Recently I noticed several mappers who convert roads, that are mapped as one usual two-way way to roads that are mapped as two one-way ways. This is necessary, when road has some sort of physical separation, but when is it not? If the opposite directions separated only by road marking line, can such road be mapped as two separated ways? (roads with 4 or more lanes) I said that this does not corresponding to the “Map what’s on the ground”, but I got the answer that highway=* means not the road at all, it’s just the line of the routing graph, and it’s more practical, because the restriction relations are not needed. What do you think about it? Thank you.
Routing is only one aspect. And I do not at all agree to the notion highway=* only represents a routing graph.
Some arguments for you: When you look at all the tags that can come with a highway tag you soon realize that it really represents the physical object: sidewalks, cycle ways, crossings, area=yes, etc. That becomes clear when you consider that a routing graph is very transport mode specific while our way with the highway tag is not. Modeling a not physically separated road as two ways effectively excludes very valid routing options. Even if vehicles might legally not be allowed to cross the middle line this definitely is not the case for all transport modes. Pedestrians might be allowed to cross at all points. Cars might be allowed to pull into a driveway on the other side. Or you might just chose to ignore the legal restriction (which OSM should not claim for itself to enforce).
It’s an established principle to use separate ways if, and only if, there’s a physical separation, and I would encourage people to follow it.
Creating a routing graph for cars is only one of many use cases of highway=* ways. Others include rendering in 2D and 3D (where the existence of separation is relevant for realistic appearance), but also routing or other modes of transport such as pedestrians crossing the street or emergency vehicles which do not need to follow the same restrictions as regular cars. (There are some fire brigades using OSM data here in Germany, for example.) And of course there are many others.
In general, things work best when OSM data follows predictable standards, because that’s the kind of foundation we need to build working software on top of it. So I appreciate that you talk to mappers when you notice such non-standard practices.