[Tagging] My "weirdly unnatural aversion to relations"

Kevin Kenny kevin.b.kenny at gmail.com
Tue Oct 2 15:37:34 UTC 2018

On Tue, Oct 2, 2018 at 11:15 AM Paul Allen <pla16021 at gmail.com> wrote:
>> Why selecting buildings and tagging them to site relation is easier than selecting building and adding them to  a multipolygon realation?
> I can't even begin to comprehend how that would possibly work.
> Well, maybe I can.  If we make the outer role of the polygon "not university" then we can add the
> individual, scattered buildings as inner role "university."  Seems bizarre to me, but feasible.  It's
> analogous to the way you can use a multipolygon to define a wood with ponds in it.  Except
> "wood" is a concrete term everyone understands but "not university" is not.
> If we make the outer role "university" then the inner roles have to be all the places that the
> university buildings are not, with role "not university."  Not only do we have the conceptual
> problem of "not university" but it would be very fiddly to map.
> I can't see any way of using multipolygons for this case that makes sense and is easy to do.  If
> you can, then please explain it.
> A site relation, however, is simple.  Just add each building to the relation.  Why do you consider
> this to be no easier than the multipolygon approach?

Are you labouring under the misapprehension that a multipolygon has to
have a single outer ring? Otherwise, what you're saying makes no sense
at all to me.

I routinely map state lands that have quite complicated topology, with
holes, islands, disconnected parcels, cutout rights-of-way, just about
everything you can imagine.
https://www.openstreetmap.org/relation/6362702 is an example of how
irregular things can get.  That's a single multipolygon, and that's
the approach that I'd use for the sprawling campus that you describe.

I'm willing to be convinced that site relations are useful, but your
example doesn't convince me.

I agree that relations are underused, but I also agree that PART of
the reason is that only a few types of relation so far (multipolygon,
route, possibly group) are sufficiently well structured that the tools
have something to go on.

More information about the Tagging mailing list