Ways are assigned to incorrect changesets when their nodes move

[mojodna]

When a node (501609671) associated with a way (41113036) is moved, the way appears in Overpass's augmented diff output as though it changed (which its geometry did), e.g.

<action type="modify">
<old>
  <way id="41113036" version="1" timestamp="2009-09-19T10:30:53Z" changeset="2532203" uid="110076" us
    <bounds minlat="41.8432372" minlon="12.6208508" maxlat="41.8432812" maxlon="12.6210011"/>
    <nd ref="501609670" lat="41.8432812" lon="12.6210011"/>
    <nd ref="501609671" lat="41.8432372" lon="12.6208508"/>
    <tag k="highway" v="service"/>
  </way>
</old>
<new>
  <way id="41113036" version="1" timestamp="2009-09-19T10:30:53Z" changeset="2532203" uid="110076" us
    <bounds minlat="41.8432288" minlon="12.6208433" maxlat="41.8432812" maxlon="12.6210011"/>
    <nd ref="501609670" lat="41.8432812" lon="12.6210011"/>
    <nd ref="501609671" lat="41.8432288" lon="12.6208433"/>
    <tag k="highway" v="service"/>
  </way>
</new>
</action>

However, the only changeset visible in this fragment (2532203) is the changeset in which the way was originally created. As a result, osm-adiff-parser credits the original changeset (2532203) with the modification rather than the changeset which actually modified the geometry (54135745, per http://www.openstreetmap.org/node/501609671).

The clue here that the way shouldn't be credited to its original changeset is that version didn't change. Given that, it becomes necessary to look through the current augmented diff for entries corresponding to any of the referenced ways and use the associated changeset id as the target.

This is touched on indirectly by drolbr/Overpass-API#407. Addressing the problem upstream would likely involve including some indicator about which component was modified such as additional metadata for each nd ref (version, changeset, timestamp, etc.).

[nrenner]

A nice example for this issue is changeset 54743987.

Current

Changeset Map:

Expected

achavi:

See also OSM History Viewer.
By the way: History Viewer has a nice feature highlighting only the changed segments of a way (old red, new green), with unchanged segments in a different color (blue).

Examples

Four ways are shown, because their node refs changed and therefore they are included in the changeset, e.g.:

• way 547767010 Deep Diff: v2 replaced node ref 5292498502 with 1250563352
• way 18349363 Deep Diff: v14 added node refs: 4674218721, 5292498514

All other ways are not shown, because only the coordinates of their nodes changed, e.g.:

• way 18349346: node 189138670 moved
• way 153245700: node 1659531661 moved

I wondered whether the chached JSON files on S3 of the wrongly attributed, original way changeset are updated and how, so I checked for way 153245700 and changeset 53896180:
https://s3.amazonaws.com/mapbox/real-changesets/production/53896180.json

• contains way 153245700 only once --> Ok
• apparently also wasn't updated:
  • closed_at="2017-11-18T10:14:49Z" (changeset 53896180)
  • Last-Modified: Sat, 18 Nov 2017 10:18:40 GMT (curl --head .../53896180.json)
  • closed_at="2017-12-18T22:22:12Z" (example changeset 54743987 a month later)
  • --> Ok

Context

Achavi - when filtering for a specific changeset - includes a way if its versions are the same and one of its nodes is in the changeset.

This issue also applies to relations when members change their geometries:

Relations are considered as changed when their members have changed or one of its way members has changed by its members or coordinates or one of the node members has changed its coordinate. By contrast, changes on members of type relation don't propagate to their parent relations.

https://wiki.openstreetmap.org/wiki/Overpass_API/Augmented_Diffs#Contained_data

cc @batpad @willemarcel