Multipolygon validation failures

[pch14@…]

Two cases, where validation does not catch errors in multipolygons from uploading.
Screenshots and OSM file to be attached.

1) Move a node from an inner ring outside of the outer ring
2) Make a section of the inner ring also part of the outer ring

Both times, trying to upload will not display a warning or an error.
Neither will starting validation do so, when nothing is selected at all.

The Error will only show, when starting validation and the relation itself selected.
In case two, this only after having downloaded the missing members thereof.

The error messages differ
1) Intersection between multipolygon ways
2) Multipolygon outer way shares segment with other ring

[pch14@…]

Quick start for testing MP validation

[pch14@…]

Case 1) Intersection

[pch14@…]

Case 2) Shared Segment

[GerdP]

This is a general weakness of the validation that is done before upload. It only looks at modified objects, and the relation itself is not modified and thus it is not tested or uploaded.
The intersection is found when validator is run with empty selection, the shared segment is only detected when the members are complete. This could be improved.

[pch14@…]

Funnily, it depends, on what was selected previously. The intersection e.g. is not detected with empty selection, when the new point is added, while the relation was selected last, which changes both the adjoined screes at the same time.

If the new point is only added to the northern scree there will also be a warning about "Overlapping identical Natural Areas". I understand, that this is quite tricky. Can there be a "dirty" flag for relations, that gets set, when a member is changed?

Fortunately, the error seems to be rare and geofabrik toolbox detects it and a busy kartler quickly repairs that :)

[GerdP]

Replying to pch14@…:

Funnily, it depends, on what was selected previously. The intersection e.g. is not detected with empty selection, when the new point is added, while the relation was selected last, which changes both the adjoined screes at the same time.

If the new point is only added to the northern scree there will also be a warning about "Overlapping identical Natural Areas". I understand, that this is quite tricky. Can there be a "dirty" flag for relations, that gets set, when a member is changed?

I don't understand/believe the part regarding selections. If nothing is selected all elements are tested. If something is selected on the selected elements are tested, but some tests are smart enough to collect additional data needed to test the selection. The MultipolygonTest is not one of them.
You probably created different changes while testing.

Reg. the "dirty" flag: I agree that it would be good to validate also parent objects of changed elements on upload. Still, some tests would not work with incomplete data.

[pch14@…]

Replying to GerdP:

I don't understand/believe the part regarding selections.

I meant that the new node in case 1 will be different, depending on what was selected at the time, that I pulled it from centre of line. But you are right, the error is reported, probably I just forgot to deselect the node.

Still, unlike validation with empty selection, upload validation does not catch the intersection. Shouldn't this be the same? Yet, as soon as the downloaded area has work size map content loaded, validation with empty selection reports lots of stuff, that mappers truly should not be bothered with. So generally, it is good, that its not the same.

PS: Perhaps case 2 is about ordering of the tests - If the "incomplete" test returns a warning, the "sharing" test does not run?

[GerdP]

Yes, with the current code the "shared segment test" is skipped for incomplete multipolygons. I'll try to change that.

[GerdP]

In 16761/josm:

see #19526: Multipolygon validation failures

• find shared segments with outer ways when multipolygon relation is incomplete. Relies on the roles of the complete way members, an empty role is considered to be "outer".

[skyper]

Replying to GerdP:

In 16761/josm:

see #19526: Multipolygon validation failures

• find shared segments with outer ways when multipolygon relation is incomplete. Relies on the roles of the complete way members, an empty role is considered to be "outer".

Nice, see #19534 for the corresponding update request for route relations.

[pch14@…]

If I understand fully, in the next release, shared segments will trigger an error even on incomplete multipolygons, when the validator is started with an empty selection. I understand further, that uploading tests will be more shallow, rightly so, yet therefore not show errors on intersections and shared segments of competing rings, as long as the relation itself was not modified.

In conclusion, this here actually is an "enhancement" type issue, like "Mark multipolygon relations for validation on upload, if members were modified"? From personal experience, having knocked some square kilometres of forest from the map, I see some value in that. In the mean time I will try to validate with empty selection before upload, so as to keep palms from getting sweaty ;)

[GerdP]

The change in r16671 doesn't fix the general problems with the validation of selected or modified objects. It only fixes the problem mentioned in comment:5

[pch14@…]

Single node both inner/outer in MP

[pch14@…]

Hello there, no idea if this applies here, I probably made another mistake, not sure, yet OSM Inspector-AREAS seems to think so; I noticed because it got corrected in ​https://www.openstreetmap.org/changeset/89906991 - in the lower a node of an inner pond touches with the outer area.
Attached osm file of state before correction "MP-single-io-node", no errors/warnings reported whatever I try

[anonymous]

Replying to pch14@…:

Hello there, no idea if this applies here, I probably made another mistake, not sure, yet OSM Inspector-AREAS seems to think so

Hmm, Reading ​https://wiki.openstreetmap.org/wiki/Relation:multipolygon osmi is wrong on this, inner and outer polygons may touch at isolated nodes, it says - mapnik also could render just fine

[GerdP]

I think we can close this as fixed?

[pch14@…]

You are welcome. Feel free to do so. At least a non trivial part is said to be fixed, and I believe that. The remaining, possible enhancement, might be better described in another issue anyways. I do not see the urge for even that, though.