R devel - Jan 2006 - R 2.2.2-1 RPM build problem and solution on RH AS 4 x86_64

I have a dual Xeon x86_64 system running Red Hat AS 4. There are no
x86_64 rpms in http://cran.us.r-project.org/bin/linux/redhat/el4/ (the
i386 ones are a point release behind anyway) , and the fc4 rpms have a
whole web of dependencies I don't want to pull in. So I decided to
build http://cran.us.r-project.org/bin/linux/redhat/SRPMS/R-2.2.1-1.fc3.src.rpm
.

When I ran rpmbuild. one of the make-check tests failed.

from /BUILD/R-2.2.1/tests/p-r-random-tests.Rout.fail:
> dkwtest("weibull",shape = 1)
weibull(shape = 1) FAILED
Error in dkwtest("weibull", shape = 1) : dkwtest failed
Execution halted

I was able to build the rpm after removing "--enable-r-shlib" from the
spec file.

http://cran.us.r-project.org/bin/linux/redhat/SRPMS/ReadMe says:
"The new SRPM for R 2.1.1 builds the shared library version of R. This is,
unfortunately, slower than the version without the shared library."

It doesn't say why, if it's slower, it builds it that way. Can anyone
shed some light on the subject?

Dan

On Wed, 11 Jan 2006, Dan Lipsitt wrote:
> I have a dual Xeon x86_64 system running Red Hat AS 4. There are no
> x86_64 rpms in http://cran.us.r-project.org/bin/linux/redhat/el4/ (the
> i386 ones are a point release behind anyway) , and the fc4 rpms have a
> whole web of dependencies I don't want to pull in. So I decided to
> build
http://cran.us.r-project.org/bin/linux/redhat/SRPMS/R-2.2.1-1.fc3.src.rpm
> .
You might consider the fc3 RPMS: is RHEL4 not closer to FC3 than FC4?
(They may not have made your mirror yet, but they are en route.)
> When I ran rpmbuild. one of the make-check tests failed.
That test is documented to be random and fail sometimes.  Please try it
again.
> from /BUILD/R-2.2.1/tests/p-r-random-tests.Rout.fail:
>> dkwtest("weibull",shape = 1)
> weibull(shape = 1) FAILED
> Error in dkwtest("weibull", shape = 1) : dkwtest failed
> Execution halted
>
> I was able to build the rpm after removing "--enable-r-shlib"
from the
> spec file.
>
> http://cran.us.r-project.org/bin/linux/redhat/SRPMS/ReadMe says:
> "The new SRPM for R 2.1.1 builds the shared library version of R. This
is,
> unfortunately, slower than the version without the shared library."
>
> It doesn't say why, if it's slower, it builds it that way. Can
anyone
> shed some light on the subject?
This _is_ discussed in the the R-admin manual, to which the INSTALL file 
refers you.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

On Wed, 2006-01-11 at 17:26 -0500, Dan Lipsitt wrote:
> I have a dual Xeon x86_64 system running Red Hat AS 4. There are no
> x86_64 rpms in http://cran.us.r-project.org/bin/linux/redhat/el4/ (the
> i386 ones are a point release behind anyway) , and the fc4 rpms have a
> whole web of dependencies I don't want to pull in. So I decided to
> build
http://cran.us.r-project.org/bin/linux/redhat/SRPMS/R-2.2.1-1.fc3.src.rpm
> .
> 
> When I ran rpmbuild. one of the make-check tests failed.
> 
> from /BUILD/R-2.2.1/tests/p-r-random-tests.Rout.fail:
> > dkwtest("weibull",shape = 1)
> weibull(shape = 1) FAILED
> Error in dkwtest("weibull", shape = 1) : dkwtest failed
> Execution halted
> 
> I was able to build the rpm after removing "--enable-r-shlib"
from the
> spec file.
> 
> http://cran.us.r-project.org/bin/linux/redhat/SRPMS/ReadMe says:
> "The new SRPM for R 2.1.1 builds the shared library version of R. This
is,
> unfortunately, slower than the version without the shared library."
> 
> It doesn't say why, if it's slower, it builds it that way. Can
anyone
> shed some light on the subject?
Well, I did it because people were asking for it. You need the shared
library to use embedded R or use a GUI. I considered that most people
using R on the command line would pay the speed penalty (or not notice)
and that people who really need the speed could always compile their
own. The penalty is not so bad (~10%) on x86_64 anyway.

Martyn



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