[MITgcm-devel] testing with mpi

[Jean-Michel Campin]

Hello Ed,

I have a suggestion regarding testreport and the testing with mpi:
generally (at least on myrinet cluster) the run always return 
a zero output code (RETVAL=0), as if everything was OK, even when 
the run stop immediately.

It means that in testreport (arround line 430):
>        ( eval $COMMAND ) > run.log 2>&1
>        RETVAL=$?
>        if test "x$RETVAL" = x0 ; then
>            tail run.log
>            echo successful 1>&2
will not necessary catch a failure in running the model.

And in fact, it's what happens with fizhi-gridalt-hs in all
recent mpi testing on cg01.

Could we save also STDERR.000? (in $CDIR, in addition to Makefile_head 
and summary.txt) so that we would have some idea about what's wrong. 
And the advantage is that those files (STDERR.000?) are generally 
very small (most of the time empty) so that it will not require 
too much disk space.

Thanks,

Jean-Michel