[Rcpp-devel] using an unloaded package's lazy-load data in Rcpp function

Jack Wasey jack at jackwasey.com
Mon Apr 16 21:48:42 CEST 2018


Thanks for the reply. The solution of requiring the namespace from within Rcpp hadn't occurred to me.

I had used Environment::namespace_env("pkg_name") before, but I think the problem here is that lazy-loaded (and maybe other elements we expect in R) are not available if this is done from Rcpp. E.g.

# (using R --vanilla)
Rcpp::cppFunction('Rcpp::Environment getUnloaded2() { return Rcpp::Environment::namespace_env("nycflights13"); }')
ls(envir = getUnloaded2())
# empty, whereas the following:

nycflights13::flights
# pulls the data, with the package not being loaded or attached.

To close this thread, I think what the latter command does differently to Rcpp is to load the package, because sessionInfo() after running the second command, includes nycflights13 in the section "loaded via a namespace (and not attached):" This is not the case after pulling the environment into an Rcpp variable.

Thanks again for steering me,
Jack



On 04/15/2018 10:18 PM, Kevin Ushey wrote:
> The environments called 'package:foo' are associated with attached packages; that is, they're located on the search path. You likely want to find the actual package namespace.
> 
> I think we have a static function Environment::namespace_env() for this. Or you can just call back to R with requireNamespace().
> 
> Kevin
> 
> On Fri, Apr 13, 2018 at 10:10 PM, Wasey, Jack O <jack at jackwasey.com <mailto:jack at jackwasey.com>> wrote:
> 
>     Dear all,
> 
>     It seems I can reference lazy-loaded data from an unloaded, unattached package from R, but not using the equivalent Rcpp as described in the quickref document. I checked out: http://dirk.eddelbuettel.com/code/rcpp/html/Environment_8h_source.html <http://dirk.eddelbuettel.com/code/rcpp/html/Environment_8h_source.html> and even the scant references to environments in the C API section of "Writing R Extensions" and doesn't see what I'm missing. I suspect this is a limitation of R's C interface, not necessarily Rcpp itself.
> 
>     I use the package nycflights13 as an example. Running `sessionInfo()` shows the package nycflights13 is not loaded, then I run:
>     R --vanilla
>     > a <- nycflights13::flights
> 
>     Which completes without error. In Rcpp, I cannot even get a handle on the package namespace if it is not loaded.
> 
>     Rcpp::cppFunction('Rcpp::Environment getUnloaded() { Rcpp::Environment env("package:nycflights13"); return env;}')
>     getUnloaded()
>     Error in getUnloaded() :
>       Cannot convert object to an environment: [type=character; target=ENVSXP].
> 
>     library(nycflights13)
>     Rcpp::cppFunction('Rcpp::Environment getUnloaded() { Rcpp::Environment env("package:nycflights13"); return env;}')
>     getUnloaded()
>     <environment: package:nycflights13>
>     attr(,"name")
>     [1] "package:nycflights13"
>     attr(,"path")
>     [1] "/Library/Frameworks/R.framework/Versions/3.4/Resources/library/nycflights13"
>     >
> 
>     I also found that simple using `loadNamespace` was insufficient for the Rcpp approach to work, but after `library` or `attachNamespace` the above function does work.
> 
>     Am I treading in another rare code path? All I want to do is use a lazy-loaded data set in an Rcpp function when the package may not be loaded when called. Should I just be passing it as a parameter to the function? This seems ugly.
> 
>     Using Rcpp 0.12.16, R 3.4.3 on Mac.
> 
>     Any advice appreciated. Thanks,
> 
>     Jack
> 
> 
>     _______________________________________________
>     Rcpp-devel mailing list
>     Rcpp-devel at lists.r-forge.r-project.org <mailto:Rcpp-devel at lists.r-forge.r-project.org>
>     https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel <https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel>
> 
> 


More information about the Rcpp-devel mailing list