[R-pkg-devel] Calls to exported dplyr functions from DBI/dplyr backend

[Duncan Murdoch]

On 08/01/2016 8:20 PM, Imanuel Costigan wrote:
> Hi
>
> I am looking to release an updated version of RSQLServer and noticed that when checking the package, I was getting a large number of errors of the style:
>
> * checking R code for possible problems ... NOTE
> auto_copy: no visible global function definition for ‘same_src’
> b

> I have dplyr in Description > Imports and call dplyr functions such as `same_src` without them being imported into my package’s NAMESPACE (or making qualified calls using `::`, the latter being my usual preference). My thinking is that my package’s (dplyr) backend functionality should fail if dplyr itself is not attached on the search path (by for e.g. calling `library(dplyr)`). I think the latter is the behaviour that Hadley has advocated elsewhere (I can’t find reference in places I would expect).

If it makes no sense to use your package without dplyr on the search 
path, you should have

Depends:  dplyr

in your DESCRIPTION file rather than Imports:  dplyr.  You should 
explain this choice when you submit to CRAN, because using this means 
that calling library(RSQLServer) will add dplyr to the search path if 
it's not there.  That could mess up something else that a user is doing, 
so should normally be avoided.  It can also cause your functions to 
fail, if a user has already got dplyr on the search path, as well as 
another package that masks some of its functions.  Your package will use 
the first copies it sees (unless you use :: even in this case).

If your package has functions that users might want to use on their own, 
then you should import the dplyr functions that you need, or call them 
with ::.

It doesn't sound like this applies to you, but another possibility is 
that most of your package doesn't need dplyr.  Then you might use

Suggests:  dplyr

and check for it (using "if (requireNamespace('dplyr'))" ) before each 
use with ::.

Duncan Murdoch

> I am looking for guidance on what to do in this circumstance. I see at least two possibilities:

> * I import all these dplyr functions into my package (or call with ::) and make CRAN errors go away
> * I stick with behaviour described above and risk wasting CRAN maintainers' time and have package rejected.