Warn about unusually long footway crossings

[gaben]

If a way tagged with footway=crossing + highway=footway and it's longer than around 30-40m it is highly likely an error. The validator should warn about these ways.

The longest crossing in the world is probably in Shibuya, Tokio, which is about 44m long according to ​this source.

[Klumbumbus]

Is this a common error in the database? Can you count the objects with overpass?

[gaben]

Ah, that's not how it works, unfortunately. If a whole crossing is drawn as a single way around eg. a 4-way intersection, it creates too many false-positive results. But with increasing the length it doesn't detect shorter ways which are wrong, just checked on overpass.

Replying to Klumbumbus:

Is this a common error in the database? Can you count the objects with overpass?

Yes, a moment. The global query timed out with 180s.

[Klumbumbus]

Yes, I noticed this too, did some analysis in Germany.

• 55m - ​https://www.openstreetmap.org/way/172439472 (simple crossing with tram)
• 111m - ​https://www.openstreetmap.org/way/321834286 (not even all 4 sides)

With a threshold of 150m these are 57 out of 23017 footway=crossing ways in Germany.

[Klumbumbus]

• 162m - ​https://www.openstreetmap.org/way/538540146
• 170m - ​https://www.openstreetmap.org/way/430495101

[gaben]

I see, thanks for the analysis. My global query timed out once again now I'm rate limited.

Maybe configurable info level check possible? The threshold would be up to the user.

[Klumbumbus]

Replying to gaben:

Maybe configurable info level check possible? The threshold would be up to the user.

That doesn't help to separate the right from the false positives

[GerdP]

I think this would require a lot of logic. You have to detect parts going into one direction and measure the length. If one part is longer than x meters create a warning. Can be done with java but I don't see why.

[gaben]

Yeah, that's the only thing I come up with. Too much work for a little benefit. Wontfix?

[Klumbumbus]

Replying to gaben:

Wontfix?

Yes except someone comes up with a new idea.

[gaben]

I have a new idea. The length check + if a footway shares a node with a higher importance way (residential and up) between the ends. If not, the rule can be applied. What do you think?

[GerdP]

Yes, would reduce the false positives. On the other hand, if no such shared node exists even a short way might be flagged. Thinking about it maybe someone may split a footway=crossing to map a different surface on a crossing island.

[skyper]

Just for the record, do not forget other values for highway like path and cycleway:
highway={0} + {0}=crossing

[gaben]

Replying to GerdP:

Yes, would reduce the false positives. On the other hand, if no such shared node exists even a short way might be flagged. Thinking about it maybe someone may split a footway=crossing to map a different surface on a crossing island.

That's true, but those ways are short, at least shorter than the suggested length (30-40m).

Now I see a different problem. If a footway is parallel to another highway and long (crosses multiple minor roads), the check wouldn't work. Meh :/

Edit: But checking the distance with a different threshold between shared nodes maybe could work. It's going to be very complex again...