[R-SIG-Mac] Using multiple CRAN R-versions for building binary packages in a single machine?

[Reijo Sund]

I try to maintain a build system for the R packages so that I can easily produce all required Windows and Mac OS X binaries (for R versions 2.15.x and up) whenever I update my packages. I acknowledge that excellent services such as CRAN or RForge could be used, but to maximise the flexibility required by my packages, I prefer to have my own private building system. Currently the build system uses several virtual machines and a separate VM is needed for Mac OS X Leopard, Snow Leopard and Mavericks. I would like to use the CRAN versions of R to maximise the compatibility of binaries.

Unfortunately I have had problems to install multiple CRAN versions of R to the same VM. The R-admin manual (chapter 4) gives instructions for installing multiple versions (subchapter 4.3), but even if I use pkgutil to forget the old version and use command line such as /Library/Frameworks/R.framework/Versions/3.1/Resources/bin/R to run R, it always launches the version installed latest.

A quick look to the R ”binary" (for which the command line refers to) reveals that it is actually a shell wrapper for R executable that contains paths not including the version information. It is not difficult to change "hard coded" paths R_HOME_DIR, R_SHARE_DIR, R_INCLUDE_DIR and R_DOC_DIR in the wrapper and then indeed the correct version launches.

Unfortunately, the building of packages including C code using different versions seems not to work for other than the version installed latest even after changing the paths; or actually the compilation phase seems to work fine and linking paths etc. are conditional to the version, but the "testing if installed package can be loaded” results in a segfault. The similar procedure works fine for me if I (instead of installing CRAN version of R) manually build R from sources to some directory and use it for building of packages (I use this technique to build against Aqua Tcl/Tk), but then the compatibility of builded binary packages will be restricted to my configuration if I have understood correctly.

Currently the only working solution I have found has been to install a new VM for each (major) version of R. That is actually a good way also to to update the toolchains, but probably and hopefully that is not required for each version of R as each new VM increases the burden for maintaining a building system.

So even after reading the manuals and FAQs several times, I have not so far been able to sort this out myself (i.e. how to use several CRAN R versions to build packages containing C code on a same Mac OS X VM), so any tips would be appreciated. 


Best wishes,

Reijo Sund
Centre for Research Methods,
Department of Social Research,
University of Helsinki,
Finland