Preparing for R 2.12.x

3 replies [Last post]
mspiegel's picture
Offline
Joined: 07/31/2009

The R developers are aiming for a release of 2.12.0 sometime this fall. Here's what I would like to do with the OpenMx binary releases when the new version of R is released.

On Windows: the R tools build system will switch from gcc 4.2 to gcc 4.5. The R tools system is not designed to support multiple versions on the same machine. That means we intent to drop support for all versions of R on windows that are less than 2.12. The good news is we'll finally get 64-bit windows releases. Also, we'll need a new compile of libnpsol for gcc 4.5 and windows. And some kind of libnpsol cross compile for 32-bit and 64-bit systems.

On Mac systems: support for R 2.9.x will be dropped. Either upgrade your systems, or install the build tools to compile OpenMx from source. See http://openmx.psyc.virginia.edu/wiki/howto-build-openmx-source-repository.

tbates's picture
Offline
Joined: 07/31/2009
Most R packages warn you

Most R packages warn you about their being built on newer Rs than you are running, but run just fine on quite old installs.

Many students and even staff these days are in managed environments where they can't upgrade software - so you'll have people using office 2003 or something, and R 2.09.

On the 64bit front, many people have an idée fixe that drives them want 64-bit, but have yet to see anyone actually pushing the memory limit in 32bit R, and 32 often runs faster, not having to drag that usually empty upper word around the processor.

So if we force 2.12 only for a handful of edge users (most of whom will want SVN versions and could do the compile themselves) but a much larger group of users stuck on whatever system is in their PhD office that might not help us grow the user-base.

It would be kind if OpenMx supported two trailing major releases: say 2.10-2.11 as well as 2.12 (which sounds like what will be the case for OS X, but not Win?). Of course with a hard-pressed developer team, your needs would come first in my mind: Perhaps the gcc build increment is worth it?

Best,
tim

PS: Do we know what our windows/*nix/OS X user profile is like?

mspiegel's picture
Offline
Joined: 07/31/2009
tbates wrote: Most R

tbates wrote:

Most R packages warn you about their being built on newer Rs than you are running, but run just fine on quite old installs.

I agree with you on this point. OTOH, R 2.9 systems cannot read the documentation format in packages compiled with R 2.10+. We've been doing the courtesy of compiling OpenMx with R 2.9 and R 2.10 to support documentation on all versions. With the release of 2.12, I'd like to stop compiling on R 2.9. It's just me that has to compile all these versions, and it's getting out of hand. Maybe if we setup an automated build system with multiple virtual machines. There's a learning curve to create an automated build system that is not top priority at the moment.

Also, I suspect that R packages on windows compiled for R 2.12 will simply not work with earlier versions of R. They are changing the gcc compiler, I am assuming that parts of the R platform rely on an ABI, which will probably break. Again, if we had a volunteer who wanted to build windows binaries for versions earlier than 2.12, I'd be happy to give them instructions.

g.avenin's picture
Offline
Joined: 10/13/2009
To confirm your last

To confirm your last suspiçion MASS package 7.3.9 requires 2.12 and can no longer be compiled with 2.10.