[Tagging] natural=bay as nodes are evil

[moltonel 3x Combo]

On 27/10/2014, Christoph Hormann <chris_hormann at gmx.de> wrote:
> On Monday 27 October 2014, moltonel 3x Combo wrote:
>> >
>> > This extremely simple approach will probably result in reasonable
>> > polygons for label placement in more than half the cases.  You can
>> > easily improve the algorithm of course to properly deal with
>> > various special cases, in particular the case of small islands
>> > within a bay deserves consideration.
>>
>> That's a very fragile algorythm.
>
> Have you tried it?
>
> On the contrary - due to its simplicity it is a very robust algorithm,
> it will hardly ever generate something completely wrong and fail
> gracefully in difficult cases.  And as said it is strait away to extend
> this approach to specifically take care of cases where it does not work
> so well.

Since AFAIK it's not implemented anywhere yet (?), neither of us has
tried it. And I guess we have different thresholds as to what
constitudes "completely wrong". I'd be happy to be proved wrong on all
accounts.

>> [...] And until you get something working
>> reasonably well upstreamed in all data consumers, we mappers should
>> bite the bullet and map bays as areas
>
> No, that is not how OSM works.  The mappers can choose a method to map
> they deem appropriate - which in this case quite clearly is nodes (less
> than 0.5 percent ways and relations according to taginfo).  If you want
> to get the mappers to change their mapping you need to convince them
> that it is better to do so

That's what this thread is about : convincing mappers that a
particular way of mapping bays is better. It seems we agree on the
process, if not (yet) on the result.

I wouldn't read too much into the current node/way statistics: they
can either be a sign of imports, or simply of approximate mapping that
hasn't been refined yet (which is the norm in OSM).

> and just making it easier for those
> rendering maps is not a convincing argument, even in cases where unlike
> here there is no additional work involved.  Of course by not rendering
> bays mapped as nodes in the standard style you could 'encourage'
> mappers to change their approach.  This however would be mapping for
> the renderer which is generally frowned upon.
>
> I can't help but notice that in the whole discussion here no argument
> has been put formward indicating a practical advantage of bays mapped
> as polygons other than the ease of rendering labels.

I really don't care that much about rendering. If that was the main
concern, adding some kind of "importance" tag to the bay nodes would
be better. And implementing your algorythm for the renderer, while
quirky, would probably yield very similar rendering results to the
mapping-as-polygon method, with no mapper effort.

I actually care about mapping correctness. A bay is not a point in
space, it's an area. Mapping it as such makes deciding which ones to
render at low zoom easyer, but that's just a bonus.


>> (in other words, not treat them
>> any different than any other area-like obbjects in osm).
>
> You mean like place=town, place=city etc?

Yes, and like everything that can be mapped as a polygon:
amenity=hospital, leisure=pitch, natural=wood, etc etc etc.