[Talk-us] [OSM-talk] Addressing Question

[Stellan Lagerstrom]

Ian Dees wrote:
> * Ok, not "impossible", but the import size would triple and the CPU
> time to compute the new addressing-only ways might make it hard for
> the "regular mapper" to do.
But for no added code and editor complexity.

IMHO the only decent alternative is using a relation for each address
interpolation-range, with the nodes at the ends of the range and the way
itself as members.
This is roughly half as expensive as the Karlsruhe interpolation ways,
and reduces the visual clutter in the editors. It is also insensitive to
way direction reversal.
One of the worst problems is when you split a way which has more than
one such relation (or even one, if the split is NOT between the two nodes).

Consider a way A from node 1 to 2 to 3; interpolation relation R has
members 1[from], 2[to], A[via] and the range info.
Relation S has members 2[from], 3[to], A[via].
A is split into A and B at node 2. If B gets all tags and relations from
A, things get ugly - R and S now have 2 "via" members.
So the editor needs to know how to remove the extra relations so that
only ways that contain both the nodes stay members.

Note: This problem already exists in reverse for turn restrictions. Both
the "from" and "to" ways are required to end at the "via" node. So, say,
Market St crosses 1st St. Market and 1st are both split at the
intersection so a no_left_turn can be inserted. If some well-meaning
editor goes and rejoins the two halves of Market, the relation is at
least formally wrong.

/Stellan