Open dieterdreist opened 7 years ago
Thanks @dieterdreist for bringing this up. I guess the problem is that routing software doesn't support (multi)polygons at all which would explain the effect you have seen. Medium term I should think that simplifying multipolygon code (because old-style mps don't have to be account for any more) makes it more likely that routers will support (multi)polygons in the future.
I guess we should inform the router developers about the changes happening with multipolygons. Lets start a list here and I'll contact them.
Any others?
Itinero, Brouter, Routino, and maybe pgRouting preprocessors osm2pgrouting and non-free osm2po
I have looked at the data a bit, here are my findings: There are about 24,000 multipolygon relations with a highway
tag with together about 110,000 ways. About 50,000 of these ways have no tag at all, only about 6,000 ways have a highway
tag. So from these numbers, this isn't a small thing. But just looking at the data it seems a lot of these areas have separate ways going through them. About of a third of those multipolygon relations are new in 2017 or were touched in 2017, possibly due to the area fixing effort.
in case of separate ways going through the highway areas: often(?) the areas are needed as well as cross connections.
While I generally agree with changing old style multipolygons to new style multipolygons, I want to mention that there is an issue which let me refrain from doing so in the past in certain cases: the pedestrian routing across areas tagged as highway=pedestrian, area=yes. When I last checked, none of the routers worked if these tags were put on multipolygon relations, but they did when they were on the outer way (routing along the edges of the square, which is not perfect but acceptable, at least compared to having a gap). I agree it is an issue of the routers and not of the data, but I wanted to mention that the current cleanup will likely lead to issues in this field.