[R] building RPMs for R packages

[Cooper Bethea]

On Sat, 2003-06-14 at 04:12, Martin Maechler wrote:

> I'm not really answering your primary question, 
> BUT
    - don't you share some file system(s) between all clients?

*share library directory across cluster*
> 
>     - if you don't share any file systems {quite improbable} AND

*distribute libraries over rsync*

> Hence, no really deep need for RPMs there, I think.

thanks for the input, but i don't agree with you here.
for 1) - it keeps network traffic down to keep libraries on local disk.
some of our computations are not even close to embarrassingly parallel,
and this would impede performance.
2) some of /etc directories are shared across the cluster because
they're small files that are easily cached and it's more convenient that
way. however, this is a more reasonable idea.

but those are still totally nasty hacks if you ask me. the way to
install software on an rpm-based system is via rpm.

the reason i need to build a package here (and the reason any other sys.
admin on an rpm-based system should) is because packages not installed
under rpm have a chance of breaking during system upgrades. in order to
maintain my sanity, i need to know that package R-foo needs to be
upgraded when i upgrade packages R, dependency blah, dependency blah,
and i need rpm to know that so my install tool will do it automatically.
if it doesn't my users' jobs will break and they will yell at me and i
will cry. the methods you propose are not scalable.


> 
>     Cooper> so could someone who's built R extension RPM
>     Cooper> packages give me some pointers? a SRPM for a package
>     Cooper> would be invaluable.