On the talk page of junction=roundabout, a discussion has started on the presence of [dubious – discuss] tags in the paragraph “Connecting ways” of the wiki of junction=roundabout. Because of the many Maproulette challenges “Incorrect junctions on roundabouts” (such as this one), I thought it would be uncontroversial to remove these [dubious – discuss] tags as I never doubted the need for these challenges. However, some mappers think otherwise so I am starting this thread to ask a wider audience for their opinion.
I don’t have the necessary background myself to understand why it is necessary for routers to work properly to include a section of the roundabout way in the route and it is not sufficient to have only a node of the roundabout in the route. Maybe someone from TomTom can explain? Ping @sbaido
My non-expert belief: routing does not require entry separate entry and exit nodes. It only matters for navigation instructions. If the entry node is also the exit node, the router does not ‘see’ a junction=roundabout tag, so it cannot announce to enter the roundabout and take the first exit. It will say: turn right.
I don’t know if navigations usually use or could use other methods to detect that the route actually enters and exits a roundabout. I guess it could flag all junctions in the route as specific junction types for use in instruction text.
I have mapped a lot of roundabouts, and there is much variation in what (in my mind) people would say to indicate what a driver experiences. Sometimes they would feel that they do enter the roundabout and take the first exit. In that case, I think separate entry and exit nodes best reflect the actual situation. Other times, even though they have to stop for the roundabout traffic, it feels like they turn right before entering the roundabout. Then I would think combining the nodes would be the best representation of reality..
(I also encountered a few situations where an entry way actually intersects an exit way outside the roundabout circle. )
Pull requests welcome**. More seriously, I’d suggest that OSM’s “show roads as directional ways” requires a connecting bit of way on the roundabout if you actually ever join the roundabout and leave it again.
** Actually it’s not clear whether a “fix” would be needed there or at mkgmap itself, but it’d definitely be needed somewhere.
For me specifically: I am not signing up for that right now, I have managerie of hobby projects I have started. And want to finish at least several of them before starting new ones.
Right. It is also related to the tendency of some mappers to mimic vehicle flow by bending the exits/approach ways, instead of having them reflect the exit/entrance angle of the road.
Curious how you would map to “reflect the exit/entrance angle of the road”, to follow OSM guidelines for physical separation and the best practice of aligning to middle of roadway, and also keep a part of the roundabout in the right turn/first exit at roundabouts like this, here?
(I’ve realigned both examples a bit vs how they are mapped in OSM now to best match the imagery, I didn’t upload it because it’s for sake of illustration only)
On this kind of roundabout, the exit and the entrance basically intersect at an angle, and if you want the first exit, you just sort of skim along at the edge of the roundabout.
I tried putting in a “bit of way on the roundabout” for the northeast corner of the second example while keeping reasonable geometry, and it came out 20 cm long.
I’ll be enjoying the popcorn I got for my last message.
Literally the point of those examples though is that the drivers turning right do really enter the roundabout. Physically, they may do so only fleetingly (“skimming” the edge of the roundabout is still physically on the roundabout), but more importantly, legally they are absolutely and unambiguously entering the roundabout. Traffic turning right must yield to traffic already on the roundabout; this would not be the case if said traffic was “not really enter[ing] the roundabout”.
Hence why the existing guidance is that entrance and exit ways should not be directly connected. Such directly connections would not reflect ground truth.
Indeed, the supposed counterargument, that mapping extremely short ways on the roundabout would be “tagging for the router” has the line of logic entirely backwards. Directly connected entrance/exit ways may cause routers to give incorrect instructions because that’s a symptom of that mapping not reflecting ground truth.
People are choosing to forget that the entire purpose of mapping highways as linear ways, instead of as areas, is to form a functional routing map, not to map a physically accurate depiction of the highway. Any argument using the physical geometry of the roundabout as justification for directly connecting entrance/exit ways is therefore missing the point.
By contrast, here’s an example of a roundabout where it would be permissible to connect entrance/exit ways directly (example from the UK, so turning left instead of right, but you get the idea):
On the entrance coming from the southwest, and turning left onto the northwest route, the traffic will “skim” the roundabout. But crucially, the leftmost lane is separated by a painted gore, meaning left-turning traffic does not have to yield to traffic already on the roundabout.
Since there is no physical separation, by convention we would not map this slip lane as a separate way. But we can connect the entrance way directly to the exit way on the roundabout, to denote the distinction between this design and standard roundabout.
Different mappers look at this differently. It’s up to the mapper to judge the situation and apply the best representation.
The linear way representation of roads serves multiple purposes, including routing and rendering. I think no one purpose can be claim to be The One Purpose. Sometimes it’s not easy to combine different purposes in one representation. Then I say: give it your best shot, knowing how others solve this.
I personally put the roundabout way inside the furthest out lane of the roundabout, and use a placement=* tag to say it’s in the outside lane (placement=middle_of:1). This means that the joining roads have more space to play with.
Here is another case where in my opinion entrance and exit should be directly connected because on the southern side you don’t enter the roundabout for a right turn.
Looks like there is support that sometimes connecting the entrance to the exit is appropriate to reflect the real world situation. Is it an idea to work this into the wiki page? It’s about this section:
Do not directly connect approaching with departing roads.
Roundabout with correct connections
Each way which intersects with the roundabout should be connected to it with a separate node. Two ways which enter and exit a roundabout should never connect to the same node on the roundabout.[dubious – discuss] This helps routing applications to provide correct directions, otherwise the routing software may not correctly recognise the roundabout. With a segment of the roundabout between the entrance and exit, routers can safely recognise that a section of the roundabout has to be driven through.[dubious – discuss]
“Do not connect”… and “should never connect” are too absolute. We could add "As a rule, ", then give a clear exception with the reason why entrance and exit share a node.
PS 2026-08-26 Added to the wiki “As a rule..” and described the exception that traffic just skims the roundabout. Could someone with better wiki skills please add the example provided by @eginhard?
Mandating that ways entering a roundabout must not connect to ways leaving a roundabout inevitably leads to mangled geometry where mappers following this guidance feel they have to modify existing mapping just because of what the wiki says. As an example:
I support changing the wiki page to be more neutral and less prescriptive. The wiki page was explicitly cited as reasoning for flagging such issues in the above maproulette challenge, so the wiki page as written is actively causing bad edits.
