To give a bit more comprehensive and nuanced reply than I previously did:
I think there are really two separate discussions going on here:
- type=multipolygon relations vs. closed ways when possible
- Changing one way or the other
So let’s tackle them one at a time:
type=multipolygon relation vs. closed ways
While I prefer multipolygon relations over closed ways for most landuse/landcover, I can definitely understand people having different preferences. Editor support for both ways of mapping is not the best it could be, though also not as terrible as some describe (especially those opposing multipolygons).
Purely theoretically (this should not be the primary consideration, though it should also be taken into account) I also prefer multipolygons, as a border between two landuses/landcovers is by definition a single border, and should thus be represented as a single way.
For free-standing polygons like buildings, parking areas, etc., I definitely think closed ways are the right answer unless strictly necessary. However, landuse/landcover is quite different:
- Given a complete dataset, it should cover the entire area of interest (in OSMs case the entire planet), thus all areas should border against another area (in practice in OSM this is not always the case, as e.g. highways are rarely mapped with a dedicated area and often are a “gap” in the landuse/landcover)
- Areas are generally large and with many nodes (a normal building has maybe 4-10) while landuse/landcover can have hundreds or even thousands
I’ll refer to @Tomas_Marklund’s post and illustration above on why this matters, but to also put it in my own words I’ll give a few example scenarios:
When newly mapping an area (or mapping an area that was previously only very coarsely mapped)
Using multipolygons means I only have to click once per node per landuse/landcover border. Mapping using closed ways would mean twice the number of node clicks on each area (as noted in the first point because each border is shared by two areas, and as noted in point two this usually means very many extra clicks), as well as a risk of misclicking.
When adding something between two existing areas (like adding a beach between a forest and water)
Using closed ways
- It starts with the two existing areas (closed ways)
- First I need to select every single node that should be disconnected and disconnect them (luckily I can do all at once, though I still need to select them each), and then pull each of them apart
- Finally I need to click on each node in the new opening to create the new closed way/area
Using multipolygons
- It starts with the two existing areas (multipolygons with a shared border)
- Next I’ll draw a new way which will be the border between beach and forest and cut the old forest/water border
- I remove the forest relation from the part of the old border that will now be between beach and water and add the new way to the forest relation
- Finally I can just select the two ways and create a new multipolygon for the beach
Using multipolygons might seem like they have more steps, however in practice they are almost always faster and easier (once you get used to them), as drawing new ways is a lot faster than adjusting existing nodes.
When moving a border between two existing areas
When free-standing there is no significant difference between using multipolygons and closed ways as both mean moving each node manually. It can however be worth it to draw a completely new border if it is very long (though this will lose the history) if it is a multipolygon, which is a slight advantage.
Parks, place=island, etc. I generally prefer to handle the same as landuse/landcover, as they usually share a border with the landuse/landcover (place=island has the same outer members as the water has inner members for that particular island).
The exception to the rule
There is also a case where I map landuse/landcover using closed ways rather than multipolygon; when it is completely enclosed by one other multipolygon. For example farmland impediments and islands of uniform landcover (the closed way also being an inner member of the farmland/water multipolygon), overlapping landcover on landuse (like natural=wood on landuse=meadow/residential, no inner member in this case). This is mostly for convenience in iD, as I can just draw the the area and tag it, rather than draw line, create multipolygon relation, tag it.
Possible editor improvements
All this is written from the point-of-view of a primary iD-user, I can’t really comment on how these arguments are different in JOSM or other editors.
There are definitely improvements that can be done to iD, both for making multipolygons and closed ways easier to handle. Some examples:
- Downloading of entire multipolygon relations (automatically or on user command)
- This would be a requirement for better validation as well as get rid of the invalid rendering of large multipolygon relations
- Improved validation of multipolygon relations, and possibly preventing saving entirely on some issues
- Automatically assigning roles when possible (example: when creating a new multipolygon relation and all included ways form one valid outer ring, they should all be given the outer role)
- Better editing of existing ways (this would be an improvement to both approaches), something like “select existing start node, draw new geometry, end on existing node”
Changing one way or the other
Having covered that, I think we can agree on one thing: Both approaches have their merits, and there is no clear community consensus. As such, I think doing edits for the sole purpose of changing the way of mapping is definitely wrong. If either camp wants to change things, these things should be done first:
- Get clear community consensus (not “everyone agrees”, but a majority agrees or at least finds it acceptable)
- Adjusting wiki etc. to reflect this change
- Do mass-edits
I quite strongly think that @Friendly_Ghost should immediately cease these edits until the above steps have been followed.
There is also the history to consider. I somewhat frequently check the history of landuse/landcover elements (something like “based on this satellite imagery I think this should be farmland, not residential, but the imagery might be old and not should new development, so I need to know when it was added and based on what source”). And these mass-edits make this practically impossible past the mass-edit.
I have personally considered doing this kind of mass-edit on areas I’m mapping, only in the other direction (changing landuse/landcover closed ways into multipolygons), but have not done so because of a lacking consensus (and me not having the time). Though all this has made me start re-considering…
There might also be completely different solutions like adding a dedicated area datatype, though that would be a quite hard problem to solve well. (see also)
This does not seem to be related to Caspers project, however.
