[OSM-talk] Roundabouts - why is a separate segment required?

Dave F davefoxfac63 at btinternet.com
Fri Feb 16 01:20:03 UTC 2018

On 15/02/2018 08:52, Tom Pfeifer wrote:
> On 14.02.2018 17:39, Dave F wrote:
>> https://www.openstreetmap.org/node/5408566797
> It appears that you already engage in an edit war, although half a 
> dozen people here tell you, from a variety of perspectives, that you 
> are wrong.
> https://www.openstreetmap.org/changeset/56352276

The previous mapper move the roundabout to an *inaccurate* location in 
order "to fudge" a separating section. I legitimately reverted it so it 
was accurately located as per a survey I perform on the ground.

> It would help to maintain objectivity to leave out any rants about 
> particular programmers of particular software. I have seen other 
> routing engines fail when the situation occurs that you have created.
> On 14.02.2018 18:27, Dave F wrote:
> > This would only occur if there was no check to see if it's a 
> roundabout first:
> >   * Enter
> >   * Check if roundabout
> >   * (While still on the same node) Start counting entrances/exits
> In which routing engine did you implement this? From which experience 
> do you speak?
> Or is it just pseudocode that fell off your sleeve without being 
> tested in an implementation?

If a router is unable to perform simple checks it's not worth its salt.

> On 14.02.2018 21:44, Mark Wagner wrote:
>> In the general case, a router only needs to consider the ways that a
>> route actually passes over when creating directions.  By mapping a
>> roundabout entrance and exit sharing a single node, you've
>> introduced a special case: the router now needs to check all ways
>> connected to that node to see if any of them is part of a roundabout.
> Yes and to reiterate, to separate roundabouts from non-roundabouts, 

How does the 'roundabout' tag not  "separate"?

> you would need to check that special case not only at roundabouts, you 
> would need to check its absence at _any_ node connecting _any_ two 
> road segments, even if these are not junctions. You would need to 
> check if there starts a roundabout segment or not, and if all segments 
> of such roundabout loop back to the original node.
> As the check you propose is against the basics of graph theory which 
> is behind routing algorithms, it would create an immense performance 
> burden on the algorithms. That would make you mourn about the skills 
> of the programmers, again.
> May we ask you to undo your revert in CS 56352276? You still have not 
> explained how the two node solution "fudge OSM".

No one's explained why the single node fails. It works fine at crossroads.

More information about the talk mailing list