Re: [R] Fedora Core 4

From: <plummer_at_iarc.fr>
Date: Mon 20 Jun 2005 - 06:29:48 EST

The Fedora upgrade process should normally install backward compatibility libraries when it finds an RPM linked to a library (or library version) that isn't in the new release. In this case "compat-libf2c-32" provides libg2c on FC4, and it should be installed on your system if you previously had R for FC3 installed. So R should continue to work. But it will break when you start compiling packages on the new system because the configuration information in the file /usr/lib/R/etc/Makeconf is obsolete. Hopefully this situation will occur rarely, because there isn't really a way to deal with it automatically.

I'm pretty sure the current R RPM for FC4 on Red Hat Extras is broken. The RPM build process adds a lot of compiler flags. R 2.1.1 doesn't pass "make check-all" when built with these flags (and if you try the legacy compilers Marc mentioned, it doesn't even get past the configure step). A close look at the spec file from Red Hat shows that this line has been commented out, which was a mistake. I'll try to ensure that their next release works. In the mean time, I am bumping up the Epoch number of my RPMs to 1 so they should be considered an "upgrade" of the Red Hat ones.

Martyn.

Compiling R 2.1.1 on FC4 with gcc4

Quoting Marc Schwartz <MSchwartz@mn.rr.com>:
> In follow up to Prof. Ripley's post, having participated in the
> referenced r-devel thread, I would note that I have been able to build
> Version 2.1.1 beta (2005-06-17) and all CRAN packages from source using
> both FC4 fortran compilers.
>
> However, I am back to using g77 based upon Prof. Ripley's
> recommendations.
>
> As mentioned, the problem below is likely the mixing and matching of
> compiled R executables from FC3 and FC4.
>
> Jon, I would recommend re-compiling a full and clean installation of R
> (including any CRAN packages), using which ever version of R you are
> running at the moment.
>
> If you did not install the referenced compatibility RPMs during the
> initial installation, it is available via the Add/Remove Applications
> GUI:
>
> GNOME menu -> Desktop -> System Settings -> Add/Remove Applications
>
> Once the GUI is up, scroll down and select "Development". Then select
> the 2 packages in the "Legacy Software Development" set, which includes
> the two "compat" sets of programs. You will need your install CD/DVD to
> load the RPMS.
>
> Alternatively, you can use:
>
> # yum groupinstall "Legacy Software Development"
>
> from a console as root.
>
> HTH,
>
> Marc Schwartz
>
>
> On Sat, 2005-06-18 at 13:00 +0100, Prof Brian Ripley wrote:
> > There has been discussion of this on R-devel in the context of the release
> > of 2.1.1.
> >
> > FC4 contains gfortran, not g77 (unless you install everything when you
> > get g77 from gcc-3.2.3 in some compatibility RPM). This is not a bug: R
> > compiled on one OS (or even machine) is not expected to work on another.
> > -lg2c is part of gcc-3.x.y, and gfortran uses something like -lgfortran
> > (FC3's gcc4 didn't use the standard name, so I am being cautious).
> >
> > I think it is fortuitous that R runs in your tests without crashing:
> > libRlapack.so is linked against libg2c.0.0 and actually calls entry points
> > in it. (The R executable is also linked but seems to contain no such
> > entry points.)
> >
> > I would suggest installing 2.1.1 on Monday from the sources, or waiting
> > for a 2.1.1 RPM. I believe gcc-3.4.4 is still a better choice than
> > gcc-4.0.0 on ix86: it runs faster and there are several cases in which
> > 4.0.0 appears to give incorrect answers. I am waiting for the eventual
> > release of 4.0.1 to test these more thoroughly. (See the notes in the
> > R-admin manual: fortunately FC4's gcc4 is just recent enough to avoid the
> > showstopper bug.)
> >
> > On Sat, 18 Jun 2005, Jonathan Baron wrote:
> >
> > > I had installed R from source on Fedora Core 3. Then I upgraded
> > > to Fedora Core 4, but left R alone. R worked fine, until I trued
> > > to update.packages(). Even then, many packages updated just
> > > fine, but two of them, cluster and mgcv, failed with the
> > > following error message (using cluster as an example):
> > >
> > > gcc -shared -L/usr/local/lib -o cluster.so clara.o daisy.o
> > > dysta.o fanny.o meet.o mona.o pam.o spannel.o twins.o -lg2c -lm -lgcc_s
> > > /usr/bin/ld: cannot find -lg2c
> > > collect2: ld returned 1 exit status
> > > make: *** [cluster.so] Error 1
> > >
> > > I tracked this down as far as I could and got to this page:
> > > http://gcc.gnu.org/fortran/usage.html
> > > in the section about "Compatability with g77." The problem seems
> > > to have something to do with Fortran.
> >
> > Interesting spelling!
> >
> > > Before I invest more time, I wonder if someone knows what to do.
> > > Options:
> > > 1. Re-install R from source.
> > > 2. Try the RPM for Fedora Core 4. It isn't in CRAN, but it is in
> > > Fedora "extras". R is now part of the Fedora distribution,
> > > albeit one of the many things they have put in "extras" in an
> > > effort to limit the distribution to 4 CDs.
> > > 3. Report a bug. But where? Is it a bug in R?
> > > 4. Something else.
> >
>



This message and its attachments are strictly confidential. ...{{dropped}}

R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Mon Jun 20 06:36:09 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:32:51 EST