octave-control configure fail

[Michael Dickens]

Yes, there would be no difference so long as gcc5+ uses libgcc or
libgcc-devel (which they seem to do). The change actually works pretty
well in my testing; in the pre-configure stage from octave-1.0.tcl,
change the current "configure.env-append" to instead do:
{{{
    set GFORTRAN [glob ${prefix}/lib/libgcc/libgfortran*]
    set QUADMATH [glob ${prefix}/lib/libgcc/libquadmath*]
    set FLIBS [exec ${prefix}/bin/mkoctfile -p FLIBS | sed -e "s at -L\[^
    \]* @@g" -e "s at -lgfortran@${GFORTRAN}@g" -e
    "s at -lquadmath@${QUADMATH}@g"]
    set FLIBS "-L${prefix}/lib/libgcc ${FLIBS}"
    set LAPACK_LIBS [exec ${prefix}/bin/mkoctfile -p LAPACK_LIBS | sed
    -e "s at -L\[^ \]* @@g" -e "s at -lgfortran@${GFORTRAN}@g" -e
    "s at -lquadmath@${QUADMATH}@g"]
    set LAPACK_LIBS "${FLIBS} ${LAPACK_LIBS}"
    configure.env-append \
       FLIBS='${FLIBS}' \
       LAPACK_LIBS='${LAPACK_LIBS}'
}}}

I'm not sure the above is the -best- way to go, but it works in my
testing. mkoctfile allows one to overload its settings via environment
variables, so I just retrieve the current settings and tweak them to use
libgcc files instead of going through any specific gcc port's files.

Anybody want to test? - MLD

On Tue, Jul 7, 2015, at 03:18 AM, Ryan Schmidt wrote:
> On Jul 5, 2015, at 12:50 PM, Michael Dickens wrote:
> > I've been working on a fix to this issue, but haven't finalized it yet.
> > The problem is that mkoctfile hard-codes -L and -I paths, when what it
> > should actually do is glean them at runtime just before compiling or
> > linking. My way around this is to override the settings in mkoctfile by
> > those determined from the active libgcc port. Since all of the gcc4*
> > ports link with the libraries provided by libgcc, we should not have to
> > rev-bump octave* ports for updates to gcc4*.
> 
> And there would presumably be no difference for gcc5 or gcc6 or later
> either.
> 
>