64-bit R-build on Mac OS X 10.4 - make check failures

I am not sure why Simon thought you should ask here

(and not on R-sig-mac,
say): it is almost certain these are problems with your system not R.
For the first, use your debugger to find where the segfault  
occurred (it
might be inadequate stack space).  For the second, look at the  
values of
dbeta(0, 0.9, 2.2, ncp = c(0, a)), and trace the C code to find out  
why
you are not getting Inf (it is probably a broken libm function).

I'd be worried about the first: the second is probably esoteric.

On Wed, 7 Nov 2007, Steven McKinney wrote:

Hi all,

I compiled 64-bit R on an Apple Mac G5 running OS X, but it failed
make check.  Simon Urbanek suggested I post results to R-devel.

On Nov 6, 2007, at 10:23 PM, Steven McKinney wrote:

Hi Simon,

Would you be able to give more guidance on how to compile 64-bit
libiconv for Tiger,

You can get the sources from Apple and compile it:
http://www.opensource.apple.com/darwinsource/tarballs/other/ 
libiconv-13.2.tar.gz

If you want to save yourself the hassle, my 64-bit build can be
obtained and installed as follows:
curl -s http://r.research.att.com/libiconv-quad.tar.gz | \
sudo tar fvxz - -C /usr/lib

Many thanks to Simon Urbanek for providing a 64-bit libiconv for
use on the Apple Mac OS X platform.  This allowed compilation
on the Mac without having to use the '--without-iconv' switch
(for those of us who are not yet proficient at compiling 64-bit
libraries on the Mac).

Compilation of 64-bit R succeeded, but make check did not.

First failure was in ok-errors.R, and generated this output
in ok-errors.Rout.fail

## bad infinite recursion / on.exit / ... interactions
bar <- function() 1+1
foo <- function() { on.exit(bar()); foo() }
foo() # now simple "infinite recursion"

   *** caught segfault ***
  address 0x7fffeff7ffe30, cause 'memory not mapped'

  Traceback:
   1: foo()
   2: foo()
   3: foo()
  ...
  2154: foo()
  2155: foo()
  2156: foo()
  aborting ...

Second error was in d-p-q-r-tests.Rout from this test:
  ## dbeta(*, ncp):
  a <- rlnorm(100)
  stopifnot(All.eq(a, dbeta(0, 1, a, ncp=0)),
            dbeta(0, 0.9, 2.2, ncp = c(0, a)) == Inf
            )
  ## the first gave 0, the 2nd NaN in R <= 2.3.0

and generated this
output in d-p-q-r-tests.Rout.fail

## dbeta(*, ncp):
a <- rlnorm(100)
stopifnot(All.eq(a, dbeta(0, 1, a, ncp=0)),

  +           dbeta(0, 0.9, 2.2, ncp = c(0, a)) == Inf
  +           )
  Error: dbeta(0, 0.9, 2.2, ncp = c(0, a)) == Inf is not all TRUE
  In addition: Warning message:
  In dnbeta(x, shape1, shape2, ncp, log) : NaNs produced
  Execution halted

Everything else passed, so I'm not sure how serious these failures  
are.

I'd appreciate any suggestions about how to figure out what caused  
these
failures (an inappropriate file somewhere in the path, a possible
bug, improper compilation commands etc.)


version information of compiled R (compilation information at end):

sessionInfo()

R version 2.6.0 Patched (2007-10-29 r43302)
powerpc64-apple-darwin8.10.0

locale:
C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base

other attached packages:
[1] lme4_0.99875-8    Matrix_0.999375-3 lattice_0.17-2

loaded via a namespace (and not attached):
[1] grid_2.6.0      rcompgen_0.1-17

R.Version()

$platform
[1] "powerpc64-apple-darwin8.10.0"

$arch
[1] "powerpc64"

$os
[1] "darwin8.10.0"

$system
[1] "powerpc64, darwin8.10.0"

$status
[1] "Patched"

$major
[1] "2"

$minor
[1] "6.0"

$year
[1] "2007"

$month
[1] "10"

$day
[1] "29"

$`svn rev`
[1] "43302"

$language
[1] "R"

$version.string
[1] "R version 2.6.0 Patched (2007-10-29 r43302)"

.Platform

$OS.type
[1] "unix"

$file.sep
[1] "/"

$dynlib.ext
[1] ".so"

$GUI
[1] "X11"

$endian
[1] "big"

$pkgType
[1] "source"

$path.sep
[1] ":"

$r_arch
[1] ""

I built 64-bit R on Mac OS X 10.4.10 with these commands:

export PATH='/usr/local/lib64/X11/bin:/usr/local/lib64/X11/etc:/ 
usr/local/lib64/X11/include:/usr/local/lib64/X11/lib:/usr/lib:/usr/ 
bin:/usr/local/sbin:/usr/local/bin:/bin:/usr/sbin:/sbin:/usr/local/ 
teTeX/bin/powerpc-apple-darwin-current:~/perl:~/bin'

./configure --host=powerpc64-apple-darwin8.10.0 --build=powerpc64- 
apple-darwin8.10.0 \
--prefix=/usr/local/lib64 'CC=gcc-4.0 -arch ppc64' 'CXX=g++ -arch  
ppc64' \
'FC=gfortran-4.0 -arch ppc64' 'F77=gfortran-4.0 -arch ppc64' \
'CFLAGS=-g -O3 -mtune=G5 -mcpu=G5' 'FFLAGS=-g -O3 -mtune=G5 - 
mcpu=G5' \
'LDFLAGS=-arch ppc64 -m64 -L/usr/local/lib' 'CXXFLAGS=-g -O3 - 
mtune=G5 -mcpu=G5' \
'FCFLAGS=-g -O3 -mtune=G5 -mcpu=G5' --disable-R-framework --enable- 
R-shlib \
'--with-blas=-framework vecLib' --with-lapack 1>  
configure.R.ppc64.out.20071107.Run01.txt 2>&1

sudo make 1> make.R.ppc64.out.20071107.Run01.txt  2>&1

sudo make check 1> make.check.R.ppc64.out.20071107.Run01.txt  2>&1




Steve McKinney

-- 
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