[Talk-us] Addresses and Tiger

Alan Millar am12 at bolis.com
Mon Jun 1 01:57:55 BST 2009

> > While a relation on the street could be useful for most addresses, it is
> > not sufficient for all.  There is nothing stopping anyone from
> > implementing both, though.

> i don't understand what this would mean in practice -- doesn't it just
> mean that data would have to appear twice in the database, and/or the
> renderers would need to render based on two different schemas?

Implementing both does potentially mean that data could appear twice in
the database.  I wouldn't worry about the computer processing costs of
that.  I would worry about the data accuracy and human processing costs of
maintaining it.  (If the database contains both a Karlsruhe-style
addressing way and some other relation-based form, and a user moves an
addressing node or interpolation point, does the user update both data
sets or will it require smarter tools in JOSM, Potlatch, and Merkartor to
maintain it automatically?)

The renderers would not necessarily need to render both forms.  OSM as an
architecture says that the Osmarender and Mapnik renderers are an
important but only partial part of the total usage of OSM data.  Either of
the two could display addresses on a rendered map using either Karlsruhe
nodes and ways, or a relation-based system.  As far as I know only the
Karlsruhe form is implemented but like all open-source projects, it is
just waiting for an implementor for the other.

However, the bigger issue under discussion, I think, is not how those two
renderers draw pretty pictures, but rather how will routing software make
use of the address information to provide routes.  There are a number of
routing programs which are looking for this data.  There is
openrouteservice.org, local software like Navit and Traveling Salesmen,
and some people's Holy Grail of routing, Garmin GPS onboard routing.

Any one of these users could be implemented to make use of either
Karlsruhe-format data, relation-based data, or both.  None of them must
implement both but are free to do so if they want; it is up to the authors
of the various tools to choose which they want to make use of.

So this is a parallel-path of how to people want to enter and store the
data, and how people want to extract and make use of the data.  OSM can
certainly support both, but I have a feeling that people are likely to
converge on one or the other, and Karlsruhe seems to have the edge at the

Has anyone actually done ANY implementation of a relation-based addressing
system in OSM yet?  I've yet to see any actual examples of either data
collection or use; I've only seen wiki proposals so far.

- Alan

More information about the Talk-us mailing list