tagging validity

[imagico]

The documentation here just talks about new and only style MPs in the pure form:

• tags on relation and no tags on outer rings: new style, good
• identical tags on all outer ring ways and no tags on relation: old style, bad

So far so good but it would be important to clarify what kind of tagging combinations that fall into neither of these categories is considered valid and/or desirable. Obviously bad are:

• different tags on different outer ways and no tags on relation (broken old style tagging)
• identical tags on all outer ring ways and relation (both old and new style together)

But there are also variants that are generally considered correct

• tags on relation and different tags that can apply to linear ways on some or all outer ways
• tags on relation and different tags on closed ways that are outer rings: two different areas with different tags

For mappers it would be important to have a clear guideline what kind of constructs in tagging are OK and which not.

[joto]

This all falls under one simple rule: Put the tags where they belong and then everything makes sense. All those combinations you mention are logical once you only have the new tagging style. Tags on the relation are for the multipolygon, tags on the member ways are for those member ways only. I don't really know how to better describe this, but welcome any contribution to better wording here, in the wiki, or anywhere else.

[tyrasd]

identical tags on all outer ring ways and no tags on relation: old style, bad

To be precise, the wiki never actually allowed the kind of multipolygon tagging with no tags on the relation and more than one outer way. Initially multipolygons only allowed one outer ring where the tags are put on that one (for backwards compatibility reasons, I guess). The schema was later expanded to also allow advanced multipolygons (with the possibility of having multiple outer ways, but tags are now required to be on the relation). At that point the "old" style tagging was still kept (again for backwards compatibility reasons, I presume):

If you have one closed way making up the outer ring and it does not describe something in its own right, you may also put these tags on the outer ring and leave the relation untagged. If you have more than one outer way (see "Advanced Multipolygons" below), then this does not make sense. Therefore it is suggested (for consistency) to always put the multipolygon tags on the relation.

(The first part of this paragraph has only very recently been marked as deprecated in the wiki.)

So, it would make sense to also consider the following cases differently:

• identical tags on a single closed outer way and no tags on relation: old style, deprecated
• more than one outer way and no tags on relation: broken old style tagging

[imagico]

The problem i see is that mappers get conflicting feedback here. On the one hand in the long term we should eliminate old style polygons and then everything is clear in terms of tags and you can freely use tags on member ways as they apply to the individual ways.

On the other hand right now most applications allow old style MPs in some form and then the use of tags on member ways is complicated, even more because different applications interpret them differently. The only thing you can reliably recommend to mappers is in fact to avoid tagging ways that are members of MP relations at all. This is of course often fairly awkward in reality.

The approach taken at openstreetmap-carto recently - see https://github.com/gravitystorm/openstreetmap-carto/pull/2527 - is to treat any MP relation with any tag other than type as a new style MP and therefore ignore any tags on the ways for MP interpretation. This seems a fairly clear rule that is easy to communicate and you can tell mappers: as soon as you tag the MP relation you are in the new style domain and any tags on ways only apply to the ways themselves.