A few days ago Gong Yu alerted this list to the possibility of
building a 64-bit R for Windows under a recent MinGW-w64 toolchain,
something we had failed to make work in 2007, 2008 and Feb 2009.
We've now completed the port in the R-devel (SVN trunk) sources and
are able to successfully complete 'make check-all'. An experimental
installer based on this version is available at
http://www.stats.ox.ac.uk/pub/RWin/Win64/R-2.11.0dev-win64.exe
but people who want to experiment may prefer to build from the R-devel
sources, and all users should read the new 'Building R for 64-bit
Windows' section in the R-admin manual in R-devel.
Some notes:
- The tcltk support is a stub. We have been able to build Tcl/Tk with
MinGW-w64, but just as with a 32-bit MinGW build, it is liable to
crash (which is why we use Tcl/Tk built under VC++6). It ought to be
possible to build Tcl/Tk under the Platform SDK, but that is left for
an interested party (it does not work out-of-the-box).
- There is no support for binary packages and the default package type
is "source". For packages without compiled code the 32-bit binary
packages will work (but such packages can also simply be installed
from their sources). For compiled code you will need to adjust
etc/Makeconf unless you are using the mingw-w64-bin_i686-mingw
cross-compiler (running under 32-bit Windows) mentioned in the R-admin
manual. That compiler uses dynamic libraries, and C++ users (and
perhaps others) will need to have the toolchain's bin/ directory in
their path or copy the DLLs to RHOME/bin. (Other toolchains can be
used, including building MinGW-w64 from the sources with static
libraries.)
- We found this build to be as fast as, and sometimes a bit faster
than, the 32-bit CRAN windows build.
- There is no commitment to support this other than in the sources,
and in part we are making experimental builds available to judge
interest and attract support. We see this as only being of interest
to Windows users with a 64-bit OS and substantially more than 4GB of
RAM who only use a few relatively simple packages (and there are other
options from R redistributors).
- Feedback and offers of support to R-windows at r-project.org,
please. At this stage we are looking for serious testers who are able
to compile from the sources and submit patches, and offers to
e.g. resolve the Tcl/Tk issues.
- Thanks to Gong Yu and especially Kai Tietz (the MinGW-w64 lead
developer) for help along the way.
Brian Ripley
--
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
