Native calls fail after load_all ("not resolved from current namespace")

[mb706]

devtools::load_all unloads previously loaded DLLs and therefore reorders the LoadedDLL array, but the getCallingDLLe result that R relies upon when looking up DLL objects does not change in the same way. This sometimes leads to not resolved from current namespace errors when an unlucky package calls a .C function.

In the following example, I have cloned the ParamHelpers package to a subdirectory, and I have the lhs package installed in my library. R needs to be run with the --vanilla option to prevent interference from a possible .RData file.

$ R --vanilla
> devtools::load_all("ParamHelpers")
Loading ParamHelpers
Re-compiling ParamHelpers
'/usr/lib64/R/bin/R' CMD INSTALL '/home/user/ParamHelpers' --library='/tmp/RtmproDXbZ/devtools_install_588768c7107c' --no-R --no-data --no-help --no-demo --no-inst --no-docs --no-exec --no-multiarch --no-test-load
make: Nothing to be done for 'all'.
> lhs::maximinLHS(n=2, k=2)  # implicitly load the lhs DLL; works fine here.
          [,1]      [,2]
[1,] 0.6008410 0.9491948
[2,] 0.4723376 0.3303989
> devtools::load_all("ParamHelpers")
Loading ParamHelpers
Re-compiling ParamHelpers
'/usr/lib64/R/bin/R' CMD INSTALL '/home/user/master/ParamHelpers' --library='/tmp/RtmproDXbZ/devtools_install_58874478c51e' --no-R --no-data --no-help --no-demo --no-inst --no-docs --no-exec --no-multiarch --no-test-load
make: Nothing to be done for 'all'.
> lhs::maximinLHS(n=2, k=2)  # does not work                                                                                                                                                                                                               
Error in .C("maximinLHS_C", as.integer(n), as.integer(k), as.integer(dup),  : 
  "maximinLHS_C" not resolved from current namespace (lhs)

P.S.

• One could also argue this is R's fault, and dyn.unload should take care of this. This looks like the same issue and was filed 1.5 years ago at the R project's bugzilla. Someone with an account could add a back reference to here.
• This was also an issue here, see my comment.
• A possible workaround on the side of the package ('lhs' in this example) is to use the PACKAGE argument when calling .C.
• I'm running on a x86_64 linux system, Windows might very well be a different story.

[hadley]

Yeah, I think this is out of scope for devtools, since:

• It's really R's fault
• There's already an existing best practice (using PACKAGE or registration) that avoids the problem.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/