On Jan 6, 2010, at 12:35 PM, Max Kuhn wrote:
> My company is trying to manage R installations across a number large
> SMP machines. We're thinking out the best way to manage the packages
> installs and updates. They would be happy if we could work out RPM's
> for package installations (traceable, easily facilitated with existing
> sw management tools).
>
> I don't know a lot and RPMs beyond how to use them, but it seems
> plausible to write R code to create the RPM package. If we need to
> update package X, which triggers and update of ancillary packages Y
> and Z, it should be possible to use available.packages() to figure out
> the dependencies, download the sources, write out the RPM headers and
> package things up.
>
> Before I try to write any code, does anyone see any issues with this
> (or has it already been done)? Is this a ridiculous approach?
>
> Thanks,
>
> Max
Max,
It would certainly be beneficial for you or someone at your shop to
get intimately familiar with creating RPM packages for Fedora/RHEL if
that is the way you want to go. There are various detail issues
involved in creating RPMs and for R packages specifically.
The following page from the Fedora wiki would likely serve as a
starting point for creating and maintaining R related RPMs:
https://fedoraproject.org/wiki/Packaging:R
The Fedora and RHEL/EPEL folks (basically the same people) are
providing standardized RPMs for selected CRAN packages and it would
make sense from a consistency perspective, not to mention not re-
inventing the wheel, to learn from their activity.
The R package managers involved in maintaining the R related RPMs have
their own e-mail list which you might find helpful, albeit the volume
there is low:
https://www.redhat.com/mailman/listinfo/fedora-r-devel-list
and of course we have our own Fedora SIG:
https://stat.ethz.ch/mailman/listinfo/r-sig-fedora
HTH,
Marc Schwartz