Re: f2c obsolete?

[Thomas Treichl]

John W. Eaton schrieb:
> On 24-Oct-2007, Michael Goffioul wrote:
>
> | On 10/24/07, John W. Eaton <address@hidden> wrote:
> | > On 24-Oct-2007, Michael Goffioul wrote:
> | >
> | > | This is something that can also be handled in fort77 itself. In the end,
> | > | the libf2c I'm using needs that -MD flag, so it make sense to add it
> | > | directly in fort77 script. For me it's the same, as long as it works...
> | >
> | > How does it need it?  In what command?
> | 
> | The -MD flag tells the MS compiler about the C runtime library to use
> | (in this case, the DLL version). When you compile different modules and
> | try to link them together, you should always use the same runtime lib
> | for compiling all modules to avoid weird problems (it's even needed if you
> | want to share FILE* streams between modules).
> | 
> | I always use the -MD flag to compile everything (octave and all deps),
> | including libf2c. Because the code generated by fort77 is intended to
> | be linked with libf2c, fort77 should also use by default the -MD flag.
>
> So is -MD supposed to be passed to the C compiler or the linker?
> Either way, it looks like you should be able to use -Wc,-MD or -Wl,-MD
> to send the options through fort77 without having to modify the fort77
> script itself.
>
> If that works, then the only modification that I see that should be
> necessary is the one I posted earlier to make -v print actual verbose
> output from the compilers so that fort77 will play nice with the
> AC_F77_LIBRARY_LDFLAGS autoconf macro.
>
> jwe

Hi,

I'm late (as always) in replying to this thread. A first setup of the fort77 
program did work for me (with modifications - see below) and it also worked to 
rebuild a new version of Octave.app (that is compiled like Michael's MSVC 
version with f2c). What I needed to do is to comment out the "BLAS library 
calling convention compatibility" test result in the configure.in script because 
the test fails for me. See yourself what happens (the -framework vecLib is found 
and needed, this is okay - but the '-framework vecLib' seems to not be passed 
through correctly through the configure test).

I changed the script the following way to get some more output:

line 886:
  if $have_fortran_compiler; then
    LIBS="$BLAS_LIBS $FLIBS $LIBS"
    AC_LANG_PUSH(Fortran 77)
  echo "--->"
  echo "-1-> $ac_compile"
  echo "-2-> $F77"
  echo "-3-> $FFLAGS"
  echo "-4-> $ac_ext"
  echo "--->"
  echo "-5-> $ac_link"
  echo "-6-> $ac_exeext"
  echo "-7-> $FFLAGS"
  echo "-8-> $LDFLAGS"
  echo "-9-> $ac_ext"
  echo "-0-> $LIBS"
    eval "$ac_compile"
    eval "$ac_link"
    AC_LANG_POP(Fortran 77)
  elif $have_f2c; then

And the output then is:

  --->
  -1-> $F77 -c $FFLAGS conftest.$ac_ext >&5
  -2-> fort77
  -3-> -O3 -ftree-vectorize
    -I/Applications/Octave.app/Contents/Resources/include -mieee-fp -mieee-fp
  -4-> f
  --->
  -5-> $F77 -o conftest$ac_exeext $FFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5
  -6->
  -7-> -O3 -ftree-vectorize
    -I/Applications/Octave.app/Contents/Resources/include -mieee-fp -mieee-fp
  -8-> -L/Applications/Octave.app/Contents/Resources/lib
    -L/Applications/Octave.app/Contents/Resources/lib/pkgconfig
  -9-> f
  -0->  -framework vecLib -L/Applications/Octave.app/Contents/Resources/lib
    -lf2c -lhdf5 -lz -lm
     MAIN foo:
     MAIN foo:
  /usr/bin/ld: can't locate framework for: -framework -c
  collect2: ld returned 1 exit status
  /Applications/Octave.app/Contents/Resources/bin/fort77: aborting compilation
  ./configure: line 1: ./conftest: No such file or directory
  no
  configure: WARNING: Your BLAS library was apparently compiled with a Fortran
  configure: WARNING: compiler that uses a different calling convention from
  configure: WARNING: the one used by the selected compiler, fort77.

Any ideas what I can do now?

  Thomas