oneway=-1 is broken?

[emiltin]

handling of oneway=1 works.
but it seems handling of -1 is somehow not working, despite the code in ExtractorCallbacks.h?

      | highway | oneway   | forw | backw |
      | primary |          | x    | x     |
      | primary | nonsense | x    | x     |
      | primary | no       | x    | x     |
      | primary | false    | x    | x     |
      | primary | 0        | x    | x     |
      | primary | yes      | x    |       |
      | primary | true     | x    |       |
      | primary | 1        | x    |       |
      | primary | -1       |      | x     | <--- expected
      | primary | -1       | x    |       | <--- actual

Failed Scenario: Oneway street
Time: 2012-02-05T14:02:24Z

<?xml version="1.0" encoding="UTF-8"?>
<osm generator="osrm-test" version="0.6">
  <node id="1" version="1" uid="1" user="osrm" timestamp="2000-00-00T00:00:00Z" lon="1.0" lat="1.0">
    <tag k="name" v="a"/>
  </node>
  <node id="2" version="1" uid="1" user="osrm" timestamp="2000-00-00T00:00:00Z" lon="1.001" lat="1.0">
    <tag k="name" v="b"/>
  </node>
  <node id="3" version="1" uid="1" user="osrm" timestamp="2000-00-00T00:00:00Z" lon="1.002" lat="1.0">
    <tag k="name" v="c"/>
  </node>
  <way id="4" version="1" uid="1" user="osrm" timestamp="2000-00-00T00:00:00Z">
    <nd ref="1"/>
    <nd ref="2"/>
    <nd ref="3"/>
    <tag k="highway" v="primary"/>
    <tag k="name" v="abc"/>
    <tag k="oneway" v="-1"/>
  </way>
</osm>

[test]
cycleway = 19
trunk = 19
trunk_link = 19
primary = 19
primary_link = 19
secondary = 17
secondary_link = 17
tertiary = 15
residential = 15
unclassified = 15
living_street = 13
service = 12
track = 12
path = 12
footway = 10
pedestrian = 5
pier = 5
steps = 3
ferry = 5
obeyOneways = yes
useRestrictions = yes
accessTag = bicycle
excludeFromGrid = ferry
defaultSpeed = 17
obeyBollards = no

Expected: {"from"=>"a", "to"=>"c", "route"=>""}
Got: {"from"=>"a", "to"=>"c", "route"=>"abc"}
Query: http://localhost:5000/viaroute&start=1.0,1.0&dest=1.0,1.002&output=json&geomformat=cmp
Response: {"version": 0.3,"status":0,"status_message": "Found route between points","route_summary": {"total_distance":110,"total_time":22,"start_point":"abc","end_point":"abc"},"route_geometry": "_ibE_ibE?gE","route_instructions": [["Head","abc",111,0,210,"0","NE",22.5]],"via_points":[],"transactionId": "OSRM Routing Engine JSON Descriptor (v0.3)"}
Parsed:
["version", 0.3]
["status", 0]
["status_message", "Found route between points"]
["route_summary", {"total_distance"=>110, "total_time"=>22, "start_point"=>"abc", "end_point"=>"abc"}]
["route_geometry", "_ibE_ibE?gE"]
["route_instructions", [["Head", "abc", 111, 0, 210, "0", "NE", 22.5]]]
["via_points", []]
["transactionId", "OSRM Routing Engine JSON Descriptor (v0.3)"]

Expected: {"from"=>"c", "to"=>"a", "route"=>"abc"}
Got: {"from"=>"c", "to"=>"a", "route"=>""}
Query: http://localhost:5000/viaroute&start=1.0,1.002&dest=1.0,1.0&output=json&geomformat=cmp
Response: {"version": 0.3,"status":207,"status_message": "Cannot find route between points","route_summary": {"total_distance":0,"total_time":214748365,"start_point":"unknown street","end_point":"unknown street"},"route_geometry": "","route_instructions": [],"via_points":[],"transactionId": "OSRM Routing Engine JSON Descriptor (v0.3)"}
Parsed:
["version", 0.3]
["status", 207]
["status_message", "Cannot find route between points"]
["route_summary", {"total_distance"=>0, "total_time"=>214748365, "start_point"=>"unknown street", "end_point"=>"unknown street"}]
["route_geometry", ""]
["route_instructions", []]
["via_points", []]
["transactionId", "OSRM Routing Engine JSON Descriptor (v0.3)"]

[DrVanScott]

i can't reproduce this, but i also found that there is something wrong. I did the following tests with the OSRM WebInterface and a minimalistic test file like emiltin's above.

if the way is defined by points a->b->c and oneway=yes, routing is only possible from a to b. The segment b-c is not routable. The blue line (route) stops at point b. c to a, or b to a are not possible.

If the way is defined by points a->b->c and oneway=-1: Routing is only possible from c to b. The segment b-a is not routable.

If the way is defined by points a->b->c->d and oneway=yes: Routing is only possible from a to c. c-d is not routable.

If the way is defined by points a->b->c->d and oneway=-1: Routing is only possible from d to c (sic!)

[emiltin]

some of this might be more of a bug in the map than the router: if you go into a oneway that's also a dead end, you can never get out again. in your example, if you go to c, you're stuck.

to avoid these 'dead-end' issues, i changed the test to use four nodes for each way, and do the routing between the two middle nodes.

it might still be important to test these cases, since such mapping could easily occur in real use. just need to decide what the desired result should be.

[emiltin]

btw, i'm curious how you test with the gui using custom data?

[DrVanScott]

if you go into a oneway that's also a dead end, you can never get out again. in your example, if you go to c, you're stuck.

Good point. But this would apply to the whole dead end way - not just for its last segment. Also note the strange difference on my last example (4 node way, oneway=-1, two segments not routable). Maybe these problems disappear if the oneway is embedded in between normal ways as you described. I can't test it at the moment. Anyway, the difference (1 segment not routable, 2 segments not routable) could be the outcome of a implementation bug, which could be more seriously in other situations. It's just a feeling...

it might still be important to test these cases, since such mapping could easily occur in real use. just need to decide what the desired result should be.

In my opinion osrm should ignore the oneway attribute for such "stuck dead ends", as this seems to be very likely a mapping error. A warning message could help to fix such errors in osm.

During my tests with real data i came also across situations like this (arrow describes the oneway direction):

-a->b<-c-

Also kind of a dead end.

btw, i'm curious how you test with the gui using custom data?

I referenced my testing data on a clearly identifiable coordinate, not (0,0) or (1,1). The way segments also define a evenly spaced "zig zag", which helps to distinguish them.

[emiltin]

yes it seems something odd is going on. i hope to soon have a version of the cucumber test you can try if you want. being able to share failing tests might make it easier to debug

[DennisOSRM]

please test this one again with the latest source.

[emiltin]

ok i will

[emiltin]

as far as i can tell, there's still a problem. but osrm is also returning empty routes, which messes with the cucumber tests.

[DennisOSRM]

According to your pull request with the cucumber tests, you are missing the lines in your branch that revert the path in case of oneway=-1. Did you remove that on purpose?

[emiltin]

no, not on purpose. i suppose there's some confusion on my part due to the branch juggling.

[DennisOSRM]

I reran the example from above and detected no error. Emil, if you have no objections, I'd like to close this issue.

[emiltin]

no problem. if i find any problems, i'll let you know.