[Patch] Detect all self-intersecting ways (multipolygons, less common tags, etc.)

[naoliv]

While searching for objects with invalid geometries (using ST_IsValid() with a database) I saw that some objects aren't reported as wrong/invalid in JOSM.

For example, validating the attached file in JOSM nothing is displayed, while I can see a Self-intersection at or near point -49.642125 -21.152750000000001 from ST_IsValid()

Search for problem=yes in the attached file and you should see the intersection.

Shouldn't JOSM catch intersections like this?

JOSM:

Build-Date:2016-06-28 00:41:59
Revision:10501
Is-Local-Build:true

Identification: JOSM/1.5 (10501 SVN pt_BR) Linux Debian GNU/Linux testing (stretch)
Memory Usage: 247 MB / 10206 MB (63 MB allocated, but free)
Java version: 1.8.0_91-8u91-b14-2-b14, Oracle Corporation, OpenJDK 64-Bit Server VM
VM arguments: [-Dawt.useSystemAAFontSettings=on]
Dataset consistency test: No problems found

[mdk]

see also #12843

[Don-vip]

Ticket #12843 has been marked as a duplicate of this ticket.

[Don-vip]

Ticket #13500 has been marked as a duplicate of this ticket.

[GerdP]

I've already posted a patch for multipolygons, see #13307. I think we have to add a test similar to CrossingWays to the SelfIntersectingWay test and maybe filter those self-intersections in the other tests so that they are not reported twice.
Will look into this.

[GerdP]

I've added a patch which
1) improves SelfIntersectingWay so that it also finds 2self intersections at first or last node, Unit test added
2) improves CrossingWays to check each way for self-intersection

One open problem: Should we use the same error code for both ? I think no, but did not yet find a good description.
Once the way crosses itself without a node at the intersection, once there is a node.

Will post a modified patch for #13307 as well.

[GerdP]

I've noticed that the first patch replaced a warning like "crossing barriers" by "Self-intersecting ways". This was not intended. The new patch fixes this and uses the message "Self-crossing ways" instead of "Self-intersecting ways".
I can also create a unit test similar to that for coastlines if needed.

[naoliv]

Ticket #13602 has been marked as a duplicate of this ticket.

[bastiK]

In 11136/josm:

applied #13086 - Detect all self-intersecting ways (patch by Gerd Petermann)

[Klumbumbus]

There is a false positive validator warning for ways which end at a node, which is part of the way and not the first node. This is a common case for highway=* e.g. ​way/209238741

Since I can't remember to see this warning before I assume this patch introduced it.

[bastiK]

Agreed, this is generally not an error.

@Gerd Petermann: I think we should revert the part of the patch for SelfIntersectingWay.java and adapt the tests.

[GerdP]

Yes, I also wondered if highway=* should be treated special. I'll have a look at it tomorrow.

[Klumbumbus]

This is not specific to highways. I think all linear features can have such a shape (railways, barrier,...).

For linear features it should not warn and for areas and multipolygons such a case will be catched by ohter warnings (uncloses area / unclosed multipolygon). So I don't see a need for a warning for this case.

[GerdP]

Please comment the case in new attachment first_and_last.osm
Should this be flagged or not? I think if the p-shaped ways are okay than this is also no problem?

[GerdP]

The patch 13086_mod.patch allows p-shaped and 8-shaped (as in first_and_last.osm) linear ways when the intersection node is first or last node of the way (or both) and modifies the unit tests accordingly.

[bastiK]

In 11149/josm:

applied #13086 - changed self-intersecting ways detection (patch by Gerd Petermann, minor javadoc changes)

[bastiK]

Replying to Gerd Petermann <gpetermann_muenchen@…>:

The patch 13086_mod.patch allows p-shaped and 8-shaped (as in first_and_last.osm) linear ways when the intersection node is first or last node of the way (or both) and modifies the unit tests accordingly.

Thanks, I would say the 8-shape is unusual, but not necessarily wrong. Maybe a Validator report at INFO-level would be okay. (But not for the P-shaped highways and barriers, ... They are too common.)

[Klumbumbus]

I think we should keep it as is now. I don't think we need the 8-shape info level warnig. I assume this would produce more false positives than real errors.

[Don-vip]

Regression in #14202