[Tagging] Through_route next steps

[Eckhart Wörner]

Hi Rob,

Am Samstag, 15. Juni 2013, 13:33:24 schrieb Rob Nickerson:
> The next steps with any tag proposal that reaches a hung jury is to read
> through the comments and update the proposal to address the issues raised.
> 
> In this case I think the wiki page needs to be clearer about what this tag
> is for (a few photo/aerial image examples would help), and how it differs
> from other tags.

I don't think the rejection of the proposal is based on missing illustration.
In my eyes, the proposal almost completely misses the underlying problem: the representation of data in OSM is unsuitable for inferring turn instructions.

The closest thing to a junction in OSM right now is a node: if there are three or more points connected to a node, we call it a junction.
Let's talk a bit about that node: how should we infer routing instructions from the constellation of ways connected to that node? Should we take the classification of the road into account? Most OSM routing programs are doing an exceptionally bad job at this, and yet they are not to blame; since the only way they are actually able to work is by applying heuristics, and the reason for that is simple: there are no conventions *at all*.

The second problem is that a junction is not necessarily node-like at all. Think about divided highways. Or think about links, and you'll soon realize that turns can have an extent as well.

A third aspect of the problem is that even if you are perfectly able to deduce appropriate turn instructions from the geometry, you may still end up with turn instructions that differ from what the signs say. Deviating from the signs is probably not a good idea unless your plan is to confuse the user.

Eckhart