Splitting 4 turn restrictions

I mapped this today Relation: 14409823 | OpenStreetMap - Coming from R-Street one must not turn left into A-Street via the node shared by both streets.

I was told, that I have to split way to at the via node, otherwise, the relation cannot be processed successfully. I see no reason for doing this: In my eyes, that via is an endpoint of from, is enough, to make the turn restriction unambiguous: any of no-*-turn and also only-*-turn at least. I rather not make gratuitous splits. Streets are already split so much.

I might be logically handicapped. So maybe I am missing something. My question therefore: Am I wrong, and splitting the To way necessary? If so, please explain, for whom, and why that is so.

While you are correct in stating that this is unambiguous, that doesnÔÇÖt imply that it is correct tagging or will be correctly interpreted by applications.


If you can return from Andechsstra├če (when driving from the northwest to the southeast) to Reut-Nicolussi-Stra├če, but you cannot return to this road when driving in the opposite direction (from Reut-Nicolussi-Stra├če to Andechsstra├če when driving southeast), then you can do without the relationship: just divide the way Andechsstra├če by 2 at the intersection with the ÔÇťupperÔÇŁ Reut-Nicolussi-Stra├če and add oneway=yes to the segment (from south to north) that connects the 2 Reut-Nicolussi-Stra├če roads. Yes, there is an additional line here, but the relation is not used.

By the way, is it possible to turn around on the AAA road?

1 Like

OTG, there is a double line, one side solid, the other broken on the connecting way between the two carriageways of A-Street, so U-Turns ruled out. It is not enough to map the Only-Right-Turn sign on location. There are no One-Way Signs, and there is no Not-Left-Turn sign either, I just made that up, this relation. Maybe splitting off the connection into an unnamed highway with a made-up one-way would be the most portable mapping. But I have more of the kind on my Osmose scold list, that is not that easy to solve.

I will wait a few days, and report, how the most prominent applications (routers) react.

Preliminary report: Brouter was the quickest, it honored the restriction the other day. Today (4 days after), also Graphhopper got wind of the change. OSRM still wanting, though they usually outpace graphhopper, but from their Debug Map I can see, that other changes in the area still are missing too. So just have to wait a bit more. Openrouteservice also not there yet. OsmAnd maps are also date July 1st.

Is OSM used in proprietary car navigation systems?

PS: I was also browsing the map around the area of my local knowledge for such restrictions. The splitting has been performed quite thoroughly. creating a number of pieces of five to ten meters length primary highway.

Looking once again at the changeset that created the relation, I see clear evidence, that OSRM must have it in its data. I may have overlooked that yesterday. So indeed OSRM does not use the no-left-turn restriction, if VIA is not an endpoint of TO.

I can only speculate as to why: The requirement, that VIA must be an endpoint of TO effectively means, that the algorithm does not need to know, if the restriction is no-left-turn or no-right-turn. That may speed up finding routes. Thankfully the directions given include handedness, otherwise they might be quite unsatisfying.

Openrouteservice still does not have the current state of the map data. I rather not wait for them to catch up, the need to split is already proven by how OSRM processes the data.

To conclude, in this very case, I consider splitting R-Street and mapping a short stretch of one-way the most organic mapping of the OTG restriction.

PS: The OSRM bike profile ignores no-*-turn restriction all the time. It even directs against one-ways, e.g. on stretches as long as 50m, of a close by street. One can always push a bicycle.