vulnerability to PROJ6/GDAL3

[rsbivand]

Running revdeps from sp (sp (my github fork) with development rgdal from R-Forge):

> packageVersion("sp")
[1] ‘1.3.4’
> packageVersion("rgdal")
[1] ‘1.5.3’
> rgdal::rgdal_extSoftVersion()
          GDAL GDAL_with_GEOS           PROJ             sp 
       "3.0.2"         "TRUE"        "6.2.1"        "1.3-2" 

a test fails:

> library(testthat)
> library(spex)
> 
> test_check("spex")
── 1. Failure: extent and crs works (@test-spex-extent.R#5)  ───────────────────
==... isn't true.

══ testthat results  ═══════════════════════════════════════════════════════════
[ OK: 35 | SKIPPED: 0 | WARNINGS: 16 | FAILED: 1 ]
1. Failure: extent and crs works (@test-spex-extent.R#5) 

Error: testthat unit tests failed
Execution halted

See:

http://rgdal.r-forge.r-project.org/articles/PROJ6_GDAL3.html
r-spatial/sf#1231
r-spatial/sf#1187
r-spatial/sf#1146
r-spatial/discuss#28

for background. See:

r-spatial/discuss#28 (comment)

for a way of testing fixes in a docker container contributed by Jakub Nowosad.

[mdsumner]

Thanks Roger.

Small thing, any idea why rgdal::rgdal_extSoftVersion() returns 1.3.2 for sp rather than 1.3.4 as reported by packageVersion? (It doesn't for me)

[mdsumner]

I don't think this has anything to do with new GDAL/PROJ, the projargs was logical NA and must be character. (Strictly spex should hack its own way into the crs/wkt2 here and not rely on raster)

#devtools::install_github("rsbivand/sp")
#devtools::install_github("rforge/rgdal/pkg")
packageVersion("sp") #[1] ‘1.3.4’
packageVersion("rgdal") #[1] ‘1.5.3’
rgdal::rgdal_extSoftVersion()
# GDAL GDAL_with_GEOS           PROJ             sp 
# "3.0.2"        "FALSE"        "6.2.1"        "1.3-4" 

[mdsumner]

I think the failing travis problem is due to new version of stringi (2020-01-09)

[mdsumner]

stringi fixed by not using trusty 94b24b9

[rsbivand]

The underlying problem is that:

raster::projection(spex(raster::extent(lux), raster::projection(lux)))
# [1] "+proj=longlat +ellps=WGS84 +towgs84=0,0,0,0,0,0,0 +no_defs"
raster::projection(lux)
# [1] "+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0"

so are not string-equal. Even rgdal::compare_CRS() - WIP - sees them as different.Unsure what to advise, maybe mute the test until things calm down. rgdal will be submitted to CRAN soon.

[mdsumner]

Eh, thanks sorry to miss that - I'll remove that pointless test

[mdsumner]

presumably was fixed by raster

[mdsumner]

I had thought rgdal was released already

[rsbivand]

I'm moving towards release, still a few revdep packages unresponsive.

[mdsumner]

I think we're good now, thanks!