Hi,
On 11/16/2011 06:23 PM, Alfredo Ferrari wrote:
> no simple solution. One possibility is for the user to upgrade their
> gfortran compiler to 4.6.x (very uncomfortable to say the least), or for
> us to build a version on some remaining machines still with gfortran 4.5.1
> and provide two versions if required (maybe the one compiled on 4.5.x
> links on 4.6.x). It is clear that this kind of incompatible changes
> between compiler sub-versions if confirmed are a major pain...
It would first be good to mention it on the documentation, wich only
says "gfortran >=3D 4.4 as I spent quite some time yesterday wondering
what I may have done wrong (I'm new to Fluka).
If the dependency is to the version of the compiler, the tar.gz should mention it in its name
(something like fluka-2011.2.7-gfortran4.6.tar.gz so that you may publish several ones
and us being able to differentiate them. This is also an occasion to
mention that on the download page, it looks like an archive can silently
change it's content without changing its name (as the latest '.7' is not
part of the name) which also doesn't seems good policy. The fact that
that archive deploy itself in the current directory (and not in a subdir) is also something usually considered bad.
> I would like also to warn all users running with gfortran 4.4.x
> (eg those on Scientific Linux) that we have evidence that gfortran 4.4.x
> can miscompile Fluka routines. It is not an issue if you only link the
Many thanks for this advice. So for SL5 users like me, recompiling a
4.6.x gfortran seems to be the way to go anyway.
Yannick
Received on Wed Nov 16 2011 - 21:43:37 CET
This archive was generated by hypermail 2.2.0 : Wed Nov 16 2011 - 21:43:38 CET