Non-Connected way checks in validator needs to be projection independent

[Nakor]

If you run validator the attached file in 3782 it correctly detects "Way end node near other highway".

In 3976 it does not detect it.

[skyper]

In r3992 it does not find any "way end node near other way" anymore.

[skyper]

strange just some minutes ago it worked for some ways but now it does not.

r4001

[Zibik]

Hello

I have the same problem with validator...

[Nakor]

3921 is OK
3925 has the bug

[stoecker]

NOTE: Error due to scaling factor change for mercator code?

[Nakor]

I tried to narrow it down to a precise revision:

• 3921 works fine
• 3922 is broken

[Nakor]

patch

[Nakor]

@stoecker this is the reason. Attached is a patch that fixes the issue

[Nakor]

There is something more. The patch works on small examples but get stuck on big files.

[stoecker]

In [4034/josm]:

see #6102 - partial fix for nearby node checks

[stoecker]

The code depends on display projection. This is not really usable. It either should rely on direct LatLon calculations (e.g. greatCircleDistance() or force e.g. mercator projection for calculations.

[GerdP]

In 15344/josm:

see #6102, #11778: Improve UNCONNECTED_WAYS test

• avoid to fill QuadBuckets structure with nodes outside downloaded area
• reduce dependence on projection, calculate distance as greatCircleDistance
• check if barrier is between unconnected highways
• don't ignore nodes which are only "connected" via a long detour (4 x mindist), this replaces the method Node.isConnectedTo() which counted hops in a rather unpredictable way.
• fix error in Geometry.addIntersections() which didn't return the intersection node when parameter test was true
• don't report ways with different layer=* values

Test now typically produces a lot more warnings, esp. for sidewalks and for service roads like service=parking_aisle. Not sure if the latter should be supressed...