[Tagging] tagging large intersections cross-road sections

[Lauri Kytömaa]

Florian Lohoff wrote:
> On Wed, Jun 24, 2015 at 10:51:08AM +0200, Tom Pfeifer wrote:
>> Do we have better alternatives? Should we leave the snippet sections unnamed?
>
> I'd love routing people to speak up - I'd prefer to not name them and
> tag them according to be a crossing link.

Three points here:
I thought the originally linked video said that they already detect and deal
with the short sections within intersections and they only tell the correct,
final, road name. So the problem in this discussion has already been
solved, for most cases(?) anyway.

The other point is that the data model probably just doesn't fit all cases, but
it should be possible to derive the correct instructions in either case. The
name of the software escapes me now, (or was it Traveling salesman?) but
years back there was already a solution which, when it was building the
local routing database, detected dual/separated carriageway roads, i.e.
which two highway ways were in fact describing a single "road". It sounds
easy but there were some edge cases etc., as far as I know.

But, given that knowledge, it's possible for the routing directions algorithm
to detect that the name of the crossing road changes in that intersection,
and then ignore the short inter-intersection way's name tag when spelling
out the instructions. It can't universally ignore short named sections,
because the distances between carriageways can be longer than some
single carriageway T-junctions really close to each other.

And, lastly, although not designed for this, there's a not yet so commonly
used tag, or combination of two tags, that implicitly tell us the section is
(at least in part) within an intersection: (as far as I know stopping/halting
in intersections isn't allowed in any countries?)

parking:lane:both=no_stopping (i.e. no halting on either side)
parking:condition:reason=junction (i.e. the no halting rule here is not from
explicit traffic signs, but the intersection rule - it can't be reliably derived
from present data, if one wants to e.g. make a map)

-- 
alv