Roundabouts are not analyzed correctly by relation analyzer, when they are connected to oneway part

[PetrDlouhy]

Roundabouts are not analyzed (sorted nor visualized) correctly by relation analyzer, when they are connected to oneway part like this:

 |
 |
 ⋀
| |
| |
| |
 O
 |

More examples are on ticket #5109, see testing cases in roundabout-and-oneway-loop_(tricky).osm​ and example 1 in more complicated cases in relation_analysis.osm​.

[PetrDlouhy]

There is little controversy about this, which I am not sure, how to treat:

When the orientation of oneway loop is opposite to orientation of the roundabout, one part of the roundabout will be driven twice (or never in case of same orientations), when going from beginning of the relation to end and backwards.

This could be contradiction to system, in which oneway parts of the relation are understood. It inconsistency appear more intensively, when trying to split the roundabout into several forward/backward parts - the twice (or never) driven part should, probably be included twice (or never) in the relation.

This will lead to inconsistent relations, when somebody the roundabout without any knowledge about the relations on them.

[stoecker]

NOTE: Validator component reports roundabouts in routes as broken as well.

[PetrDlouhy]

Hm, even plain roundabouts are leading to inconsistency with forward/backward logics, when spitted. I am not sure, how JOSM should deal with that.

There should have been wider agreement about the relation treating in whole OSM, not just JOSM. I didn't find anything on OSM wiki, but I hope my improvements will provoke some discussion about this as well as other implementations.

[bastiK]

I don't see the problem, can be more specific?
For me, there are 2 cases:

I) Simple roundabouts that consist of a single closed way. For the analysis of connections I consider them equivalent to a single node. The current sorting and rendering algorithm takes this abstraction into account.

II) Split roundabouts. These can usually be treated as route with forward/backward parts.

I'll add examples for II).

[bastiK]

The example has a normal highway (A), a 2 lane part (B/C), a roundabout (D,E,F) and a normal continuation (G). The first relation shows the normal way to drive from (A) to (G). The lower is the twisted case, although it seems unusual to me.

[PetrDlouhy]

Yes, the image is how I taught it should be represented with forward/backward roles.

The problem occur, when you have DEF represented as simple roundabout and want to split it without knowledge about any relation, that it is part of. Then you will end up with inconsistent relation without editing it.

This problem could be solved by another message box with complicated automatic correction. But maybe, there is another possibility, how to design the forward/backward feature.

[bastiK]

If you split a simple roundabout that is member of route relations, then you should better know what you are doing. Automating this process sounds like a major task.

[bastiK]

What is the status of this ticket? Can you give a clearer description of the issue you like to address? What component of JOSM is affected?

[skyper]

Anything left that is not covered by #6166, #6920 and there was another ticket with nice examples I do not find right now.