[R-pkg-devel] Undeclared packages ... in Rd xrefs

[Ulrike Grömping]

Am 02.02.2021 um 22:37 schrieb Duncan Murdoch:
> On 02/02/2021 9:13 a.m., Ulrike Grömping wrote:
>> Am 02.02.2021 um 02:38 schrieb Duncan Murdoch:
>>> On 01/02/2021 5:03 p.m., Ulrike Grömping wrote:
>>>> Dear package developeRs,
>>>>
>>>> under the Fedora clang checks, I find the note
>>>>
>>>> "Undeclared packages ‘FrF2’, ‘DoE.wrapper’, ‘sfsmisc’, ‘DoE.MIParray’,
>>>> ‘planor’ in Rd xrefs"
>>>>
>>>> for my package DoE.base. I understand that package planor has been
>>>> archived from CRAN; I don't understand what is wrong with the other
>>>> xrefs; only the R-Devel Fedora clang flavor seems to complain, and 
>>>> I did
>>>> not find an explanation in the section on cross references in 
>>>> Writing R
>>>> Extensions. Can someone explain the meaning of this note?
>>>
>>> There's a line in the Writing R Extensions manual section 2.5
>>> Cross-refernces for R-devel that says:
>>>
>>> "Packages referred to by these ‘other forms’ should be declared in the
>>> DESCRIPTION file, in the ‘Depends’, ‘Imports’, ‘Suggests’ or
>>> ‘Enhances’ fields."
>>>
>>> The other forms are the forms of links to other packages.  So
>>> presumably you don't mention those packages in your DESCRIPTION file.
>>> Generally that means they should be listed in Suggests, which doesn't
>>> force them to be installed, but they will be installed during tests.
>>> You might also argue they should be in Enhances, though that seems a
>>> worse fit.
>>>
>>> Duncan Murdoch
>>
>>
>> Thank you for the explanation! I will remove the xref to sfsmisc
>> (because it does not make sense to suggest that package). The three
>> other still available packages all depend on or import DoE.base. Thus,
>> if I incorporate them in Suggests, I will create a circular dependence
>> structure. Doesn't that create problems?
>
> It creates small problems, not serious ones. You just need to make 
> sure that a version of your package is available when those packages 
> are installed for testing.  So things need to happen in something like 
> this order on a totally new machine:
>
>  - install the "hard" dependencies for your package.
>  - install your package
>  - install the "soft" dependencies for your package.
>
> I just discovered that the remotes::install_deps(dependencies = TRUE) 
> function can't currently do things in the right order, but 
> remotes::install_local(dependencies = TRUE) is fine.  I didn't check 
> the base function install.packages(dependencies = TRUE).
>
> Duncan Murdoch

Thank you! I don't like that very much, because exactly those users who 
are totally new to using R are extremely put off by this type of 
difficulty. I think that I will try to solve this differently, e.g. by 
showing the code for calling these help files from R without linking to 
them from within the Rd file. I am not sure what the benefits are of 
this demand - but I do see the drawbacks here.

Best,
Ulrike Grömping

-- 
##############################################
## Prof. Ulrike Groemping
## FB II
## Beuth University of Applied Sciences Berlin
##############################################
## prof.beuth-hochschule.de/groemping
## Phone: +49(0)30 4504 5127
## Fax:   +49(0)30 4504 66 5127
## Home office: +49(0)30 394 04 863
##############################################