R 3.0.2/3.0.3 source install problems on OS X 10.9.2

On 11/03/2014 07:40, Berend Hasselman wrote:

On 11-03-2014, at 05:32, Bryan Renne <brenne at gmail.com> wrote:

Dear Simon:

On Mon, Mar 10, 2014 at 11:23 AM, Simon Urbanek
<simon.urbanek at r-project.org> wrote:

Bryan,

based on the gist output I think I have a candidate - please check that
you're not re-defining TMP or TMPDIR to a relative path - that would break
pretty much anything. Check with

set | grep 'T*MP'

This must be either undefined or the paths *must* be absolute, i.e.
start with a slash /

Other replies below, but I think the above may be the main issue.

That's it!  I can't believe you noticed that.  Thank you so much for your
help.

I had executed "export TEMPDIR=." for some other program (I think
latex2html?) that requires some non-null setting and, after clearing
this setting so that TEMPDIR is unset, the installation of recommended
packages completed and the make script finished without errors.  Looks
like an absolute path should be used as you suggest.

Isn't TMPDIR set automatically by OS X?
I've never seen it to be null on my machines.

Yes, but you can reset it (and sometimes it is necessary because of its
length or the non-alphanumeric characters Apple used to put in it).

What puzzles me is that TEMPDIR (with an E) is ever used: it is not used by
R.

Berend

Problem solved.  Thank you so much for your help with this.

On Mar 10, 2014, at 12:53 AM, Bryan Renne <brenne at gmail.com> wrote:

Hi Simon.  Thank you for your reply.

On Sun, Mar 9, 2014 at 6:41 PM, Simon Urbanek
<simon.urbanek at r-project.org> wrote:

Bryan,

On Mar 9, 2014, at 3:59 PM, Bryan Renne <brenne at gmail.com> wrote:

Thank you for your reply, Peter.

On Sun, Mar 9, 2014 at 3:31 AM, peter dalgaard <pdalgd at gmail.com>
wrote:

On 09 Mar 2014, at 00:21 , Bryan Renne <brenne at gmail.com> wrote:

Hi there.  I am having compilation problems with R 3.0.2 (and R
3.0.3) on
OS X 10.9.2.

On homebrew.... (Would have been useful to say up front.)

Actually, no.  This is *not* a homebrew specific problem,

It *is* as you even point yourself further down. It doesn't occur on a
clean OS X system.

OK, so to rule out any possible homebrew issues, I used the homebrew
uninstall script to delete homebrew and all homebrew-installed stuff
(https://gist.github.com/mxcl/1173223).  This script wipes the system
clean of homebrew.

Thanks, it would be sufficient to just rename /usr/local and re-install
the gfortran from CRAN.

I cannot reproduce the issue with exactly your settings on a clean
machine, so it must be some custom configuration of yours. I suspect you may
have set TMP/TMPDIR to . which is invalid - please check.

Thanks,
Simon

Acknowledged.  Will take the point about /usr/local under advisement
going forward.

Thank you again for your time.

I then installed gfortran 4.8.2 binaries using the provided pkg file
for OS X Mavericks (http://gcc.gnu.org/wiki/GFortranBinaries#MacOS).

I then downloaded R-3.0.3.tar.gz from the main page
(http://cran.r-project.org/).  I untarred, configured (with options
"--with-aqua --enable-R-framework --with-lapack") and ran make.  I
received the same error as before.  Here is the tail end of the "make"
output:

begin installing recommended package MASS
Warning in read.dcf(fd <- file.path(pkg, "DESCRIPTION")) :
cannot open compressed file
'./RtmpwOScIp/R.INSTALL588267e46267/MASS/DESCRIPTION', probable reason
'No such file or directory'
Error: error reading file
'./RtmpwOScIp/R.INSTALL588267e46267/MASS/DESCRIPTION'
make[2]: *** [MASS.ts] Error 1
make[1]: *** [recommended-packages] Error 2
make: *** [stamp-recommended] Error 2

Please see the following gist for full details of the system
configuration and output of "configure" and "make":

https://gist.github.com/bprenne/9459571

In particular, you'll notice in the section for make the error I
quoted above.  Also, at the very bottom of the gist, I have provided
what I think shows that the error is happening with the "R CMD INSTALL
MASS.tgz" step in the installation of recommended packages (somewhere
deep within the Makefile steps).

[[...snip...]
I also tried setting R_INSTALL_TAR to the system tar or to a
homebrew-installed gnutar.  Neither helped.

Any suggestions on how I might proceed?

Hmm, difficult to tell. If push comes to shove, you could try
getting more info out of the INSTALL process (inject a
debug(tools:::.install_packages), I suppose, but it is a bit beyond me
currently). But first, I'd check what Makeconf sets the TAR variable to.

Thank you for these suggestions.  Makeconf contains this line:

TAR = /usr/local/bin/gtar

This is the homebrew-installed gnutar.  However, is the R binary
(which has properly compiled and is the source of the error in
library
installation) reading Makeconf to determine how to untar the library
.tgz?

No, it is set in etc/Renviron which is in turn populated via configure
by the value you see in Makeconf.

Try setting

TAR=/usr/bin/tar

since system tar normally works. But that may be only the tip of the
iceberg, we can't tell what else you installed that breaks otherwise working
system. If in doubt, share the list with us so we can try to replicate it.

Cheers,
Simon

Thanks for this suggestion with setting the TAR variable.  As you can
see from the gist (https://gist.github.com/bprenne/9459571), this is
now the default value because homebrew has been wiped from the system.
However, even with this value, the problem still obtains.

I've tried to provide everything I can think will be useful in the
gist.

Is there something more I can provide or some way I can get more
fine-grained debugging information to help in diagnosis?

I appreciate your time and any thoughts you might have that you think
would help.

--
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