[OSM-talk] Maximum recommended length of ways tagged with layer
ricoz.osm at gmail.com
Sat Mar 22 17:38:52 UTC 2014
On Sat, Mar 22, 2014 at 03:33:03PM +0100, colliar wrote:
> On 22.03.2014 11:01, Richard Z. wrote:
> > On Fri, Mar 21, 2014 at 05:05:21PM -0500, Paul Johnson wrote:
> >> On Mar 21, 2014 4:59 PM, "Richard Z." <ricoz.osm at gmail.com> wrote:
> >>> Example of a problem this should catch: I have seen cases where someone
> >>> wanted to tag a simple bridge with layer and added the layer to the wrong
> >>> segment - tagging a hundred or more miles of road accidentally, possibly
> >>> affecting crossings far away for an area not downloaded. The validator
> >> will
> >>> not detect it and in most cases the renderer will work around this bug
> >> very
> >>> well so it is only discovered by accident in most cases.
> >>> This is not limited to layer, I have seen the same problem with culverts
> >> and
> >>> bridges.
> >> This seems like something a validator should be able to catch without
> >> overly complicating how levels work.
> > I am not trying to complicate how layers work right now but trying to codify
> > how they already work in >99% of cases in easy to follow rules that could be
> > utilised by validators.
> > Yes, the validator should be able to catch such situations. Just how?
> > It doesn't right now. I see some possible approaches:
> > * warn user if tagging excessively long ways with "layer". Here the problem
> > is to judge what is excessively long.
> As judgement is difficult and it will still depend on other cases, I do
> not think this will help that much
it is not my preferred approach either, but it might still catch some cases
which might otherwise be hard to catch.
For example I think it is reasonable to assume that if someone tags 200 miles
of a road with layer=1 it is an accident and the rate of false positives would
be very low.
Could this limit be much lower without generating many false positives?
For layers 2,3,4,5 it appears reasonable to assume higher limits because these
are more likely to be used for longer bridges etc.
> > * warn user if applying layer to a way that exceeds the size of downloaded area
> > because in this situation the validator is unable to do even the basic checks.
> even though this will lead to false warnings when working with
> incomplete data, I would give this solution a try.
yes, incomplete data is another problem.. fortunately people downloading partial
data through overpass usually know what they are doing pretty well.
> > * warn user if applying layer to a way without tunnel/bridge/covered/indoor or
> > similar tags.
> Covered is an example where it does not work e.g. you tag the
> building=roof with layer and not the way underneath.
in this case there is nothing to solve.
However "covered=yes" can be used with and without layer depending on situation,
so the validator should allow that.
For JOSM I have the following search string to find suspicious way/layer combinations:
(highway | railway=rail |waterway) -layer=0 layer=* -tunnel=* -bridge=* -culvert=yes -*=steps -*=elevator -covered=yes -indoor=yes
Surely the validator could do a better job than a simple search string but
I am not sure about false positives. If you look at this area
there are quite a few matches and I have no idea if all of those could be tagged
> >>> What kind of underground areas are that in Kansas, do you have a pointer?
> >> I'm not exactly sure where exactly it is, but there's apparently a pretty
> >> extensive underground industrial and office district entirely underground
> >> complete with drivable underground streets in KCK thanks to repurposing an
> >> old mine.
> > interesting, I will have a look when I have some time.
> Could you post a link, please. I wonder how mapnik will work with that
> as there are already problems with a single underground floor/parking.
Found it meanwhile..
looks like nothing is mapped yet? Mapping it will be tricky with
a handheld GPS or Bing - maybe there is some PD data available?
I believe that as it is essentially an underground building it should
be mapped using "level" instead of "layer".
Same for most parking lots and many similar examples.
More information about the talk