Skip to content

[R-pkg-devel] recreating CRAN Testing environment

7 messages · Jonathan Callahan, Ben Bolker, Avraham Adler +3 more

#
I am (hopefully) at the end of a series of back-and-forth submissions of a
package to the CRAN upload site.

I have done my best to test things by:

   - using devtools
   - running "R CMD check --as-cran" on the command line
   - checking my package on OSX, win-builder.R and various 32 and 64 bit
   linux virtual machines at DigitalOcean

But I have never managed to recreate all of the notes and warning messages
that the CRAN testing comes up with.

In an effort to save time and effort on both ends, can anyone advise on
what I need to do to generate ALL the warning messages that the CRAN
testing will find?
#
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 15-07-20 01:50 PM, Jonathan Callahan wrote:
The most important thing that you haven't explicitly mentioned is
to test on the most recent development version of R (check out via
Subversion/configure/make/make install from scratch).  Can you adjust
your DigitalOcean settings accordingly?

  Ben Bolker


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iQEcBAEBAgAGBQJVrTcSAAoJEOCV5YRblxUHNqUIAKMpuDZ5viQSBlyXpIqjtjEW
Na6LhC5O/XcjmmrZHDftXAat2gTrRM0OJAn0cfzSZttdcJKowbMYfJzDWe/5oBCE
QRQ0zwxbJY1cFHFyYR6r1zCe4ah4SzoGt+FmBYjtAKefvjm12tfQubhQlxPxZ1kV
XFIfYTWHmHWNGlhZuokPE/GtdXK5/4MNF/32HxseNBRzc2SWqP3i6VKPnbN7LFgl
xrjXsHQdqU3UaY37D44mInl+qm5y5reRSUi6NO0SrSh2WcVWTnqZ3HsUkFPND5tM
NyzrfhtGN2wISsFq1J+dC87ajyeh6D8LTYBiokUnWQFOj8tPi1Zhff0LuTJi7Vc=
=gytX
-----END PGP SIGNATURE-----
#
Even if you don't use Windows, it pays to submit your package to the
Winbuilder test of buth R-release and R-devel. The error reports, if any,
are often not-OS dependant and will serve as a good gaage as to how the
real CRAN will respond.

Avi
On Mon, Jul 20, 2015 at 1:59 PM, Ben Bolker <bbolker at gmail.com> wrote:

            

  
  
#
Jonathan,
On 20 July 2015 at 13:59, Ben Bolker wrote:
| On 15-07-20 01:50 PM, Jonathan Callahan wrote:
| > In an effort to save time and effort on both ends, can anyone
| > advise on what I need to do to generate ALL the warning messages
| > that the CRAN testing will find?
| 
|    The most important thing that you haven't explicitly mentioned is
| to test on the most recent development version of R (check out via
| Subversion/configure/make/make install from scratch).  Can you adjust
| your DigitalOcean settings accordingly?

Seconded.

Recipes how to build R-devel from source have circulated for years. [1] "As
it works for me" I still keep a svn checkout which I update maybe once a week
(or whenever I become aware of a new issue) to rebuild.

For those familiar with Docker, Carl and I also provide both a small image for
Rocker ('drd' for daily r-devel) and a larger one ('r-devel', still contains
development libraies). [2]  Use of those should be straightforward on eg
digital ocean.

For what it is worth, I had about five different packages go onto CRAN in the
last week. None required a second iteration -- but testing with R-devel is a
key part of the process.

Dirk


[1] Here is one which Google just found for me. I may have posted this
    earlier too:
    https://stat.ethz.ch/pipermail/r-sig-debian/2012-August/001935.html

[2] https://github.com/rocker-org/rocker on GitHub and
    https://registry.hub.docker.com/repos/rocker/ at Docker
#
On Mon, Jul 20, 2015 at 1:59 PM, Ben Bolker <bbolker at gmail.com> wrote:
[...]
If you want an easy way to build and check your package with R-devel,
and your package is on GitHub, consider using
https://github.com/metacran/r-builder

By default it builds and checks your package with R-oldrel, R-release
and R-devel, on Travis CI.

win-builder.r-project.org is another possibility, somewhat complementary.

Disclaimer: r-builder is my project, and I have no financial interests in it.

Best,
Gabor
[...]
#
Thanks to all so far for excellent suggestions!

On Mon, Jul 20, 2015 at 11:18 AM, G?bor Cs?rdi <csardi.gabor at gmail.com>
wrote:

  
    
#
On 20.07.2015 19:50, Jonathan Callahan wrote:
Not needed at all.
Perfect.

But the R version should be a recent version of R-devel.

Best,
Uwe Ligges