Hi all,
after a busy week I am finally starting to get back to some kind of normality
(no pun intended :-) ).
At useR! 2008 I had Martin Plummer, Marc Schwartz and Peter Dalgaard in the
audience. The talk went well and it was a fun to do. The room was small for
the attendance and in my (biased) view it seemed that I had people coming on
purpose for the talk. :-)
After the session ended I had a friendly discussion with Martin and Marc
regarding Fedora and R relationship. :-)
Some of the topics covered were (in no particular order other than my notes).
1) The redundancy between the two mailing lists devoted to R and Fedora (this
and SIG list). Actually at that time I forgot the SIG mailing list (my bad). I
always keep forgetting about the different purpose of both lists.
2) Martin and Marc suggested one scenario that involved the removal of all R
packages from Fedora where the only remaining installed package would be the R
core package. Then using some kind of package, the better analogy that I found
was a backend to PackageKit providing automatic installs from CRAN.
My opinion (that I have also expressed back then) was that this would be a no
go for Fedora for two reasons, first the packages dependency outside of CRAN
would be hard to get right (think about packages like R-hdf5 that require
linking with an external package). The second and probably the most important
issue from our (Fedora) point of view would be the liability issue. We would
be connecting to a repository that we have no control about, I tried to
express the different reasons why this is not possible.
3) There are discussions at R about the next generation CRAN meta-information
(improved DESCRIPTION format) and the opinions of distributions are welcome.
4) Is there a link in fedora bugzilla to connect upstream for R bugs reported
to Fedora? I am not sure if this is possible with the new bugzilla but I
suppose that there are not so many bugs like that reported at Fedora.
5) One problem usually reported in the R users mailing list is the
installation procedure used to install R packages (not in Fedora) using the R
interface.
This problem has been reported again this weekend FWIW. Instead of providing
R-devel in the R package as Martin and Marc have suggested I think that we
could create a FAQ for R in Fedora. Something like:
*********************** Draft to R - Fedora FAQ **************************
Q: I tried to install the following R package in Fedora but R failed with
horrendous error messages, what should I do?
A: Look to the content of the R package you are trying to install (link to
CRAN).
1) Does it contains C or C++ files (.c or .cpp ending filenames)? If yes
then install R-devel if it is not yet installed.
[Instructions how to install R-devel using yum or packagekit]
2) Does it contains Fortran files (.f or .F ending filenames)? If yes then
install gcc-gfortran if it is not yet installed.
[Instructions how to install R-devel using yum or packagekit]
3) If the steps above are not enough and the package depends on an external
library you need to guarantee that you have installed the files necessary to
compile the package. Do not forget that the -devel subpackages are required to
be installed to compile and link with the external libraries.
******************************************************************************
This is my fable attempt to document some of the possible of the problems
found by our users. :-)
We could then expand the FAQ content to other topics like how to search Fedora
R packages. The target of the FAQ would be naturally the users and not just
the developers.
Comments are welcome to all the points above,
--
José Abílio