a shared plan

[mdsumner]

Hi there, I'm very happy and impressed that you're single-handedly taking on all the various parts we need to build a decent spatial toolkit! But, did you see PROJ?

https://github.com/hypertidy/PROJ/

I think we need a bit of a map of who's doing what and where and why?

My approach has been to wrap this with reproj, which is just for matrix/dataframe input - but also includes a method for "sc" classes, which is just my models in silicate (they store vertices on a dataframe in a standard way, so there's no package-specific wrangling for the format). This has been working great for me, I use proj4 if there's no PROJ-lib >= 6, and PROJ if it's 6 or 7. (proj4 works fine with 6 and 7 btw).

I aligned with rwinlib/proj to get the windows binaries, and that's only at 6.1.0 but I'd assume to get rwinlib to release 7.0.0 rather than bundle the source into a new package.

Anyhoo, very keen to work together and maybe we need more stuff written down about where we are going. My approach to some other components is below (not a full list by any means):

• GDAL - vapour package
• distances (Karney etc) - geodist package
• get in-memory CRS - crsmeta package
• constrained triangulation decido package (or anglr, laridae for the more hard core meshing)

There's lots of others, by lots of other folks - but yours and @dcooley's are the ones I'm most impressed by, for building a spatial landscape of tools.

I find a fair lack of discussion on this topic, but I've been talking along these lines with @mpadge for a few years - and I'm very heartened by the increase in activity coming along.

Also, I've tried to be careful to keep deps down but the tidyverse was instrumental in making a lot of things work that I needed, so dplyr is often there - when there's a hard-dep I didn't want, I would simply make another package for the next feature - that's why anglr lets every dep in , just so I could get what I wanted working - but silicate as the basis was free of raster, RTriangle, etc. I'm very affected by @dcooley's take on headers, and I'l be trying to learn to take that approach as much as possible.

[paleolimbot]

Don't take this too seriously! To get it to pass CMD check I needed a Title Case Title...it takes a lot of experimenting to know whether this could do something different than PROJ, proj4, sf, etc., and trying to get this to compile was mostly trying to understand PROJ. I think the whole "geocrs" thing would be more useful, but its very close to crsmeta so maybe that's a better home for it.

I plan on finishing the first go at geovctrs and poking away at "geogeos" (or whatever "geom" gets renamed to)...geovctrs is mostly providing S3 generics (as_geovctr() and restore_geovctr()), and "geogeos" (or whatever) will hopefully provide enough functionality to convince maintainers to use them. The whole point of a good modular system is that it spreads out the maintainer load, and I'm hoping to keep my part to those two bits.

I too like the header library approach...still learning though!