Build R with OpenBLAS and OpenMP support on MacOS 10.14 (Mojave)

These instructions have been updated (August 2019) to account for the fact that Homebrew no longer allows optional installation arguments. Initally, I just rolled with the changes since was I still able to set up OpenBLAS support. But after a helpful email exchange with Luis Puerto in which he pointed out that the standard (and now optionless) Homebrew R did not have full capabilities(), I’ve updated this post. I broadly follow his process, which can be found here and here, but with the addition of my Makevars and .Renviron files. For reference, the obsolete version of these instructions can be found here.

These instructions allow you to compile R from source on MacOS 10.14 (Mojave) so that it both uses OpenBLAS libraries (for faster computations) and has OpenMP support (to take advantage of multiple cores).

Rationale

While the Clang compilers shipped with R finally support OpenMP for parallel processing, they require their own unique flags. The GNU family of compilers (gcc, g++, etc) don’t play nice with all R packages when building them from source. I figured both out the hard way.

To get R to work with both OpenMP (after compiling with OpenBLAS) and packages built from source, I use a newer, non-Apple version of the LLVM compiler.

But why?

Why not?

Seriously, don’t try this if you don’t like the command line or have a low tolerance for potentially borking your R installation. This worked for me and was a (mostly) fun project while setting up a new machine, but YMMV. Of course, I’ve had to update since Homebrew changed its default behavior, so keep that in mind if you decide to go this route.

Tools

You’ll need:

XCode
Homebrew
~/.R/Makevars
~/.Renviron

XCode

Install Command Line Tools through XCode
You may need to type xcode-select --install in the terminal after installation, but I got this message which makes me believe this is no longer necessary:
```
xcode-select: error: command line tools are already installed, use "Software Update" to install updates
```
Figure out where your missing header files are located on your machine. Since 10.14, installing the Command Line Tools doesn’t place necessary header files (e.g., math.h) in the /usr/include directory, where they have historically been and R still looks for them when you build from source. They are instead kept within Xcode.app.

There are two steps:
1. At the time of this post (August 2019), you can still just copy the header files to /usr/include using this process. Specifically:
```
  sudo installer -pkg /Library/Developer/CommandLineTools/Packages/macOS_SDK_headers_for_macOS_10.14.pkg -target /
```
  I don’t know that this approach will continue to work with future Apple updates, but it’s necessary at the moment.
2. Figure out the path to the SDK headers using the following command:
```
  xcrun --show-sdk-path
```
  On my machine, the path is
```
  /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.14.sdk
```
  This is the standard path at the moment and so should be the same on your machine unless you’ve installed Xcode in a non-standard location. Take note of this path because we’ll add it to our Makevars file.

Homebrew packages

Install Homebrew
brew install some apps/libraries from homebrew-core:
- brew cask install xquartz
- brew cask install adoptopenjdk
  - NOTE: You can also choose oracle-jdk or not to install Java at all; if you choose not to install Java, then drop the --with-java flag when installing R
- brew install openblas
- brew install llvm
- brew install ccache
- brew install gcc (should already install with llvm)
- brew install libomp
- brew install libgit2
- brew install openssl (should already install with libgit2)
- brew install gettext
brew install some optional libraries from homebrew-core
- Note: Some of these are dependencies of others so you may get messages saying a library is already installed, which is fine):
- brew install liblwgeom
- brew install gdal
- brew install geos
- brew install proj
- brew install udunits
- brew install qt
Many R packages look for these libraries on your system if built from source, so unless you have compelling reason not to do so, I would highly recommend installing them. In particular, the sf package will need these in order to work.

NOTE There’s a Homebrew tap dedicated to QGIS and other spatial libraries currently under development (h/t again to Luis Puerto). You may find these more suitable for your geospatial needs than the versions provided by homebrew-core.
Add Seth Fore R tap:
```
 brew tap sethrfore/r-srf
```
brew reinstall Cairo from sethrfore/r-srf tap (you must reinstall since the Homebrew-core version of Cairo is already installed as a dependency above):
```
 brew reinstall sethrfore/r-srf/cairo
```
NOTE You need to install Cairo from this tap if you want to have X11 support.

Now install R using from sethrfore/r-srf with additional arguments:
```
 brew install sethrfore/r-srf/r --with-openblas --with-java --with-cairo --with-libtiff
```

If everything has gone well, you should be able to open up R and see the following when you run capabilities() (i.e., all TRUE):

> capabilities()
       jpeg         png        tiff       tcltk         X11        aqua 
       TRUE        TRUE        TRUE        TRUE        TRUE        TRUE 
   http/ftp     sockets      libxml        fifo      cledit       iconv 
       TRUE        TRUE        TRUE        TRUE        TRUE        TRUE 
        NLS     profmem       cairo         ICU long.double     libcurl 
       TRUE        TRUE        TRUE        TRUE        TRUE        TRUE

Set up ~/.Renviron and ~/.R/Makevars

Create .Renviron in home directory (~) and Makevars file in ~/.R directory that looks like the files below. This step is important. If you don’t add these files, you’ll not be able to install packages since they won’t be able to find the system libraries they need.

Install packages

Open R and install packages like normal using install.packages(). I would recommend using the option, dependencies = TRUE, if you want to make sure everything is compiled correctly.

I have successfully installed the following packages (and their dependencies) using this set up: