[R-pkg-devel] Vignettes & _R_CHECK_DEPENDS_ONLY_=true

[Duncan Murdoch]

On 09/06/2022 12:23 p.m., Henrik Bengtsson wrote:
> On Thu, Jun 9, 2022 at 8:53 AM Arne Henningsen
> <arne.henningsen using gmail.com> wrote:
>>
>> Dear all
>>
>> My R package "systemfit" has a 'vignette' that uses the R packages
>> "plm" and "sem", while the actual code of the "systemfit" package does
>> not depend on these two packages. Currently, the "systemfit" package
>> 'Suggests' the "plm" package and the "sem" package. However, CRAN
>> started to check packages with '_R_CHECK_DEPENDS_ONLY_=true' so that
>> the checking of the "systemfit" package returns an error message
>> because the vignette cannot be created without the "plm" and "sem"
>> packages.
> 
> Also, the case for one of your example, cf.
> https://cran.r-project.org/web/checks/check_results_systemfit.html
> 
>> I am considering changing the DESCRIPTION file so that the
>> "systemfit" package will 'Depend' on the "plm" and "sem" packages
>> (although only the vignette depends on these two packages). Is this
>> the recommended way to do or is there a better way to make the
>> "systemfit" package pass the tests at CRAN?
> 
> To keep them as Suggests:ed packages, you need to run the code in your
> examples and vignettes that rely on them conditionally on the being
> installed, e.g.
> 
> if (requireNamespace("plm", quietly = TRUE)) {
>    ...
> }
> 
> FWIW, you're using Sweave, but if you'd used knitr, there are easy
> ways to do this also for vignettes.  I forgot the details, but I saw a
> simple solution the other week somewhere out there on the interwebs.
> 
> If you don't want to do this, then yes, you need to turn those
> dependencies into hard dependencies, e.g. Depends or Imports.

In knitr, you can use the chunk option `eval = requireNamespace("plm", 
quietly = TRUE)`.  I'd put a comment in the code chunk too, saying it 
won't be evaluated if plm is not available, but this is not a 
requirement, just being nice to the reader.

Duncan Murdoch