Cannot load libR.dylib with JRI on OSX

Question

I have a custom compiled R installation. I had to custom compile it in order to use MKL.

In my build.sbt I have the following line

javaOptions += "-Djava.library.path=/Users/me/R-3.3.2/lib:/Users/taylor/Library/R/3.3/library/rJava/jri

and also:

fork := true so javaOptions works right.

To test this I fire up sbt and attempt to load jri with System.loadLibrary("jri"). I get the following error:

java.lang.UnsatisifiedLinkError: /Users/me/R/3.3/library/rJava/jri/libjri.jnilib: dlopen(/Users/me/Library/R/3.3/library/rJava/jri/libjri.jnilib, 1): Library not loaded libR.dylib

So I went to see with otool what it expects -

otool -L libjri.jnilib in /Users/me/Library/R/3.3/library/rJava/jri has the following in it: (remainder omitted)

libjri.jnilib
  libjri.jnilib (compatibility version 0.0.0, current version 0.0.0)
  ...
  libR.dylib (compatibility version 3.3.0, current version 3.3.2)
  ...

I thought this looked funny, so I copied libR.dylib from /Users/me/R-3.3.2/libinto /Users/me/Library/R/3.3/library/rJava/jri to see if it was just looking around locally.

Unfortunately I still get the same error.

I'm at a complete loss on what I need to do to make this work, and it has ground my entire project to a halt. Has anyone experienced this and fixed it?

Answer 1

First, note that you shouldn't need to re-compile R in order to use MKL, because the CRAN OS X binaries are intentionally shipped such that you can point libRblas.dylib to any compatible BLAS implementation - consider reading the R documentation.

As for the error, it seems that you have compiled R without framework support and/or have not installed it, hence libR doesn't contain the full path. Also you have likely not used the recommended way to start your Java application use R CMD. You can either use

install_name_tool -id <libR> <libR>

to fix your libR manually (where <libR> is the full path to your libR.dylib), or you can use

install_name_tool -change libR.dylib <libR> \
   /Users/me/Library/R/3.3/library/rJava/jri/libjri.jnilib

to force libjri to pick up you location of libR. There are other ways as well, such as setting DYLD_... env vars before you start Java or using R CMD.