[OSM-talk] Turn restrictions ambiguity

[Teemu Koskinen]

On Fri, 24 Apr 2009 12:01:20 +0300, Ed Loach <ed at loach.me.uk> wrote:

>> What is your problem with having way sections between each
>> intersection
>> instead of one long way?
>
> I don't have a problem with splitting ways, as that is what I've
> always done to add the relevant tags to the relevant section. But I
> can understand that there is a bit of an issue with doing such a
> thing. By so doing it isn't possible, currently, as far as I know,
> to work out at any given junction which road has priority (if any).
> If we didn't have to split ways, then a way could run as far as it
> has priority. Ways crossing it that had to give way (yield for our
> American readers) could end at the way to indicate they have lower
> priority at that junction. At a 4 way stop (American again), you
> could have 4 ways ending at the same node.

That wouldn't work, as the name or the type of the way that has priority  
at the junction could change. In those cases the way must be split.  
There's also other possibilities when the way must be split and it would  
be then impossible to tell which of the ways has priority (or even cases  
when it would seem that a way has priority while it doesn't really have).


> But we do have to split ways for many reasons and I don't know how
> routing engines work out when one way at a junction has priority
> over another (or whether they even bother - I guess the best
> available at present is to compare names and/or refs).
>
> I did read something about road relations somewhere. I felt at the
> time that these, used carefully, could be used to indicate
> priorities at junction - so if a road crossed a road which had
> priority the lower priority would need a relation for either side
> for example. But this is complex and road relations I feel currently
> are probably unnecessary in most cases (I wouldn't want to create
> one for each residential road, though having said that I (Karlsruhe)
> tagged my first house numbers the other day and did an
> associatedStreet thing, so perhaps such relations will come with
> time).

I'd use relations, but we would need a good scheme for it. Maybe it could  
be done with a relation that groups the pieces of the road, and  
additionally the junction nodes where you must give way (and maybe other  
properties too). The name, ref and all the other constant properties would  
be then part of the relation. That way the renderers could be happy as  
they could just use the relation to draw the name, ref etc. of the road  
while the data was split because of some other property changes, while  
still have the ability to fine grain control for the routers.

>
> Ed
>

Teemu Koskinen