[Tagging] Admin Boundary admin_centre or label roles

[Sarah Hoffmann]

On Thu, Sep 30, 2021 at 07:57:34AM +0200, Martin Koppenhoefer wrote:
> 
> 
> sent from a phone
> 
> > On 30 Sep 2021, at 03:05, tguen--- via Tagging <tagging at openstreetmap.org> wrote:
> > 
> > A place node is effectively (and usually literally) the same as the label role. If the label role is deprecated, we should also delete place nodes and move the tags to boundary relations wherever they are present.
> 
> 
> the issue with label nodes is not that they are nodes, it’s that “labelling” is an operation that has to do with map drawing, not with collecting geographic data. It’s a semantic issue 

You are assigning a use to the 'label' role simply based on its name. I highly
doubt that anybody has ever really extracted the 'label' member of a relation
to place the name of the relation on a map. Renderers either simply render place
node names with disregard for the boundaries or they render boundary names with
disregard for the 'label' member. The longish description in the wiki is nice
but highly theoretical.

The root problem here is that we have two different representations for places:
place nodes and boundaries. Both convey slightly different information. Both
are useful. The 'label' role member helps expressing that the two OSM objects
belong together. Nominatim is using that information to deduplicate the data
and show search results with joint information from place node and boundary.
I'm not aware of any other user of the 'label' role.

I'm not particularly attached to the 'label' member. These days the wikidata tag
does a pretty good job for making the link between boundary or place node. (Also
most place nodes that should be a 'label' member are wrongly included with a
role 'admin_centre' into boundaries, as the original poster noted. That severly
limits the use of both roles.) So, Nominatim's use is covered with that.

>From an information point of view, it would be perfectly fine to get rid of the
place nodes for many boundary types and simply add a place tag to the boundary to
preserve the information about the type of place. (*) (The expressed exception here
are human settlements (cities/towns/villages). For these, the place node markes the
center of the settlement. This is a piece of information that cannot be reliably
computed.)

However, giving data users the possibility to just process simple point
information for a place instead of having to compute huge multipolygons when they
don't need exact boundaries... that's not such a bad thing either.

(*) For details why the place type is important, I refer to the first half of my
    SOTM talk: https://www.youtube.com/watch?v=NAzljx1ewZ0

Sarah