[R-pkg-devel] The correct way to put additional DLLs on windows?

I recommend that you "reply-all" to keep the mailing list included.


On July 1, 2019 11:30:23 PM PDT, Jialin Ma <marlin- at gmx.cn> wrote:

Hi Jeff,

Thanks for your reply. I would like to clarify a little bit:

A) You cannot assume the package code will have write access to the

library

directories. The user/admin who installs the package may never load

the

package.

I am downloading the necessary binary dependencies at *compile time*,
i.e. the
script is being called inside "Makevars.win". I suppose the script
should have
write access to the current directory, i.e. "inst" directory, at this
stage?
But I am not sure about it.

I am aware many packages will download static libraries on windows at
this
stage. For example the Cairo package (see
https://github.com/cran/Cairo/blob/
master/configure.win ) mentioned in Writing R Extensions will download
binary
to the source directory and link against them.
My question is about whether it is an accepted practice to put DLL

file

in
"inst/libs".

B) Downloading binary files for execution without the direct

involvement of

the user/admin is a security risk.

C) Open-source package license means the source code is part of the

package.

Downloading binaries later fails this requirement.

This however is a workaround for the windows platform due to the
difficulties
compiling the external libraries. The toolchains needed (specifically,
cmake)
was missing in Rtools and there are several dependencies. I believe
there
should be some accepted practice for the license issue, however I am
not sure.

Best regards,
Jialin


On Monday, July 1, 2019 10:49:43 PM PDT you wrote:

A) You cannot assume the package code will have write access to the

library

directories. The user/admin who installs the package may never load

the

package.

B) Downloading binary files for execution without the direct

involvement of

the user/admin is a security risk.

C) Open-source package license means the source code is part of the

package.

Downloading binaries later fails this requirement.
On July 1, 2019 10:03:07 PM PDT, Jialin Ma <marlin- at gmx.cn> wrote:

Hi everyone,

Currently my package has a script downloading additional DLLs
into "inst/libs" and being called in "Makevars.win". In this way

these

DLLs get copied to libs when installing the package and I am able

to

load them with "library.dynam".

However R CMD check gives the following warning.

 * checking package subdirectories ... WARNING

Found the following non-empty subdirectories of 'inst' also used

by

R:

   inst/libs

It is recommended not to interfere with package subdirectories

used

by

 R.

I was wondering what would be the best practice for my case, or I

am

safe
to ignore this warning (suppose I am planing for a CRAN

submission).

I know many packages will download static library and link them at
compile
time. But I haven't see any example with additional DLL files.

Thanks.
Jialin Ma

-- 
Sent from my phone. Please excuse my brevity.