[R] R CMD check barfs at 'suggested' package

[Prof Brian Ripley]

R never reloads a package if not asked to, and it is usually users (not R) 
who `get confused'.

Specifying Suggests: multtest just tells 'check' that it needs to be 
available: it does not tell R to load it.  (It also tells install.packages 
enough to install it if you asked for dependencies to be installed.)

This appears to stem from an error in multtest.  The example-checking 
tests (and not R generically) do try to reset the environment after each 
example, and that includes detaching packages that the example caused to 
be loaded.  Detaching a package does not unload its DLL, but a package's 
.Last.lib can do so.  However, multtest does so incorrectly, via a call to 
dyn.unload not library.dynam.unload.

Namespaces will not help as multtest does not have one (at least in the 
version I looked at).

I suggest R-devel would be a better list for questions about developing 
packages.  (But for now please take this up with multtest maintainer.)


On Wed, 1 Feb 2006, Alexander Ploner wrote:

> I have been running R CMD check for the first time on of our own
> packages, which otherwise builds, installs and runs just fine. The
> package depends on package akima and suggests package multtest;
> suggesting multtest is indicated because a) it loads a lot of other
> Bioconductor packages, very slowly, and b) it is only needed for one
> specific subroutine, which executes require(multtest) every time it's
> called.
>
> Now, when I run R CMD check on our package, it terminates when
> checking the examples in the documentation, claiming that
>
> Error in .C("get_stat", as.double(tmp$X), as.integer(tmp$m),
> as.integer(tmp$n),  :
>         "C" function name not in DLL for package 'multtest'
> Execution halted
>
> A closer look at  the output in the -Ex.Rout file shows that this
> does indeed happen when calling the function mt.teststat in multtest,
> to which this piece of code belongs. It also shows however, that this
> only happens in the *second* chunk of example code containing the
> calling function; before that, three calls to the offending function
> in the first chunk (ie in a different .Rd file) are executed just
> fine. Furthermore, if I change the status of multtest from Suggests
> to Depends in the DESCRIPTION file, R CMD check processes our package
> without problem.
>
> My impression is that R gets confused when re-loading multtest for
> the second chunk of example code (for what it's worth, the
> offending .C("get_stat"...) does have a PACKAGE argument).
>
> My question now: Is there any way of avoiding this problem without
> loading multtest by default? Would e.g. using a namespace help? If
> so, how specifically? Etc.?
>
> Thanks for your time (and expertise),
>
> alexander
>
>
> platform powerpc-apple-darwin7.9.0
> arch     powerpc
> os       darwin7.9.0
> system   powerpc, darwin7.9.0
> status   Patched
> major    2
> minor    1.0
> year     2005
> month    05
> day      12
> language R
>
>
> Alexander.Ploner at ki.se
> Medical Epidemiology & Biostatistics
> Karolinska Institutet, Stockholm
> Tel: ++46-8-524-82329
> Fax: ++46-8-31 49 75
>
>
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595