[Tagging] natural=bay as nodes are evil

[moltonel 3x Combo]

On 29/10/2014, Richard Z. <ricoz.osm at gmail.com> wrote:
> On Tue, Oct 28, 2014 at 05:21:06PM +0100, moltonel 3x Combo wrote:
>> On 28/10/2014, Richard Z. <ricoz.osm at gmail.com> wrote:
> well even if the issues were nonexistent, mapping the area of a bay seems
> to me like mapping an artificially introduced concept for which there is
> very little real world use or recognition otherwise.

Huh ? Forget about maps and osm for a moment. A bay is "a body of
water mostly surrounded by land". You're "in" a bay, not "at" a bay.
It has a size, it's not a point in space with a buoy marking the spot.
It's an area.

The fact that a lot of sources have simplified it down to a point is
an entirely different issue. But there's no reason that, with modern
tools and manpower, we can't make a better job than those historical
sources. And remember that when you see a rendered bay label, you
don't actually know wether the source (wether it's some vector data or
an idea in the sailor's brain) was an area or a point to begin with.

> Also bays with very
> flat or deep geometry will result in disproportionately small areas so
> mappers may feel compelled to do some ugly workarounds if the name of the
> bay isn't shown as expected.

Disproportionate compared to what ? And fairly flat coastlines are a
good example of cases that are tricky for algorythms, where the human
mapper can probably make a better decision.

> So I would say
> * if there is some other reason valid to map the bay as area, do it

pros:
 - bays are areas in real life
 - it makes geocoding trivial
 - it makes knowing which bays to render preferably easy (bigger bays first)
 - it enables representing nested bays
 - it is deterministic, as opposed to relying on a heuristic algorythm
cons:
 - relations are harder to work with than nodes
 - the extent of bays is usually fuzzy; nodes make that fuzzyness obvious
 - most of the existing data (osm and potential imports) are nodes

YMMV, those are reasons enough for me.

> * something better needs to be invented for hinting the renderer.

It's not just the renderer, I actually think that the geocoding
usacase is more important. And geocoding requires an area, wether it
is provided in readily-usable form as osm data, or by a
heuristics-based algorythm that infers it from a node.