50997/27,91529) http://www.openstreetmap.org/?mlat=60.51214&mlon=27.91898&zoom=17
This ought to be at the Geofabrik finland.osm.bz2 boundary, i.e., nothing to worry about, for now. The coastline segment belongs to Russia since WW II.
OK. Mark