[OSM-talk] Abandon all ways: a polemic
lorp at lorp.org
Tue Mar 20 15:22:02 GMT 2007
Frederik Ramm wrote:
> But a quick note about the "polemic" and your suggestion to only use
> segments. Basically, at the very low end of the database, having ways as
> ordered sequences of nodes does pretty much what you say.
The "ways as node sequences" data model is an interesting one, but
still flawed by ways representing something not particularly useful.
There's still that sticky issue that the number of nodes per way is
arbitrary. Such arbitrariness, I feel quite strongly, is a bad feature
of a data model. Are all data models equally good, as long as an ideal
app can cope with them?
> If you thing of a way as an
> attribute set for segments, then your segments will "point to"
> a set of attributes which is more or less what today's ways are.
> A segment being part of a street and a bus route would paint to two
> attribute sets (be part of two ways).
Sure, but you sound like you're arguing for a manually operated data
I don't like how error-prone we are with roads that are not
perfectly uniform: when a road comes to a bridge the way stops, a new
little way starts with one extra "bridge=yes" tag, then the way
restarts with, hopefully, the same attribute set as before the bridge.
The better OSM comes to representing the physical world, the more each
bit of road must be tagged differently from its neighbour. Nice, long
conceptual ways (how humans think of, say, the A4 or Regent Street)
disintegrate into short stretches. Whether we like it or not, we get,
on a single road, multiple ways that need multiple-way editing
controls, with the consistency problem for which the segments-only
approach is criticized.
More information about the talk