Incorrect matching with hov-only roads

[stl-mikeyalexander]

hov_trip_call.zip

Attached a call which seems to give the incorrect results. When the vehicle moves to the HOV lane, the match fails.

Maps were prepared with (uncompressed output is also attached)
ignore_hov_ways = false

[emiltin]

I verified that hov ways are not considered (ie they are accessible) when ignore_hov_ways = false, so the profile seems to be behave correctly.

(But the naming is a bit ambiguous, use_hov_lanes might be better.)

[stl-mikeyalexander]

I can also confirm that hov ways are being treated correctly in other areas, but the I-95 hov way is never getting used by the match call as far as I can tell.

[stl-mikeyalexander]

Looking at a few of the road segments on I-95.
One segment was missing the access flag (can't find that segment now, but did see it when I first was investigating).
All of the I-95 segments are listed as oneway:yes

[danpat]

You can see what's available in the routing graph by looking at the debug map:

http://map.project-osrm.org/debug/#15.79/38.9791/-76.3235

any edge that is snap-able is visible on that map as a coloured line. The background layer (in grey) is the full road network from OSM. Any place where you can see a grey line that looks like a road and isn't covered by a coloured line is a road edge that OSRM doesn't know about. HOV-only roads fall into this class, like this:

http://map.project-osrm.org/debug/#16.53/38.52032/-77.37062

at this location, the central HOV-only lanes are not routable on the OSRM demo server.

@stl-mikeyalexander you can grab yourself a copy of the https://github.com/Project-OSRM/osrm-frontend project and set up your own local map viewer to dig into this.

[stl-mikeyalexander]

Great tool. Thanks. So that confirms that the I-95 HOV lanes are not in the database and so the match call is doing the correct thing.

What is wrong with these segments that osrm-backend doesn't process them when building the map?
http://www.openstreetmap.org/way/141387703#map=17/38.57284/-77.33047

[danpat]

It's the hov=designated tag. This means those ways are HOV only, non-HOV cars cannot drive on them.

Looking at the logic here is making my brain hurt:

https://github.com/Project-OSRM/osrm-backend/blob/master/profiles/car.lua#L306

It looks like ignore_hov_ways=false will skip HOV checks and include them in the network. ignore_hov_ways=true will look at the hov tag, and if it's designated, we do not route on it.

@daniel-j-h did we accidentally flip the logic here? The true/false values seem backwards and not how I remember it being. We're definitely excluding hov=designated ways on the demoserver, but ignore_hov_ways=true in the profiles/car.lua on master. Am I crazy, or is the wording of ignore_hov_ways backwards?

[emiltin]

yeah as i mentioned above, the naming seems ambiguous, "route_on_hov_lanes" might be better. but the functionality seems correct.

[danpat]

We've re-worded the HOV avoidance flags in car.lua and it's a lot easier to follow. We're also defaulting to allowing HOV access now, but adding a routing penalty that means we'll not route through HOV ways, but will route off or onto them if a waypoint snaps to a HOV-only road (including during map-matching).

ref: #3721