On 20/03/2013 14:14, Kasper Daniel Hansen wrote:
I have heard rumors of GCC 4.7.2, has this been abandoned?
At least for 3.0.0. It did not work well enough.
A more likely move is to clang. Clearly that has a future whereas
Apple's port of gcc does not. However, R package writers seem beloved
of g++'s lack of C++ standards conformance, and quite a high proportion
of C++-using packages do not compile with clang++ (nor the Solaris CC).
Kasper
On Wed, Mar 20, 2013 at 5:23 AM, Prof Brian Ripley
<[email protected]> wrote:
R 3.0.0 is due to be released on Apr 3, and is now in beta. Simon is
travelling but before he left put up a binary distribution at
http://r.research.att.com/snowleopard/R-3.0-branch/R-3.0-branch-snowleopard-signed.pkg
. A high proportion of CRAN is available as binary packages, and we will
improve that over the next two weeks.
UseRs and especially package developers are encouraged to try this out. For
CRAN packages there are logs available at
http://cran.r-project.org/web/checks/check_results_NAME.html (replace NAME
by the name of your package).
Note that there are quite a few changes in this pre-release:
- it requires Snow Leopard or later.
- it is x86_64 only and sub-architectures are not used.
- the recommended compilers have changed and the CRAN distribution uses
rather specific names for them to make it easier to match them. Those
compilers need Xcode 4.2 or later and (if relevant) a Fortran compiler from
http://r.research.att.com. Note that the binary is set up for the matching
gfortran for 10.6: if you have Lion or Mountain Lion you will need to edit
FLIBS in etc/Makeconf.
- the relevant manuals are part of the distribution, or see
http://cran.r-project.org/doc/manuals/r-devel/R-admin.html, especially
chapter 4 and sections 6.3.2 and C.4.
I've reported most of the issues thrown up with CRAN packages to their
maintainers. Common issues are
- including headers such as R.h which have #defines before system headers
and/or without R_NO_REMAP or similar (see 'Writing R Extensions').
- misuse of 'inline' in C code: only 'static inline' is really portable.
- lack of care with OpenMP: the CRAN distribution does not support OpenMP
and this toolchain has flaky OpenMP support.
Note too that checking with clang (at least the version in Xcode 4.6) rather
than gcc shows up further problems: the R-admin manual shows you how to set
that up for yourself.
--
Brian D. Ripley, [email protected]
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
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
--
Brian D. Ripley, [email protected]
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
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
