pcarbo

So you want to install source packages on your Mac?

Then please follow these instructions and advice. However, please note that:

  1. These instructions were developed with R versions 3.4 and 3.5 in mind. If you are using a version of R prior to 3.4, please disregard this document as it is not (immediately) relevant to you.

  2. It is possible that these instructions will no longer be valid for future versions of R. So if you are using R > 3.5, follow these instructions at your risk.

Introduction to the problem

If you are using version R >= 3.4 in macOS, you should find that installing packages with the usual approach (e.g., using install.packages or devtools::install_github) works for most packages. However, you may find that you get strange errors when trying to install some packages. One possible reason (among many!) for this error is that installing these packages involves compiling C++ and/or Fortran code, and this compilation step failed.

Installing source packages which contain C/C++/Fortran code requires that compilers and related tools be installed.

As of R 3.4, the R development team decided to use a compiler that is no longer provided by Apple. There were good reasons for making this change:

Since R 3.4.0 release, we are now providing binaries for OS X 10.11 (El Capitan) and higher using non-Apple toolkit to provide support for OpenMP and C++17 standard features.

But… this has unfortunately made package installation and development more complicated for people with a Mac.

Key point: If you are using R >= 3.4 on Mac, and you have not installed the C and Fortran compilers recommended for your version of R, then you will have trouble installing some packages from source.

Here I explain how to properly (and safely) set up your Mac for installing R packages from source.

What I don’t recommend doing

Do not download and install the tools provided by CRAN. This will install these compilers in the system directories on your computer and could cause issues that will be difficult to fix.

My recommendation

Use package managers (Homebrew, MacPorts, Anaconda, Miniconda, etc.) to install the necessary compilers. This will allow you to install compilers without affecting the system directories. And if you make a mistake, you can easily reverse the changes.

Barebones setup instructions

I’ve made these instructions general enough that they should work for different versions of R. (I can’t guarantee that they will work, however.) If you are unfamiliar with the tools I recommend using (e.g., Homebrew), you will have to learn the basics of how to use them.

I also give the specific example of my setup, which is R 3.4.3 on a MacBook Pro (3.5 GHz Intel Core i7 CPU, macOS 10.13.4).

1. Install Homebrew.

Follow the installation instructions here.

2. Install MacPorts.

Follow the installation instructions here.

Search Homebrew and MacPorts for the recommended version of clang and gfortran, and install these packages (also called “ports”). (Note that clang may be bundled with an llvm package, and gfortran may be bundled with a gcc package.) For example, on my computer, I have R 3.4.3, so I should install clang 4.0.0 and gfortran 6.1 (according to here—I can’t find this information on the CRAN website any more). I found clang 4.0.1 and gfortran 6.4 (close enough, hopefully!) in MacPorts, so I ran these commands to install them:

sudo port install clang-4.0
sudo port install gcc6

You may also find suitable packages in Homebrew.

Confusingly, R 3.5 switched to clang 6.0!

4. Edit your Makevars file to tell R use the new clang and gfortran

Edit file ~/.R/Makevars to point to the newly installed versions of clang and gfortran. For example, on my computer with R 3.4.3, I have:

$ cat ~/.R/Makevars
CC       = /opt/local/bin/clang-mp-4.0
CXX      = /opt/local/bin/clang++-mp-4.0
CXX11    = /opt/local/bin/clang++-mp-4.0
CXX17    = /opt/local/bin/clang++-mp-4.0
CXX98    = /opt/local/bin/clang++-mp-4.0
F77      = /opt/local/bin/gfortran-mp-6
FC       = /opt/local/bin/gfortran-mp-6
LDFLAGS  = -L/opt/local/lib
FLIBS    = -L/opt/local/lib/gcc6 -lgfortran -lquadmath -lm

Note that all MacPorts packages are installed in /opt/local.

For more detailed advice on configuring your Makevars file, see here.

5. Test your new setup by installing a source package with C/Fortran code

For example, install glmnet 2.0-16 from source by running the following in R:

url <- "https://cran.r-project.org/src/contrib/glmnet_2.0-16.tar.gz"
install.packages(url,repos = NULL)

The glmnet package has both C and Fortran source code, and indeed when I run this command on my computer I see that it is using gfortran and clang to compile and build the package:

/opt/local/bin/gfortran-mp-6 -fPIC -g -O2 -c glmnet5dp.f -o glmnet5dp.o
/opt/local/bin/clang-mp-4.0 -I/Library/Frameworks/R.framework/Resources/include -DNDEBUG -I/usr/local/include -fPIC -Wall -g -O2 -c glmnet_init.c -o glmnet_init.o
/opt/local/bin/clang-mp-4.0 -dynamiclib -Wl,-headerpad_max_install_names -undefined dynamic_lookup -single_module -multiply_defined suppress -L/Library/Frameworks/R.framework/Resources/lib -L/opt/local/lib -o glmnet.so glmnet5dp.o glmnet_init.o -L/opt/local/lib/gcc6 -lgfortran -lquadmath -lm -F/Library/Frameworks/R.framework/.. -framework R -Wl,-framework -Wl,CoreFoundation

More info