Count actual roundabout exits, not just passed steps #4551
Conversation
… multiple roads leading away.
| // Count all the possible exits from from the step, minus one, which is the | ||
| // "continue on the roundabout" exit | ||
| const auto step_intersection = step.intersections.front(); | ||
| const auto num_exits_from_step = std::count_if(std::begin(step_intersection.entry), |
There was a problem hiding this comment.
This logic doesn't work for throughabouts. If a road (bidirectional) passes through a roundabout, the exit will be counted twice.
There was a problem hiding this comment.
Also, this would need to ignore exits that cannot be entered (exit against roundabout direction)
| | g | i | ga,ci,ci,ci | depart,abcdefa-exit-1,exit rotary right,arrive | | ||
| | g | j | ga,cj,cj,cj | depart,abcdefa-exit-2,exit rotary right,arrive | | ||
| | g | k | ga,ck,ck,ck | depart,abcdefa-exit-3,exit rotary right,arrive | | ||
| | g | h | ga,dh,dh,dh | depart,abcdefa-exit-5,exit rotary right,arrive | |
There was a problem hiding this comment.
How is l decided as being the fourth exit. It looks similar to i in terms of distance from g?
|
@danpat I though we discussed this and decided the current behavior is reasonable? Shouldn't we look for a case of a roundabout that is modeled like this and looks to be done correctly? ...the Logan circle was pretty much bad modeling. |
|
Capturing from chat with @danpat: He still thinks this is the right approach but need to revisit when he has time. |
|
I chatted with @MoKob about this. Next step is to scan the planet for instances and see if this is even a legit way to model things - there's a nonzero chance that it's never correct to exit a roundabout more than once from a single node. If it turns out that all instances of this are from poor modeling, this or should change into something that logs warnings during extraction. |
|
closing stale PR. Reopen if still relevant. |
Issue
When a node on a roundabout has multiple exits directly connected, we only count it as one exit from the roundabout, which throws off the exit number counting for following exits.
Usually this is a result of weird modeling in OSM - multiple exits from the same point on the roundabout is kind of odd, but we found one at http://www.openstreetmap.org/node/49861176, which was where http://www.openstreetmap.org/way/528145111 was connected until @willwhite adjusted it.
This PR looks at the actual exit counts at each step around the roundabout and tallies them up, rather than just counting the number of steps between the entry and the exit.
This could also be a precursor to fixing #2533 - one idea I had was possibly not counting exits that also have a mode change.
@MoKob can you eyeball this and see if my logic is sound? Also open to suggestions if there is a more elegant approach to iterating around the
entry[]array :-)TODO: proper left/right driving side lookup from the profile.
Tasklist
Requirements / Relations
Link any requirements here. Other pull requests this PR is based on?