GH-actions for macOS (#1309):

[ateucher]

• add Rcpp drat repo to work around Many packages are broken on Mac with CRAN version of Rcpp -- unless 10.11 SDK is used as on CRAN RcppCore/Rcpp#1060
• add configure.args for sf and rgdal (hopefully only necessary temporarily: MacOS Install with remotes fails looking for libproj #1312)
• don't build vignettes on macOS

I'm not sure this gets it 100%, but we can see how far it gets and I should be able to debug from there...

@etiennebr

[ateucher]

@etiennebr I'm not sure if you have admin access here, but for some reason the actions aren't running for me (only on my first two commits). Any way you can trigger it manually?

[ateucher]

Actually scratch that, I can just watch on my fork

[ateucher]

@etiennebr @edzer the check on macOS R-release (https://github.com/r-spatial/sf/runs/546256574?check_suite_focus=true) is now running and producing similar errors to those on CRAN: https://www.r-project.org/nosvn/R.check/r-release-osx-x86_64/sf-00check.html.

[edzer]

The error is not the same; afaict it is pkg tmaptools that is missing, and needed for vignette 5.

[ateucher]

Sorry @edzer, I meant this error in running examples (which I also reproduce locally):

* checking examples ... ERROR
Running examples in 'sf-Ex.R' failed
The error most likely occurred in:

> ### Name: st_crs
> ### Title: Retrieve coordinate reference system from object
> ### Aliases: st_crs st_crs.sf st_crs.numeric st_crs.character st_crs.sfc
> ###   st_crs.bbox st_crs.CRS st_crs.crs st_crs<- st_crs<-.sf st_crs<-.sfc
> ###   st_set_crs NA_crs_ is.na.crs $.crs format.crs st_axis_order
> ### Keywords: datasets
> 
> ### ** Examples
> 
> sfc = st_sfc(st_point(c(0,0)), st_point(c(1,1)))
> sf = st_sf(a = 1:2, geom = sfc)
> st_crs(sf) = 4326
> st_geometry(sf)
Geometry set for 2 features 
geometry type:  POINT
dimension:      XY
bbox:           xmin: 0 ymin: 0 xmax: 1 ymax: 1
CRS:            EPSG:4326
POINT (0 0)
POINT (1 1)
> sfc = st_sfc(st_point(c(0,0)), st_point(c(1,1)))
> st_crs(sfc) = 4326
> sfc
Geometry set for 2 features 
geometry type:  POINT
dimension:      XY
bbox:           xmin: 0 ymin: 0 xmax: 1 ymax: 1
CRS:            EPSG:4326
POINT (0 0)
POINT (1 1)
> sfc = st_sfc(st_point(c(0,0)), st_point(c(1,1)))
> library(dplyr)

Attaching package: ‘dplyr’

The following objects are masked from ‘package:stats’:

    filter, lag

The following objects are masked from ‘package:base’:

    intersect, setdiff, setequal, union

> x = sfc %>% st_set_crs(4326) %>% st_transform(3857)
> x
Geometry set for 2 features 
geometry type:  POINT
dimension:      XY
bbox:           xmin: 0 ymin: 0 xmax: 111319.5 ymax: 111325.1
CRS:            EPSG:3857
POINT (0 0)
POINT (111319.5 111325.1)
> st_crs("EPSG:3857")$input
[1] "EPSG:3857"
> st_crs("+init=epsg:3857")$proj4string
proj_create: init=epsg:/init=IGNF: syntax not supported in non-PROJ4 emulation mode
[1] "+proj=merc +a=6378137 +b=6378137 +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0 +k=1.0 +units=m +nadgrids=@null +wktext +no_defs"
> st_crs("+init=epsg:3857 +units=m")$b     # numeric
proj_create: init=epsg:/init=IGNF: syntax not supported in non-PROJ4 emulation mode
Error in st_crs.character("+init=epsg:3857 +units=m") : 
  invalid crs: +init=epsg:3857 +units=m
Calls: st_crs -> st_crs.character
Execution halted

That other one is spurious I think.

[ateucher]

Ok I have the checks on macOS (r-release and r-devel) down to just producing the error (failed example) that is the same as the error on the CRAN checks. @etiennebr has a simple fix (here), which I could incorporate into this PR if you like?

[edzer]

Yes, that would be great!

[etiennebr]

Good job @ateucher! 🎉

@edzer, if you want to perform a single review I could merge this branch in 'gh-actions', but I don't see any problem in merging this branch right away since it keeps all travis, appveyor and codecov checks, but adds linux and macOS checks.

@ateucher, are there any lessons we could use to improve the installation instructions for macOS?

[ateucher]

Whew that was a messy one. I definitely recommend doing a "squash and merge"! I also think this could be merged into master.

@etiennebr yes definitely, I plan on making a PR to the README to update the macOS install instructions. The Rcpp bug has thrown an extra wrench in the works, but hopefully that won't take too long to get to CRAN. The configure.args etc on its own is relatively straightforward...

[edzer]

If someone could help to squash this before I merge, the would be great!

[ateucher]

@edzer I usually use the "squash and merge" option in the dropdown in the green merge button. Then you can edit the commit message (I'd probably make it something like):

Add GitHub actions for macOS
* split actions into two files - one for testing with Postgres and one for macOS/Windows
* fix example that was causing errors

Or I can give it a go locally, then force-push, whichever you'd prefer!

[ateucher]

Actually this history is a mess. I'm going to create a new branch and PR...

[ateucher]

Closing in favour of #1330