Fixes for issues #129, #135

[sholly]

This pull request contains fixes for cime issues #129, #135.

[quantheory]

Can't we remove the block above? If we enter that block, $attrMatch will always be 1, so we shouldn't need to change it. Also, if the conditional above succeeds, then the one below will fail and we will just move on to the next iteration, so we shouldn't need a next, unless I'm missing something.

[quantheory]

Have you tested this fix for #135 on yellowstone, and checked that threaded runs really are using hybrid_launch instead of launch? I believe that there's an additional issue where the hybrid_launch line in config_machines.xml is broken. Currently, the line is:

MP_PE_AFFINITY= MP_TASK_AFFINITY= MP_CPU_BIND_LIST= mpirun.lsf $ENV{'CESMDATAROOT'}/tools/bin/hybrid_launch 

However, when I tried that, it seemed to be broken, because we actually want to guarantee that those variables are unset, not that they are set to the null string. I think the line should be one of these instead:

unset MP_PE_AFFINITY; unset MP_TASK_AFFINITY; unset MP_CPU_BIND_LIST; mpirun.lsf $ENV{'CESMDATAROOT'}/tools/bin/hybrid_launch

# Or this may be better:
unset MP_PE_AFFINITY && unset MP_TASK_AFFINITY && unset MP_CPU_BIND_LIST && mpirun.lsf $ENV{'CESMDATAROOT'}/tools/bin/hybrid_launch

Aside from this line being wrong, and the weird (unnecessary?) block I already commented on, it looks to me like this will work.

[sholly]

Re #135, I did figure out that this does not work: 'MP_PE_AFFINITY= MP_TASK_AFFINITY= MP_CPU_BIND_LIST=' doesn't work. I actually implemented your first suggestion just as you said:
unset MP_PE_AFFINITY; unset MP_TASK_AFFINITY; unset MP_CPU_BIND_LIST;
and this works. This will get easier when my new module implementation goes in, we'll be able to conditionally set environment variables based on the compiler, threaded vs non-threaded, etc.

[quantheory]

Oh yeah, you're right, that change is right here in this PR. I don't know why I didn't see it...

[sholly]

The spurious for loop has been removed...