[OSM-talk] Not attaching polygons to roads

moltonel 3x Combo moltonel at gmail.com
Fri Feb 28 00:41:17 UTC 2014

On 27/02/2014, Bryce Nesbitt <bryce2 at obviously.com> wrote:
> On Thu, Feb 27, 2014 at 10:57 AM, Janko Mihelić <janjko at gmail.com> wrote:
>> I think we can divide features to virtual and physical features.
>> Virtual: highway centerlines, waterway centerlines, administrative
>> borders, industrial and residental landuse, parks
>> Physical: riverbanks, buildings, meadows, forests, farm fields
>> Can we make a rule to never share points between these two groups?
> -1. I don't think that grouping is correct.
> First, centerlines model a physical feature.
> Second, what you list as physical features are in fact mostly human land
> uses.  Meadows/forests and even riverbanks are constructed and constrained
> by man.

That grouping makes sense, except that the terms "virtual/physical"
are really badly chosen. I tend to think of them as "1D/2D" or
"line/area" or even "simplified/precise".

A line can sometimes share nodes with an area, for example a
barrier=wall enclosing a natural=wood (assuming the wall is thin
enough to be considered as 1D), or a boundary=administrative running
along a landuse=meadow. And sometimes it shouldn't, such as a
highway=residential along a leisure=park. The rule of thumb is that if
a 1D is used as a simplified representation of a 2D object, then it
shouldn't share nodes with 2D objects.

Editor support for this is tempting, except that it would be fairly
complex (lots of rules to figure out 1D from 2D, problems when tags
change but not geometry, etc), and that node sharing is not "wrong"
per se, just inaccurate.

Once again : sharing nodes is fine, nobody should give out to you if
you initially share nodes between a highway and a park. But it's just
an approximation/simplification; not sharing nodes (and giving the
park its actual shape) is better. And people are entitled to give out
if you glue road to a park that was previously accurately mapped.

> Until you get to a level of micromapping that
> currently covers less than 1% of the planet, the road serves remarkably
> well as the dividing line.  There is no "gap"
> on the ground between the forest and the road: at a first level of mapping
> they abut.

Funny that we generaly agree (there is room for both techniques), but
end up marketing opposed viewpoints :) I prefer to only defend and
suggest the "don't share" approach, because it is the one that is
ultimately better. If somebody is undecided, I'd rather give him the
only technique he'll ever need, rather than giving him two techniques
and explaining the subtleties of when to use which.

I don't feel that mapping the actual park boundary is "micromapping",
and your damned li^W^Wstatistics aren't an interesting argument : when
I map somewhere, I only care about the desired level of detail, not
the current level.

More information about the talk mailing list