[Imports] ogr2osm: How to deal with multilinestrings?

[Andrew Guertin]

In my fork of ogr2osm (https://github.com/andrewguertin/ogr2osm/),
multilinestrings are currently handled by creating a relation, putting
tags on the relation, and putting subgeometries of the multilinestring
as members of the relation. This is analogous to how, e.g.,
multipolygons are handled.

Paul Norman has reported that when working with roads data from
http://www.tol.ca/Services-Contact/Open-Data/Open-Data-Catalogue, this
handling was not useful to him. Instead, the useful handling would be to
create no relation, and put the tags from the multilinestring on each way.

I've investigated the file, and found three distinct classes of
multilinestrings in it:

1) a dual carriageway
2) cases where a turning circle, loop, or branch in a road was combined
with the main road
3) complicated intersections/interchanges

In all of these, the standard representation in OSM would be to have no
relation and tags on the ways.

My question, then, is whether this is normal. Do other files have
multilinestrings like this, and would benefit from changing the default
handling--or is this file just poorly made, and should be fixed in
file-specific translation routines? Conversely, do other files have
multilinestrings that should be turned into relations, and would be
harmed by changing the default handling?


As a final note, Paul has posted a patch at
https://github.com/andrewguertin/ogr2osm/pull/2 . This patch changes the
parsing of multilinestrings to work as desired. However, I'm tempted to
say that a better change would be to continue parsing as-is, and later
modify the structures as desired. This would be a bit more flexible and
mesh better with other envisioned-but-uncoded data cleanups like
flattening any relations with only one member.

--Andrew