[Tagging] Clarification on the role link in route relations

[Minh Nguyen]

Vào lúc 08:38 2022-01-11, Dave F via Tagging đã viết:
> 
> 
> On 08/01/2022 23:02, Minh Nguyen wrote:
>> Can you provide an example of what this looks like? How does it differ 
>> from what destination:ref or destination:ref:to would be tagged on? 
>> (destination:ref:to is for where a trailblazer sign says "To ABC 123" 
>> with an arrow, as opposed to just "ABC 123" with an arrow.)
>>
>> The current documentation for "link" explicitly states that it's for 
>> link road connections between different road routes, making no mention 
>> of recreational routes. I think it's unlikely that U.S. mappers, at 
>> least, would support comprehensively using the "link" role for 
>> anything that so much as mentions a road route, considering that there 
>> are more maintainable representations of that information. There are 
>> countless "To ABC 123" signs that are many miles away from the route 
>> itself, with nothing in between mentioning the route. Here are some 
>> examples tagged with destination:ref:to:
>>
>> https://www.openstreetmap.org/way/19051231
>> https://www.openstreetmap.org/way/48889597
>> https://www.openstreetmap.org/way/60108744
> 
> 
> These are tags on ways, not route relations.
> Again, I think you are misunderstanding the purpose of route relations, 
> which are journeys, taken over multiple different types of highway.
> 
> Your examples above appear irrelevant to route relations.
> https://www.openstreetmap.org/way/19051231
> 
> destination:ref: The 'I 275 North' is the *only* road that highway goes 
> to. Competent routing software doesn't need to be informed of that.
> 
> destination:ref:to: The 'I 75' is miles away. If these refer to road 
> signs, then it has nothing to do with route relations. There are many 
> turn-offs. A driver's destination could literally be anywhere.

It isn't just my personal preference. You're unintentionally suggesting 
that everyone has mistagged most destinations throughout the entire U.S. 
for over a decade, and that most OSM-based navigation software 
(including what I work on at Mapbox) are in error for handling these 
destinations without issue.

The U.S. designates and signposts road and cycling routes entirely 
differently than you're used to in the UK. In the U.S., highway 
departments primarily number routes, although many routes are coincident 
to a single road, especially on short freeways and on county and 
township roads. By contrast, road inventory numbers are very obscure, 
reserved for internal bookkeeping and emergency response.

Maybe it'll help if I explain how to read the MUTCD guide signs that are 
tagged as destination:ref and destination:ref:to in these examples. For 
reference, here are street-level photos of the guide signs:

<https://www.bing.com/maps?osid=eb31e35e-9573-42e9-a71d-a47a296549f2&cp=39.211236~-84.676298&lvl=19&dir=354.557&pi=0&style=x&mo=z.0&v=2&sV=2&form=S00027>
<https://www.mapillary.com/app/?pKey=527983128577877>
<https://www.mapillary.com/app/?pKey=2886949251528646>

A shield that appears on a guide sign indicates a route carried by the 
adjoining roadway (destination:ref=*). A roadway may carry multiple 
concurrent routes, resulting in multiple shields side by side. Sometimes 
the highway department posts "To" routes (destination:ref:to=*) because 
most traffic is predicted to follow a particular succession of routes to 
reach the control city on the sign (primary destination in UK lingo).

The first photo above shows a fork in the road where a pair of ramps 
leads to opposite carriageways of an unnamed ring road, which is 
coincident to Interstate Route 275 (I-275). From this point, the 
counterclockwise carriageway is designated as simultaneously carrying 
the westbound lanes of I-75 and U.S. Route 52 and the southbound lanes 
of I-275. The clockwise carriageway carries the northbound lanes of 
I-275. [1]

A motorist who keeps left would end up on I-75 West/US 52 West/I-275 
South en route to Indianapolis. A motorist who keeps right onto I-275 
North is likely to be headed toward Dayton. To get there, they would 
need to follow I-75 North after several miles on I-275. The next time 
they'll see an I-75 shield is miles away at the exit for I-75. However, 
no one is forcing them to travel to Dayton; they could instead go in 
circles on a joyride around one of the largest ring roads in the 
country. [2]

Route relations carry essential, structured information about the route 
that cannot be reliably derived from parsing ref=* on the way. The U.S. 
has an incredibly complex system of route networks with often 
conflicting alphabetic prefixes and route shields with distinctive 
shapes and colors, which we distinguish using the network=* key. Route 
relations are important enough to Americans that members of the U.S. 
community are developing a new renderer that uses route relations to 
showcase correct-looking route shields, similar to OsmAnd's behavior but 
more thorough. During testing, we found visual anomalies that were 
caused by "link" members, hence my original question. [3]

In OSMUS Slack, someone pointed out that routers would find it useful to 
access the same structured tags that are found on route relations when 
processing destinations on link ways, versus trying to parse the 
pretty-printed destination:ref=* tag. [4] That would enable navigation 
software to consistently use the correct route shield whether the route 
is mentioned in a maneuver instruction or on the map. But rather than 
overload a route geometry with tangentially related members, I think it 
would be more sensible to put the link way and route relation into a 
superrelation akin to a destinationsign relation (not to be confused 
with a destination_sign relation). [5] Aside from the incredibly 
unfortunate name, this approach is incapable of indicating the 
signposted order of destination cities and destination routes, so 
destination:ref=* is still essential.

I think this thread has already clarified that UK National Cycle Routes 
are signposted in a way that's apparently more conducive to some sort of 
relation role and less conducive to destination tagging, which I'm not 
in a position to dispute. But hopefully my explanation shows why the 
wiki's motorway-specific definition of the "link" role sounded 
particularly baffling to an American like me. At this point, it seems 
like there's a consensus to describe the "link" role as being specific 
to piste routes (approved) and maybe cycling routes in some regions (in 
use), but not for more general use on route relations.

[1] For more about cardinal directions as part of signposted routes: 
https://wiki.openstreetmap.org/wiki/Route_directions
[2] The ring road's inventory number is technically HAM-IR-275, but that 
unsignposted information is too obscure to map.
[3] https://github.com/ZeLonewolf/openstreetmap-americana/issues/79
[4] 
<https://osmus.slack.com/archives/C01TUSZ1Q7J/p1641696330041000?thread_ts=1641614118.015900&cid=C01TUSZ1Q7J>
[5] https://wiki.openstreetmap.org/wiki/Relation:destinationsign

-- 
minh at nguyen.cincinnati.oh.us