[Rd] $new cannot be accessed when running from Rscript and methods package is not loaded

[Simon Urbanek]

On Feb 11, 2014, at 1:28 PM, Martin Morgan <mtmorgan at fhcrc.org> wrote:

> On 02/11/2014 09:06 AM, Winston Chang wrote:
>> To state the issue that Kirill raised in a different way... A package
>> with S4 or reference classes and Depends:methods can throw an error
>> when you do something as simple as this:
>>   Rscript -e "mypackage::foo()"
>> 
>> But this will work:
>>   Rscript -e "library(mypackage); foo()"
>> 
>> This is because when mypackage has Depends:methods, calling
>> mypackage::foo() merely loads mypackage, and doesn't attach it -- and
>> when this happens, the methods package is also loaded and not
>> attached. In this situation, you can get errors of the type that Dirk
>> noted.
>> 
>> In contrast, library(mypackage) causes methods to be attached, and so
>> you won't get those errors. This seems like a problem with the
>> behavior of the :: operator. Why should it be that when a package is
>> loaded, the packages listed in Depends are not attached?

Because the whole point is that you don't want mypackage on the search path, otherwise you wouldn't use :: in the first place. Namespaces are about solving the namespace clutter, not exacerbating it ;).


>> Presumably,
>> when someone lists a package in Depends instead of Imports, it's for a
>> good reason, as is the case here with the methods package.
> 
> To me it's worth distinguishing between the particular problem associated with the methods package, and the more general problem encountered when PkgA fails to import symbols used in PkgB, and instead relies on PkgB symbols to be found on the search path. In the latter it seems better practice to import(PkgB) (or importFrom) in the NAMESPACE and probably Import: PkgB in DESCRIPTION.
> 
>> In the following thread, it's stated that the defensive approach is to
>> use Depends:methods -- but that's still not enough for running the
>> code listed above.
>> http://r.789695.n4.nabble.com/advise-on-Depends-td4678930.html
> 
> A more defensive defensive approach might be to add
> 
>    .onLoad <- function(...) require(methods)
> 

But that will attach methods again - not good.

I suspect the problem could be classified as bug in methods which doesn't behave correctly if not attached. I didn't look into details, but there used to be a technical problem in R C API where you could not pass the right environments to make that work (that bit me when I tried to use S4 a while ago), but I think that has been solved, so conceptually I think this is a bug that could be fixed. This sounds like something that should be possible to fix - even if it means fixing something in R to make it happen. Working around it in hacky ways mentioned here doesn't solve the underlying problem.

Cheers,
Simon


> but this leads to either
> 
>    * checking dependencies in R code ... WARNING
>    'library' or 'require' call not declared from: ‘methods’
>    See the information on DESCRIPTION files in the chapter 'Creating R
>    packages' of the 'Writing R Extensions' manual.
>    ...
>    * checking R code for possible problems ... NOTE
>    File ‘PkgA/R/binding.R’:
>      .onLoad calls:
>        require(methods)
> 
>    Package startup functions should not change the search path.
>    See section ‘Good practice’ in '?.onAttach'.
> 
> with Imports: methods, or
> 
>    * checking dependencies in R code ... NOTE
>    'library' or 'require' call to ‘methods’ which was already attached by Depends.
>      Please remove these calls from your code.
>    See the information on DESCRIPTION files in the chapter 'Creating R
>    packages' of the 'Writing R Extensions' manual.
>    ...
>    * checking R code for possible problems ... NOTE
>    File ‘PkgA/R/binding.R’:
>      .onLoad calls:
>        require(methods)
> 
>    Package startup functions should not change the search path.
>    See section ‘Good practice’ in '?.onAttach'.
> 
> with Depends: methods
> 
> So is there really any WARNING / NOTE-free way of using the methods package?
> 
> Martin
> 
>> 
>> -Winston
>> 
>> 
>> On Mon, Feb 10, 2014 at 8:31 PM, Dirk Eddelbuettel <edd at debian.org> wrote:
>>> 
>>> On 11 February 2014 at 02:53, Kirill Müller wrote:
>>> | Why does it seem to be necessary to load the methods package here?
>>> 
>>> "Just use littler (TM pending)".
>>> 
>>> It (auto-)load methods automagically, thanks to Jeff Horner.  See below.
>>> 
>>>    edd at max:~$ chmod 0755 /tmp/kirill.r
>>>    edd at max:~$ /tmp/kirill.r
>>>    [1] "refObjectGenerator"
>>>    attr(,"package")
>>>    [1] "methods"
>>>    edd at max:~$ cat /tmp/kirill.r
>>>    #!/usr/bin/r
>>> 
>>>    newTest <- function() {
>>>       cl <- get("someClass")
>>>       cl$new
>>>    }
>>> 
>>>    someClass <- setRefClass("someClass")
>>>    print(class(someClass))
>>>    edd at max:~$
>>> 
>>> For Rscript, you still need to load it explicitly like any other packages you
>>> want to use.
>>> 
>>> Dirk
>>> 
>>> PS New littler release pending in a few days or weeks. The Github repo is
>>> current and working.
>>> 
>>> --
>>> Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
>>> 
>>> ______________________________________________
>>> R-devel at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>> 
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>> 
> 
> 
> -- 
> Computational Biology / Fred Hutchinson Cancer Research Center
> 1100 Fairview Ave. N.
> PO Box 19024 Seattle, WA 98109
> 
> Location: Arnold Building M1 B861
> Phone: (206) 667-2793
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>