[R] R packages install problems linux - X not found (WhiteBoxEL 3)

[Dr Mike Waters]

> -----Original Message-----
> From: r-help-bounces at stat.math.ethz.ch 
> [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Marc Schwartz
> Sent: 09 August 2004 15:13
> To: Dr Mike Waters
> Cc: R-Help
> Subject: RE: [R] R packages install problems linux - X not 
> found (WhiteBoxEL 3)
> 
> 
> On Mon, 2004-08-09 at 08:13, Dr Mike Waters wrote:
> 
> <snip>
> 
> > Marc,
> > 
> > Sorry for the confusion yesterday - in my defence, it was 
> very hot and 
> > humid here in Hampshire (31 Celsius at 15:00hrs and still 25 at 
> > 20:00hrs).
> > 
> > What had happened was that I had done a clean install of WB Linux, 
> > including the XFree86 and other developer packages. However, the 
> > on-line updating system updated the XFree86 packages to a newer sub 
> > version. It seems that it didn't do this correctly for the XFree86 
> > developer package, which was missing vital files. However 
> it showed up 
> > in the rpm database as being installed (i.e. rpm -qa | grep XFree 
> > showed it thus). I downloaded another rpm for this manually 
> and I only 
> > forced the upgrade because it was the same version as already 
> > 'installed' (as far as the rpm database was concerned). I 
> assumed that 
> > all dependencies were sorted out through the install in the first 
> > place.
> 
> OK, that helps. I still have a lingering concern that, given 
> the facts above, there may be other integrity issues in the 
> RPM database, if not elsewhere.
> 
> >From reading the WB web site FAQ's
> (http://www.whiteboxlinux.org/faq.html) , it appears that 
> they are using up2date/yum for system updates. Depending upon 
> the version in use, there have been issues especially with 
> up2date (hangs, incomplete updates,
> etc.) which could result in other problems. I use yum via the 
> console here (under FC2), though I note that a GUI version of 
> yum has been created, including replacing the RHN/up2date 
> system tray alert icon.
> 
> A thought relative to this specifically:
> 
> If there is or may be an integrity problem related to the rpm 
> database, you should review the information here:
> 
> http://www.rpm.org/hintskinks/repairdb/
> 
> which provides instructions on repairing the database. Note 
> the important caveats regarding backups, etc.
> 
> The two key steps there are to remove any residual lock files 
> using (as
> root):
> 
> rm -f /var/lib/rpm/__*
> 
> and then rebuilding the rpm database using (also as root):
> 
> rpm -vv --rebuilddb
> 
> I think that there needs to be some level of comfort that 
> this basic foundation for the system is intact and correct.
> 
> > I only mentioned RH9 to show that I had some familiarity with the 
> > RedHat policy of separating out the 'includes' etc into a separate 
> > developer package.
> > 
> > Once all this had been sorted out, I was then left with a 
> compilation 
> > error which pointed to a missing dependency or similar, 
> which was not 
> > due to missing developer packages, but, as you and Prof Ripley 
> > correctly point out, from the R installation itself. Having 
> grown fat 
> > and lazy on using R under the MS Windows environment, I was 
> struggling 
> > to identify the precise nature of this remaining problem.
> > 
> > As regards the R installation, I did this from the RH9 binary for 
> > version 1.9.1, as I did not think that the Fedora Core 2 
> binary would 
> > be appropriate here. Perhaps I should now compile from the source 
> > instead?
> 
> I would not use the FC2 RPM, since FC2 has many underlying 
> changes not the least of which includes the use of the 2.6 
> kernel series and the change from XFree86 to x.org. Both 
> changes resulted in significant havoc during the FC2 testing 
> phases and there was at least one issue here with R due to 
> the change in X.
> 
> According to the WB FAQs:
> 
> "If you cannot find a package built specifically for RHEL3 or 
> WBEL3 you can try a package for RH9 since many of the 
> packages in RHEL3 are the exact same packages as appeared in RH9."
> 
> Thus, it would seem reasonable to use the RH9 RPM that Martyn 
> has created. An alternative would certainly be to compile R 
> from the source tarball.
> 
> In either case, I would remove the current installation of R 
> and after achieving a level of comfort that your RPM database 
> is OK, reinstall R using one of the above methods. Pay close 
> attention to any output during the installation process, 
> noting any error or warning messages that may occur.
> 
> If you go the RPM route, be sure that the MD5SUM of the RPM 
> file matches the value that Martyn has listed on CRAN to 
> ensure that the file has been downloaded in an intact fashion.
> 
> These are my thoughts at this point. You need to get to a 
> point where the underlying system is stable and intact, then 
> get R to the same state before attempting to install new packages.
> 
> HTH,
> 
> Marc
> 
>From unpacking the tarball and running ./configure in the R source
directory, I obtain the fact that crti.o is needed by ld.so and was not
found. This file is not present on the system. This file, along with crtn.o
is usually installed by the gnu libc packages, I believe. However, I know
that not all *nix distributions include these files among their packages.
>From a web search, I have not been able to ascertain whether this lack of a
crti.o is due to there not being one in the distribution, or to another
incomplete package install.

So, I did a completely fresh installation of WhiteBox, followed by R built
from source, checked that it ran and then installed the R packages. Only
then did I run up2date. At least crti.o and crtn.o are still there this
time, along with the XFree86 includes.....

A bit of a cautionary tale, all in all. 

Thanks for all the help and support.

Regards

M