Thank you for all the feedback so far.

I think that tagging "logical" footpaths as being logical is _not_ "tagging for the <your use case>". If I interpret [1] correctly, this phrase refers to the mis-use of tags. The mis-used tag has no semantic meaning for the original object. Example: Using `natural=wood` to obtain a green area, although the object has nothing to do with wood.

On wiki pages like [2], paths for pedestrians always refer to a physical linear structure. I haven't found a documentation on how to deal with paths which do not refer to a physical linear structure.
The information whether a footway is "real" path or a "virtual" resp. "logical" path, is a semantic property of the mapped object. I don't want to delete those paths, as it seems that too many people rely on them. But I still think there needs to be a method to identify such routes.

As mentioned in an earlier post, one way to identify "logical" or "virtual" paths is with a relation to an area tagged `highway=pedestrian`. To me, this is a bit complicated and may cause unintended side effects when editing any object in the relation. 

So, as far as I can tell, there are three options:

1. do nothing
2. assign a special tag to a "virtual" footpath
3. identify a "virtual path" by relation to an area tagged `highway=pedestrian`.

Any thoughts on these options?

Thanks
Stephan


[1]: http://wiki.openstreetmap.org/wiki/Tagging_for_the_renderer
[2]: http://wiki.openstreetmap.org/wiki/Tag:highway%3Dfootway 

2017-10-08 15:14 GMT+02:00 Stefan Keller <sfkeller@gmail.com>:
>
> IMHO there should be no tagging for the routing enginges, very much
> like there should be no tagging for the renderer.
> In fact, I did found almost no tags related to "virtual" paths in taginfo.
> I'm confident that routing engines with pedestrian profiles will have
> optimized routing over areas/plazas in urban areas rather sooner than
> later (see "area/plaza routing for pedestrian/bike").
> One of challenges routing engines currently face, is that until now
> their preprocessing did not include areas (polygons, multipolygon
> relations), since they concentrated on linear geometries.
>
> :Stefan
>
> P.S. BTW It's also an issue when the pedestrian route starts or ends in areas.
>
> 2017-10-06 23:32 GMT+02:00 Selfish Seahorse <selfishseahorse@gmail.com>:
> > On 6 October 2017 at 22:23, Raphael Das Gupta (das-g)
> > <lists.openstreetmap.ch@raphael.dasgupta.ch> wrote:
> >> Is that so, even if the roads are also mapped?
> >>
> >> Currently, Mapzen and GraphHopper car routers on osm.org navigate along
> >> the street right through the area of highway=pedestrian multipolygon
> >
> > Sorry, my message wasn't clear. What I meant was that pedestrian
> > relations don't work for *pedestrian routing* and that pedestrian
> > areas imply that pedestrians can move freely on that area, which is
> > not true, because there are also roads (this is rather a logical than
> > a rendering/routing issue).
> >
> > As an example, GraphHopper and Maps.me route pedestrians through
> > Münsterhof in Zürich (only along the edges though) [1] but not through
> > Lindenplatz [2]. Münsterhof is tagged as a closed way, Lindenplatz as
> > a multipolygon. Mapzen isn't able to route pedestrians through both
> > squares. [3][4]
> >
> > 1. http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=47.37031%2C8.54138%3B47.37016%2C8.54017#map=19/47.37026/8.54078
> > 2. http://www.openstreetmap.org/directions?engine=graphhopper_foot&route=47.38732%2C8.48567%3B47.38783%2C8.48572#map=19/47.38753/8.48524
> > 3. http://www.openstreetmap.org/directions?engine=mapzen_foot&route=47.37031%2C8.54138%3B47.37016%2C8.54017#map=19/47.37048/8.54116
> > 4. http://www.openstreetmap.org/directions?engine=mapzen_foot&route=47.38732%2C8.48567%3B47.38783%2C8.48572#map=19/47.38743/8.48612
> > _______________________________________________
> > talk-ch mailing list
> > talk-ch@openstreetmap.ch
> > http://lists.openstreetmap.ch/mailman/listinfo/talk-ch
> _______________________________________________
> talk-ch mailing list
> talk-ch@openstreetmap.ch
> http://lists.openstreetmap.ch/mailman/listinfo/talk-ch