Re: Gfortran does not compile dynamo (and others)

[Tobias Burnus <burnus at net-b dot de>]

Hi all,

Ignacio Fernández Galván wrote:
> By the way, I also could not compile the quantum chemistry programs
> Gaussian 98 and Molcas 6.4, but since these are commercial programs, I
> cannot provide the source. Has anyone been able to compile them?
>
> Actually, the compiliation seems to go fine (after some changes in code
> and building scripts), but when running the program, Molcas stops with
> "signal 11"
I did some debugging and the reason is that Molcas sets by default the
option
   -malign-double
on i386 systems.

And of cause this clashes with libgfortran. The crash occurs with an
internal write statement.

I think we really need a big WARNING somewhere. The result (crash) is
not obvious to find. (Unless, one has read the manpage; but if a
pre-setup package has this option, one does not easily knows the side
effects of this option.)
I think we had recently at least two bug reports about the same problem
(PR 29562, 30594; resolved invalid).


Solution:

- Removing the -malign options, i.e. changing in cfg/gf.comp

   i386FLAGS="-malign-double"
to
   i386FLAGS=""

- Additionally, one needs to add the "-fno-range-check" option
for the file "src/util/pack.f", e.g. by editing "src/util/Makefile"


Tobias

PS: Someone interested in debugging Gaussian?