We are required by our CERT team to update: they view this as a service
pack and hence compulsory for machines connected (directly, not via NAT)
to ox.ac.uk.
Please note that everything here is about upgrading to 10.9: things
could be different for a fresh install (we have not yet done one).
On 25/10/2013 02:39, Simon Urbanek wrote:
On Oct 24, 2013, at 6:16 PM, Kevin Ushey wrote:
With XCode 5.0, clang is now the default compiler toolchain supplied,
and llvm-gcc-4.2 is no longer available.
Yes, but for 10.8 it will still be there if you updated Xcode, and you
can also still install the command-line tools for 4.6.3 (as I posted
earlier).
If you upgrade to Mavericks (10.9) the llvm-* compilers will still be
present but some of the dylibs they link to are not.
FWIW, I and many others have had a lot of success compiling packages
with the version of clang supplied with XCode. The only requirement
(at this point) is that you set your own ~/.R/Makevars settings, as
discussed in the SO posts -- in particular, you need
CC=clang
CXX=clang++
Please don't quote 'SO posts' (and especially not without links). The
definitive information is in the R manuals (and has been since R 3.0.0).
See
http://cran.r-project.org/doc/manuals/r-release/R-admin.html#OS-X-packages
. Apart from anything else, the people who update the manuals deserve
credit for their work.
You may need to reinstall packages from source that your packages link
to (e.g., Rcpp) and compile them with the new set of compilers for
them to play nicely together, but in general this should work.
Actually, you don't. clang and llvm-gcc-4.2 are ABI compatible for
10.8. And I have successfully tested RcppEigen (which has just been
updated) on 10.9 with clang using the CRAN binary distribution of Rcpp.
However, hopefully others can give advice on the 'recommended'
compiler toolchain now for R and Mavericks.
The recommended toolchain as far as CRAN binaries are concerned
remains the same since the binaries did not change. I cannot currently
check on my 10.9 machine so I'll have to confirm it later, but there are
several options - you can install Apple's original llvm-gcc package or
use our gcc build (which includes gfortran). Depending on the route you
take, you can also simply symlink the compiler to lvm-g*-4.2 instead of
modifying Makevars.
The Makevars route works fine.
If you want to start from scratch (i.e. compile R and packages from
sources), then you can use clang, it's known to work (modulo a few
packages that are not up to date).
A gotcha we have seen so far:
/usr/X11 exists but is no longer a link to /opt/X11 for XQuartz. So I
needed --x-includes=/opt/X11/include --x-libraries=/opt/X11/lib on the
configure line.
Cheers,
Simon
-Kevin
On Thu, Oct 24, 2013 at 3:00 PM, David Winsemius <dwinsem...@comcast.net> wrote:
From comments made on the thread regarding the segfault in the R.app GUI following the
error from TRUE <- FALSE, I'm inferring that at least three people ( Prof. Ripley, S.
Urbanek, and T. Bates) have installed R 3.0.2 on Mavericks. I have R version 3.0.2
(2013-09-25) -- "Frisbee Sailing" with R.app GUI 1.60 (6475) and XCode Version
3.2.6 in OSX 10.7.5 and have installed the Command Line Tools, so at the moment I'm
generally succeeding when compiling from source. (I see I am rather behind the current GUI
version although that was not my intention when I updated to 3.0.2.) Are there any specific
steps one needs to take (or avoid taking) in order to preserve access to (or gain updated
access to) tools for compiling packages from source?
(I've seen people posting problems on SO where the error after updating to
Mavericks was a missing copy of llvm-gcc-4.2.)
--
David Winsemius
Alameda, CA, USA
--
Brian D. Ripley, rip...@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@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
