[R-pkg-devel] Possible R CMD check extensions (Was: Possibly mis-spelled words in DESCRIPTION)

[Uwe Ligges]

On 08.05.2016 17:44, Hadley Wickham wrote:
> On Sun, May 8, 2016 at 10:15 AM, Duncan Murdoch
> <murdoch.duncan at gmail.com> wrote:
>> On 08/05/2016 10:47 AM, Dirk Eddelbuettel wrote:
>>>
>>>
>>> On 8 May 2016 at 16:18, Uwe Ligges wrote:
>>> | On 08.05.2016 16:13, carlos cinelli wrote:
>>> | > How should I proceed in this case?
>>> |
>>> | Submit to CRAN.
>>>
>>> The deeper question is if 'we all' can have a conversation about extending
>>> the directory layout / format to add things to the packaging
>>> infrastructure.
>>> What comes to mind is e.g.
>>>
>>>    - a spell-checker white list (per Carlos' initial email)
>>>
>>>    - more generally, 'white list' of warnings R CMD check can be quiet
>>> about [1]
>>>
>>>    - more hooks, ie I would like to call roxygen2::roxygenize() as well as
>>>      Rcpp::compileAttributes() when building
>>>
>>>    - [ This place intentionally left blank. Let me hear other proposals. ]
>>>
>>> We have a year to mess things up for R 3.4.0, so anybody else intereted in
>>> working on this?
>>>
>>> Dirk
>>>
>>> [1] There is no point in telling me that Rcpp creates a shared library of
>>> over 1 mb. It has been doing so for years.  C++ libraries have a
>>> footprint.
>>>
>>
>> This gets tricky.  You don't want to be told that your shared lib is over 1
>> MB, but I'd imagine you'd like to know if it grew over 1000 MB. So really
>> you'd like to tune that note rather than suppress it, just like Carlos would
>> like a white list for the spell checker rather than suppressing the spell
>> check entirely.
>>
>> And CRAN has accepted Rcpp even though it generates that note, so it
>> probably wouldn't object if you set the limit high enough to suppress the
>> note currently, but it would object if you tried to suppress the note
>> entirely (because they don't want 1000 MB shared libs either, and don't want
>> to have to check your package manually).
>>
>> So this would require work on CRAN's side to recognize when your requested
>> suppressions are acceptable.  And I'm sure there would be arguments back and
>> forth, which also suck up CRAN's time and energy.
>>
>> The technical implementation is much easier than the procedural one.
>
> Agreed, but currently CRAN has an informal whitelist in their brains
> which they have to remember when you resubmit a package.  It is
> possible that we could come up with a system that makes life easier
> for CRAN because they could see the whitelist in a standard format and
> easily diff with the last submission.
>
> This would also make life easier for people like me and Dirk who have
> to run R CMD check on 1000's of downstream dependencies. We have no
> idea whether or not we're seeing check issues that are expected or
> need to be acted upon.


Indeed, such details could be improved.

For spell checking (the original question) I do not care too much, 
because you frequently get false positives and CRAN decides by looking 
on the list. And we do not want to force users to maintain an extra 
list. But perhaps we can discuss whether we add a feature to R that a 
list could be provided within the package.

For Notes about the size: Of course, the package itself cannot say that 
it is OK to include huge components, otherwise all maintainers would 
simply declare that it is OK. So that would need to be maintained on 
CRAN. Not sure it makes sense to do that for the few exceptions we allow.

And as you see, then we need to think about exceptions for all sorts of 
Notes and how to include that both technically as well as procedural 
into the CRAN process.

Anyway, not all CRAN team members are listining here, so perhaps start 
to discuss this with R-core (where the current CRAN team is a subset of).

Best,
Uwe