[OSM-talk] Divided roads proposal

[Steve Bennett]

On Fri, Dec 4, 2009 at 3:14 AM, Anthony <osm at inbox.org> wrote:
> It doesn't seem to be general enough, but instead as an incomplete
> lump of special cases.

It covers the small, but salient, case of divided roads. Nothing more.
We're just talking about one key here.

> So routers are going to have to handle two completely different ways
> of doing the same thing?

I'm not sure what you're getting at. Topologically speaking, the way a
divided road is handled now is the same as two parallel streets with
occasional cross streets. Routers "handle" that case in the way they
handle any other street routing. With the new divider tag, a divided
road is handled just like a single street - again, a case it already
has to handle. The only difference is that it has to check before
turning right (or left in othe countries), when u-turning, and when
crossing a junction. All these checks already have to be performed,
and there's just a very small amount of additional logic.

(I haven't seen router code, I'm just speculating here, but the tests
involved seem trivial.)

>> 1) Do you think the divider widths as currently recorded are
>> particularly accurate?
>
> The ones I've done are.
>
>> 2) Do you think it's important?
>
> Not of huge importance, but I think it's something important enough
> that it should eventually be recorded.

I should also have asked whether the widths of the roads are
accurately measured. Presumably we need to know the width of each road
and the distance between them, at each point. But if you want to model
the roads that accurately, probably best to keep them as separate
lanes, or use an area tag or something.

> I guess I'm spoiled with access to Yahoo aerials.  Still, if there's
> nothing interesting about the median, why bother mapping it in the
> first place?  You can still add the turn restrictions, which is the
> important part.

Why? Primarily as a landmark, I would think. The fact that there is a
median strip is more important, relatively, than that there is a
median strip which is 83cm wide, surfaced in terracotta pavers
overgrown with moss...

Anyway, to back up slightly here, the benefits of this proposal are:
- much simpler and faster entry of minor divided roads
- better rendering at little cost
- cleaner data structure that better reflects what we're trying to map

The downsides are:
- less specificity of the width of the division

Are there other downsides I'm missing? Could you explain the lack of
generality argument - what cases can't it handle, and why is this
important?

Steve