it turns out gfortan is innocent. metview uses the magics++
package, which caches the gfortran libraries in a cmake file
fragment used by metview.
I'm also the magics++ maintainer, and working on a solution now
metview doesn't seem to use a hard-coded version of gfortran
but picks up on the error somewhere in the bowels of its CMake
code.
When built, the tools GeoTool binary uses the usual
/usr/lib/$ARCH/libgfortran.so.3 link. During building, it seems
to pick up the /usr/lib/gcc/* version as seen in the report. It
does this too on other archs, but there the '5' version is
correct and all builds fine. metview had a dependency on
gfortran but not libgfortran.
So, it appears the bug is in gfortran somewhere but I'm not sure
yet what metview / cmake did to tickle it. I have it in a schroot
on fischer and can investigate further.
regards
Alastair
--
Alastair McKinstry, <alastair.mckinstry@ichec.ie>, t: +353 91 495343
Environmental Sciences Activity Leader, ICHEC, IT 301, NUI Galway, Ireland