[R-pkg-devel] What's recommended to do on "Suggests or Enhances not in mainstream repositories: ..."?

On Sat, May 21, 2016 at 8:22 AM, Marius Hofert
<marius.hofert at uwaterloo.ca> wrote:

Hi,

I have (co-)developed two packages: 'qrmtools' and 'qrmdata'. The
latter contains data sets and is currently only available on R-Forge
(as it is larger than 5MB and thus not accepted for CRAN [although
there are much larger packages, but that's a different story...]).
'qrmtools' contains mainly tools that show what can be done with data
such as the one contained in 'qrmdata' -- and thus has 'qrmdata' in
Suggests. The check shows 'Suggests or Enhances not in mainstream
repositories: qrmdata'; I also added 'Additional_repositories:
http://r-forge.r-project.org/' to DESCRIPTION.... but of course
winbuilder reports errors as it can't find 'qrmdata'.

First, there are plenty of CRAN package using:

Additional_repositories: http://r-forge.r-project.org/

is used, cf. https://github.com/search?q=org%3Acran+Additional_repositories%3A+http%3A%2F%2Fr-forge.r-project.org&type=Code

so this should be ok. I'm using it too.

I'm wondering about what I am supposed to do: Remove everything about
'qrmdata' from 'qrmtools'? Clearly, the easiest solution, but that's a
bit dull as the 'qrmtools' can be nicely used to deal with the data
from 'qrmdata'.

It sounds that one of your vignettes, examples or package tests
requires the qrmdata package.  It is ok to use:

Suggests: qrmdata

but then you should make sure it is truly optional throughout, i.e. use

if (require("qrmdata")) {
  ## Use qrmdata functions here
}

in your tests, examples etc.

This is common practice. See also Writing R Extensions.

Maybe there's something else to it?  It would help clarify your
problem if we could see the R CMD check issues and the code (online
repository?).

My $.02

Henrik

How 'severe' is this issue for vignettes? I guess since they are
checked as well, one would run into the same problem. But then a
vignette even more so should show/replicate the functionality of a
paper, say, and there one would definitely like to work with the
data...

The only partially related thread I found was
https://stat.ethz.ch/pipermail/r-package-devel/2015q2/000002.html

In any case, I continue to pray to the R gods in the hope that the 5MB
limit will change for data-related packages (which are updated very
rarely) at some point in the future... Some 'adaptive limits' would be
nice: The less often you submit, the larger the data package can be
(for example), up to a certain larger limit more suitable for
data-related packages.

Any ideas on (all of) that?

Cheers,
Marius