Finish implementing vctrs methods for sfc vectors

[lionel-]

• Restore n_empty, crs, and precision attributes.

• Add common type implementation for sfc objects. This finds the common class based on heuristics found in c.sfc().

• Added a new vctrs help topic, so that argument names can be documented separately from the tidyverse ones.

[lionel-]

Should we move the vctrs methods into their own file vctrs.R or perhaps tidyverse-vctrs.R?

I'm now looking into other data types, it might make the source more easily browsable to keep the vctrs methods separate as we might end up with a non-trivial amount of those.

[edzer]

Yes, good idea!

[lionel-]

hmm, last Travis build failed when it shouldn't have:

  Error: processing vignette 'sf5.Rmd' failed with diagnostics:

   package or namespace load failed for 'tidyr':

    object 'cnd_bullets' not found whilst loading namespace 'vctrs'

This is because dev rlang and dev vctrs are temporarily in incompatible state.

Is it intended that Travis checks against dev rlang and dev vctrs? I was assuming we'd be testing against CRAN versions.

[yutannihilation]

It seems because some r_github_packages is used by default. I guess they were necessary for development at the time when they were added. Are they still needed now?

sf/.travis.yml

Lines 19 to 24 in fe70fea

	r_github_packages:
	- r-dbi/DBI
	- r-dbi/RPostgres
	- r-lib/covr
	- r-spatial/lwgeom
	- r-spatial/stars

[lionel-]

Following up on #1172 (comment), the CRS and precision attributes are now combined. I took the conservative approach of throwing an error if they are inconsistent. This can be changed to something more appropriate by modifying common_crs() and common_prec().

[edzer]

Looks good -- thanks so much!