adamfranco
commented
4 years ago Here's another case of staying in a roundabout too long, then making an impossible-to-drive split:
Carmel interchange roundabout
It looks like this was caused by choosing the roundabout way rather than the entrance way or maybe the join process stepped into the roundabout through the exit way, then continued around the other direction.
One potential option would be to keep track of the direction of one-ways while joining (forward/backward), then not switch one-way directions. This should eliminate at least this particular untravelable case, but wouldn't actually solve the larger loop-back problem noted on US 202.
When joining ways for routes that split and then re-join, there are many cases where the joining process will cause the resulting collection to spiral back on itself at a roundabout or divided highway merger rather than continuing along with the main length of highway.
Here are a few examples:
One potential way to fix this is that instead of ordering ways in a route solely by OSM-id before joining, instead the ways could be sorted such that two-way ways come first, followed by normal one-way ways, followed by roundabout ways, *_link ways, and other connectors. Within these groups, ways would be still ordered by OSM-id. This would allow the join process to always try to attach the two-way non-divided highway ways first when encountering a split/merge. Only if no two-way ways can be joined would the join continue on to the other-type ways.