On 31/08/2015 16:26, Davor Cubranic wrote:
Prof Brian Ripley writes:

Second, we don't have all the pertinent information such as the
configure options used and the architecture (x86_64?).  I am going to
have to guess none as none were mentioned, but using --enable-R-shlib
would be pertinent.

On 31/08/2015 05:47, Davor Cubranic wrote:
On FreeBSD 10.2, I get the following error when compiling R from the
Subversion trunk (with "configure && make"):

You have not told us which revision.  A basic check is to see if you can
build the latest released version, as the trunk is 'Under development'.

As suggested, I tried compiling from Subversion tag 3.2.2 (r69054).
I used no command-line options to 'configure', as mentioned in my
previous email, and this is the output:

   R is now configured for x86_64-unknown-freebsd10.2

     Source directory:          .
     Installation directory:    /usr/local

     C compiler:                cc  -g -O2
     Fortran 77 compiler:       gfortran48  -g -O2

     C++ compiler:              c++  -g -O2
     C++ 11 compiler:           c++  -std=c++11 -g -O2
     Fortran 90/95 compiler:    gfortran48 -g -O2
     Obj-C compiler:         cc -g -O2 -fobjc-exceptions

     Interfaces supported:      X11, tcltk
     External libraries:        readline, zlib, bzlib, lzma, PCRE, curl
     Additional capabilities:   PNG, JPEG, TIFF, NLS, cairo, ICU
     Options enabled:           shared BLAS, R profiling

     Capabilities skipped:
     Options not enabled:       memory profiling

     Recommended packages:      yes

(I thought this, and more, would be included in config.log, but please
let me know if there is other place to get the configuration details
that are relevant.)

You need to tell us exactly which commands you used: nowhere records everything.


Still the same error:

   --- tools.so ---
   cc -shared -L/usr/local/lib -o tools.so text.o init.o Rmd5.o md5.o signals.o 
install.o getfmts.o http.o gramLatex.o gramRd.o
   --- all ---
   --- shlib ---
   mkdir ../../../../library/tools/libs
   --- sysdata ---
   installing 'sysdata.rda'
   Error in dyn.load(file, DLLpath = DLLpath, ...) :
     unable to load shared object 
'/usr/home/davor/R-3.2.2/library/tools/libs/tools.so':
     /usr/home/davor/R-3.2.2/library/tools/libs/tools.so: Undefined symbol 
"R_ClassSymbol"
   Error: unable to load R code in package 'tools'
   Execution halted

Here is a series of checks for that symbol (results from a working Linux
system):

auk% nm -g bin/exec/R  | grep R_ClassSymbol
0000000000962ec0 B R_ClassSymbol

auk% nm -g src/main/main.o | grep R_ClassSymbol
0000000000000008 C R_ClassSymbol

auk% nm -g library/tools/libs/tools.so  | grep R_ClassSymbol
                   U R_ClassSymbol

auk% nm -g src/library/tools/src/gramRd.o | grep R_ClassSymbol
                   U R_ClassSymbol

Interestintly, checking for R_ClassSymbol gives the same output as on
your working Linux system:

   ~/R-3.2.2$ nm -g bin/exec/R | grep R_ClassSymbol
   00000000008f8ff8 B R_ClassSymbol

   ~/R-3.2.2$ nm -g src/main/main.o | grep R_ClassSymbol
   0000000000000008 C R_ClassSymbol

   ~/R-3.2.2$ nm -g library/tools/libs/tools.so | grep R_ClassSymbol
                    U R_ClassSymbol

   ~/R-3.2.2$ nm -g src/library/tools/src/gramRd.o | grep R_ClassSymbol
                    U R_ClassSymbol

So R_ClassSymbol is unresolved in the tools package and should be
resolved by loading into the main R executable. On Linux that is
achieved by the linker flag

-Wl,--export-dynamic

as part of MAIN_LDFLAGS in Makeconf in the top-level directory.  We have
in configure.ac

    freebsd*)
      main_ldflags="-export-dynamic"
      shlib_ldflags="-shared"

but those were from the days when FreeBSD used gcc, and it is possible
that your use of clang[*] requires -Wl,--export-dynamic . If so, add to
config.site

MAIN_LDFLAGS="-Wl,--export-dynamic"

It would also be worth trying a build with --enable-R-shlib, as that
resolves R_ClassSymbol and similar differently.

I tried 'configure --enable-R-shlib', still the same error.

I then tried adding MAIN_LDFLAGS as you suggested, and the install
worked. Thanks you very much! ("make check" fails in datetime.R, but
that's something I'll follow up in a separate email.)

Should configure.ac be changed to account for this on FreeBSD's using
clang? (I should probably also try compiling with GCC, which I had to
install anyways for gfortran.)

It is more complicated than that, in the absence of any definitive FreeBSD documentation. There are five possibilities:

-export-dynamic
-rdynamic
-Wl,--export-dynamic
-Wl,-export-dynamic
-Wl,-E

The first two work for GCC (and have for a long time, although only the second is currently documented) and are accepted by compilers claiming to be GCC-compliant (clang and icc). That clang does nothing with the first seems a clear bug in clang (at least on some OSes (which from the sources do include FreeBSD) it does support the second and maps it to the fourth). Not least as a compiler called 'gcc' may not be GCC-compliant (that on OS X is based on clang).

It seems the GNU linker supports all -WL forms (but only the first and last are documented): some other linkers require the fourth line. And that means that GCC's -rdynamic on ELF platforms is documented to generate a linker flag, -export-dynamic, that the GNU linker is not documented to support.

If FreeBSD continues to use ELF and GNU ld, -Wl,--export-dynamic seems the safest choice, but if they change linker (and there are/will be alternatives such as gold and lld) -rdynamic might be more future-proof.

It seems R has very few FreeBSD users. The core team cannot be expected to support all minority platforms and we rely on the OS teams to inform us what is required. Now it seems that there *is* a FreeBSD port for R: https://www.freebsd.org/ports/math.html albeit at R 3.0.2 but updated for texinfo 6.0 a couple of months ago, so that is the place to get needed changes recorded.

--
Brian D. Ripley,                  rip...@stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
1 South Parks Road, Oxford OX1 3TG, UK

______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

Reply via email to