[R-SIG-Mac] Choosing between Snow Leopard and Mavericks builds

[Gow, Ian]

Thank you.

In case it helps other users of the Mavericks build: The instructions for installing the gfortran binaries (<http://r.research.att.com/libs/gfortran-4.8.2-darwin13.tar.bz2>) linked to at <http://cran.r-project.org/doc/manuals/r-release/R-admin.html#OS-X> are to be found here <http://r.research.att.com/libs/>. It took me a little digging to find these instructions; I guess I had been spoilt by the gfortran-4.2.3.dmg installer available for the Snow Leopard build.

While I generally install binary packages, I was dabbling with RcppArmadillo and, while the binary version of this package installs fine, it doesn't work without the right build environment.

(Before discovering the instructions at http://r.research.att.com/libs/, I tried using the gfortran-4.8.2 I had obtained from elsewhere, but it was not where R expected to be and I needed to set FLIBS in ~/.R/Makevars to get it to work for me.)

Thanks,

-- 
Ian Gow
Assistant Professor
Harvard Business School
Morgan Hall 379
Boston, MA 02163
Phone: +1-617-495-6530
Email: igow at hbs.edu

On Apr 22, 2014, at 3:04 AM, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:

> I've noticed some confusion about which CRAN installer package to use, 
> and even reluctance to update to 3.1.0.  So some pointers for Mavericks 
> users
> 
> 1) AFAIK no users are going to be worse off using the Snow Leopard build 
> of 3.1.0 rather than the only (Snow Leopard) build of 3.0.x.
> 
> 2) The Mavericks build has support for C++11-using packages (there are 
> currently less than a handful) and some performance improvements from 
> its later compilers (and in rare cases these are large).
> 
> 3) If you rely on binary packages with are not available for the 
> Mavericks build, install the Snow Leopard one.  Currently that includes 
> CRAN packages rgdal and rjags and all the BioC packages.
> 
> 4) If you install packages from source, you will be most likely better 
> off with the Mavericks build.
> 
> 5) The two builds need different Fortran compilers installed: you can 
> have both compilers at once.  The details are at 
> http://cran.r-project.org/doc/manuals/r-release/R-admin.html#OS-X .  As 
> far as I know a Fortran compiler is only needed if you install packages 
> from sources.
> 
> Some of the Mavericks binary packages are missing for good reason.  The 
> C++ binary interface for the Xcode 5.x compilers in Mavericks is 
> completely different, so external software has to be compiled with those 
> compilers.  This affects CRAN packages
> 
> RQuantlib RProtoBuf RVowalWabbit rgdal rjags rzmq
> 
> Most of these can be gotten to work (see the results labelled 
> r-devel-osx-x86_64-clang at 
> http://cran.r-project.org/web/checks/check_summary.html) if compiling 
> from sources.  There are packages with badly-written C++ which do not 
> compile under Mavericks, but as we have been chasing the CRAN 
> maintainers for a few months there are very few left (the main example 
> is BioC's package mzR).
> 
> We had expected that by this time Mavericks users would be in a 
> majority, but it seems the proportion is in the 40s (%).  Thus for R 
> 3.1.0 most of the effort has gone into the Snow Leopard build: expect 
> that to change by R 3.2.0 in a year's time.
> 
> -- 
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
> 
> _______________________________________________
> R-SIG-Mac mailing list
> R-SIG-Mac at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-mac