sp/rgdal workflows with PROJ >= 6 and GDAL >= 3

Because PROJ >= 6 and GDAL >= 3 change the way that PROJ strings 
(representations of coordinate reference systems) are handled, steps are 
being taken to find ways to adapt sp/rgdal workflows. A current proposal is 
to store the WKT2_2018 string as a comment to CRS objects as defined in the 
sp package.

A draft development-in-progress version of rgdal is available at 
https://r-forge.r-project.org/R/?group_id=884, and for sp at 
https://github.com/rsbivand/sp (this version of sp requires rgdal >= 1.5-1). 
This adds the WKT comments to CRS objects on reading vector and raster data 
sources, and uses WKT comments if found when writing vector and raster 
objects (or at least does as far as I've checked, possibly fragile).

An RFC with tersely worked cases for using CRS object comments to carry WKT 
strings but maintaining full backward compatibility is online at 
http://rgdal.r-forge.r-project.org/articles/PROJ6_GDAL3.html.

If you have other ideas or concerns about trying to use this mechanism for sp 
CRS objects, please contribute at your earliest convenience.

http://rgdal.r-forge.r-project.org/reference/list_coordOps.html shows the 
beginning of the next step, to query transformation operations to find viable 
coordinate operation pipelines.

I'm assuming that the previous behaviour (transform without considering 
accuracy with whatever is to hand) is not viable going forward, and that we 
will need two steps: list coordinate operations between source and target CRS 
(using the WKT comments as better specifications than the PROJ strings), 
possibly intervene manually to install missing grids, then undertake the 
coordinate operation.

The fallback may be simply to choose the least inaccurate available 
coordinate operation, but this should be a fallback. This means that all uses 
of spTransform() will require intervention.

Is this OK (it is tiresome but modernises workflows once), or is it not OK 
(no user intervention is crucial)?

These behaviours may be set in an option, so that package maintainers and 
users may delay modernisation, but all are undoubtedly served by rapid 
adaptation (GRASS 7.8.1 released yesterday, libspatialite, pyproj, QGIS 
development versions all state that they list candidate coordinate 
operations).

We cannot ship all the grids, they are very bulky, and probably nobody needs 
sub-metre accuracy world-wide. Work in PROJ is starting to create a content 
delivery network for trusted download and mechanisms for registering 
downloaded grids on user platforms. We would for example not want Windows 
users of rgdal and sf to have to download the same grid twice.

Comments welcome here and at https://github.com/r-spatial/discuss/issues/28 
or https://github.com/r-spatial/sf/issues/1187

Roger