mdsumner
commented
7 years ago It's now clear that path, object are special cases of groupings. Simple features has three kinds of grouping:
- single, POINT, MULTIPOINT, LINESTRING (point is degenerate)
- double, MULTILINESTRING and POLYGON
- triple, MULTIPOLYGON
ok four:
- variable GEOMETRYCOLLECTION
The feature is always the highest grouping level. I've been missing that the kludgy encoding of holes was hiding the fact that these are already multi-level features to sc. So, counties within states is just another level: holes within POLYGON within MULTIPOLYGON within county within state within country.
Simple features is just a special case applied to the grouping kinds above.
ggplot2 doesn't have multi-levels, but dplyr does. What about ggvis? Probably not, simple features clashes with graphics when there's an extra level like this.
I am concentrating on user-flexibility for the "unjoin" on a single table, a key pattern to split out the geometry space, de duplicate and key to the remaining data. This is the basis of path topology for many data streams and is a dual to many nested table structures that are in common use. (The polygon decomposition is just a specialist input type from this perspective.) Flipping between path- and primitive- models is a key part of a grammar with all the small pieces needed for useful intermediate forms.
The grammars of graphics and analysis in modern R provide a lot of support for restructurings like this, but common pipelines that end in a graphic or other grouped structure don't go the full distance to output a "spatial data" set. The unjoin and its counterpart evert provide keystones for the path/primitive flip.