search

conveyal / r5

Developed to power Conveyal's web-based interface for scenario planning and land-use/transport accessibility analysis, R5 is our routing engine for multimodal (transit/bike/walk/car) networks with a particular focus on public transit
https://conveyal.com/learn
MIT License
285 stars 73 forks source link

Handle GTFS Pathways #803

Open abyrd opened 2 years ago

abyrd commented 2 years ago

We are seeing problems from feeds using the GTFS pathways extension. These contain stops with location_type > 1 for which coordinates are optional. This wreaks havoc with the stop spatial index used in generating GTFS vector tiles, but that's just one symptom of the underlying need to handle non-stop, non-station transit nodes that may have no coordinates.

We need to at least gracefully tolerate feeds with these extensions, and eventually fully reconstruct the pathways within the station even when they have no coordinates.

@ansoncfit mentioned MBTA use of pathways in https://github.com/conveyal/r5/issues/642#issuecomment-771647656_

derhuerst commented 2 years ago

[…] stops with location_type > 1 for which coordinates are optional.

Nit-pick: For entrances/exits (location_type = 2), coordinates are required.

The challenge described persists of course.

derhuerst commented 2 years ago

I'm curious if and how this would affect other implementations, and therefore trying to understand the problem.

Why is it a problem for routing if generic nodes (location_type = 3) and boarding areas (location_type = 4) are missing in the spatial index?

ansoncfit commented 2 years ago

Fully reconstructing pathways would help address #330

abyrd commented 2 years ago

@derhuerst it's not a problem that these nodes are not in the spatial index. The problem was that the nodes without coordinates had coordinate values of NaN which broke the sort algorithm in the spatial index. This didn't just leave them out of the index, it caused a failure when opening the feed.

Indeed my initial comment was wrong with respect to location_type = 2 - the conditions in the PR should now exactly match the rules in the spec.