Dave

no actually, here I need a 'reshape'-type function to make an O-D style object long

will use `column_positions()` as per - https://github.com/dcooley/geometries/issues/13

> but by far the messiest stuff in silicate is the de-duplication So true. I think a "unique_coordinates()" would be a really useful feature. But I haven't found an easy...

> (Trust you don't mind me using this issue for reams of examples, it's helpful for me to air it in this way). Exactly what I need too.

> with all the index juggling that's grown organically If it helps, I'm working on an update to `decido::Earcut` .hpp so it returns coordinates directly, not indices, and will return...

> but what about the two table approach as a core? One table is just geometry (xy or xyz etc) the other the run length of each linestring Is that...

then I think now my understanding is finally where yours was at a few years ago! This is also similar to the discussions around Arrow and GeoPandas. The big difference...

Leaving this code here while I experiment ```r library(Rcpp) cppFunction( depends = c("decido", "sfheaders") , include = c( '#include "decido/decido.hpp"' , '#include "sfheaders/df/sf.hpp"' ) , code = ' SEXP tris(...

still experimenting - this time without `sfheaders` and using `geometries` in stead. Performance is the same (as expected; the code is almost identical). ```r library(Rcpp) cppFunction( depends = c("decido", "geometries")...

I don't think I can confirm one way or another what the impact will be if I'm honest, as I don't know the differences between the implementations. I can't imagine...