"Why does my libgfortran build yield an error on kinds.h?"

[FX Coudert <fxcoudert at gmail dot com>]

Hi Bernhard, hi all,

[ I first wrote this message to Bernhard, and thought it was a nice  
summary of the cases where we get an error while building libgfortran/ 
kinds.h, so I send it to the list, with hopefully enough context for  
everyone to understand ]

I read the busybox thread you mentionned to me (http://busybox.net/ 
bugs/view.php?id=1164). I'm not sure this is a problem with  
autodetecting kinds... in fact, what I can say for sure is that every  
single time this exact error was reported by someone, it was simply a  
problem of the newly built gfortran compiler not working at all. The  
most common reasons for that have been:

  - the f951 binary being linked to the shared mpfr library (or  
gmp), and LD_LIBRARY_PATH not being set so that running the compiler  
simply errors out because of the shared library not being found;

  - the f951 binary being linked to GMP and MPFR libraries built for  
a different architecture than the *host* architecture (ie the one the  
compiler is executed on): including, but not limited to, 64bit GMP  
and/or MPFR libraries used in a 32bit compiler binary;

  - the f951 binary being unable to run on the host, e.g.  
segfaulting every time it's called, because it's compiled with a  
bizarre -march value that doesn't match the host architecture.

I've recently (yesterday) commited a patch that explcitly checks if  
the compiler is working and gives a better error message. Right now,  
the best way for you to determine what's going wrong is to use the  
compiler indicated in the mk-kinds-h.sh command line (the failing  
one) to build a small test program. For example, when the failing  
command-line is:

> /bin/sh /root/buildroot/toolchain_build_i686/gcc-4.2-20070110/ 
> libgfortran/mk-kinds-h.sh '/root/buildroot/toolchain_build_i686/ 
> gcc-4.2-final/./gcc/gfortran -B/root/buildroot/toolchain_build_i686/ 
> gcc-4.2-final/./gcc/ -B/root/buildroot/build_i686/staging_dir/i686- 
> linux-uclibc/bin/ -B/root/buildroot/build_i686/staging_dir/i686- 
> linux-uclibc/lib/ -isystem /root/buildroot/build_i686/staging_dir/ 
> i686-linux-uclibc/include -isystem /root/buildroot/build_i686/ 
> staging_dir/i686-linux-uclibc/sys-include -I . -Wall -fno-repack- 
> arrays -fno-underscoring ' > kinds.h || rm kinds.h

the compiler command is the first arg of mk-kinds-h.sh, that is:

> /root/buildroot/toolchain_build_i686/gcc-4.2-final/./gcc/gfortran - 
> B/root/buildroot/toolchain_build_i686/gcc-4.2-final/./gcc/ -B/root/ 
> buildroot/build_i686/staging_dir/i686-linux-uclibc/bin/ -B/root/ 
> buildroot/build_i686/staging_dir/i686-linux-uclibc/lib/ -isystem / 
> root/buildroot/build_i686/staging_dir/i686-linux-uclibc/include - 
> isystem /root/buildroot/build_i686/staging_dir/i686-linux-uclibc/ 
> sys-include -I . -Wall -fno-repack-arrays -fno-underscoring

Create a small test.f90 file, containing simply
  integer :: i
  end

and try compiling it ("large-command-line -c test.f90") and see what  
the compiler says. If you have a version of gfortran more recent than  
my patch, it gives the name of log file that you can look into: it  
will contain the error message from the compiler.

Now, I hope that we can refer people to this post in the future!

FX