[R-pkg-devel] sf subsetting with square bracket

[Duncan Murdoch]

On 05/02/2019 3:57 a.m., Berry Boessenkool wrote:
> 
> I was pointed to a similar problem in data.table:
> https://github.com/Rdatatable/data.table/issues/2341
> It turns out it is sufficient to import (any random) single function 
> from sf.
> The "[" method will then be available as well.
> E.g. with the following line included, the sf methods are available 
> without the package being attached first:
> #' @importFrom sf st_sf
> 
> Is this cleaner than requireNamespace("sf") because it only loads a 
> single function?

It loads the entire package, and it does it unconditionally.  (You can't 
*load* just one function; you get the whole namespace loaded if you ask 
for one function.  However, you will have access to just that one 
function from your own functions.)  You would need to list "sf" in the 
Imports: section, and every user of your package would have to have "sf" 
installed.

That would be appropriate if your package can't function without working 
with "sf" objects.  I don't think it's the appropriate solution for 
"osmplotr", because they want to remain small, and the "sf" package is 
big.  They want to be able to operate even if the user chooses not to 
install "sf", and apparently all but one of their functions can do so.

Duncan Murdoch

> 
> Both approaches work fine. Thanks, Duncan, for digging into this!
> 
> Btw: sf people are aware: https://github.com/r-spatial/sf/issues/970
> 
> 
> Kind regards,
> Berry
> 
> 
> ------------------------------------------------------------------------
> *From:* Duncan Murdoch <murdoch.duncan using gmail.com>
> *Sent:* Monday, February 4, 2019 15:11
> *To:* Berry Boessenkool; R package devel
> *Subject:* Re: [R-pkg-devel] sf subsetting with square bracket
> On 03/02/2019 8:26 p.m., Duncan Murdoch wrote:
>> On 03/02/2019 5:21 p.m., Berry Boessenkool wrote:
>>>
>>> Hi,
>>> my package in question can only be online next week, but here's a script
>>> for a MWE:
>>> https://gist.github.com/brry/7728b9b2d35afad7f1fc5978c3315009
>>> The script uses devtools and osmplotr to
>>> - create a package with a bare minimum dummy function
>>> - run the code showing the problem
>>> - create a solution using the ::: import
>>> - clean up, removing the package and the folder for it
>> 
>> Thanks.  I can confirm the issue.  I think the problem is that neither
>> your sample package nor osmplotr declares a dependency on sf, and it is
>> not loaded.  I don't know if it makes sense for your package to do so,
>> but osmplotr probably should, since it is creating "sf" objects.
>> 
>> If I force sf to be loaded by changing your dummy function to
>> 
>> dummy <- function(map,obj)
>> {
>>     requireNamespace("sf")
>>     add_osm_objects(map, obj[1:10,], col='green')
>> }
>> 
>> then things are fine.  In your real package, you'll want to be prepared
>> to handle the case where "sf" is not installed.
>> 
>> sf is a big package, so maybe it's intentional to not require it.  I
>> don't know these packages at all, but it seems to me that if osmplotr is
>> returning objects of class "sf", then it should probably make sure that
>> the "sf" package is loaded so that methods can be found.
> 
> I've reported this on the osmplotr Github page:
> https://github.com/ropensci/osmplotr/issues/46.  It appears that osmdata
> has the same issue, but the maintainers know about it and are working on it.
> 
> In the meantime I'd add the requireNamespace("sf") call to your package
> just to be defensive.
> 
> Duncan Murdoch