🐛

Describe the bug

1. List the steps someone needs to take to reproduce the bug.
   run filter MITgcm_ocean and perturb from a single copy

2. What was the expected outcome?

pertval depends on the which variable is being perturbed (some are clamped).

 ! Clamping: Samples obtained from truncated Gaussian dist. 
 if (.not. log_transform) pertval = max(clamp_min_val, pertval)

5. What actually happened?

The loop is assuming that a variable is contiguous in %copies.

- start_ind = get_index_start(domain_id, ivar)
- end_ind   = get_index_end(domain_id, ivar)

INDICES : do i = start_ind, end_ind
        MEMBERS : do copy = 1, ens_size

         ! Only perturb the actual ocean cells;
         ! Leave the land and ocean floor values alone.
         if( state_ens_handle%copies(copy, i) == FVAL ) cycle MEMBERS

         pertval = random_gaussian(random_seq, state_ens_handle%copies(copy, i), &
                                         model_perturbation_amplitude)

         ! Clamping: Samples obtained from truncated Gaussian dist. 
-         if (.not. log_transform) pertval = max(clamp_min_val, pertval)

         state_ens_handle%copies(copy, i) = pertval

      enddo MEMBERS
   enddo INDICES

get_index_start is for %vars not %copies. %copies is round robin across processors.
so if (.not. log_transform) pertval = max(clamp_min_val, pertval) is not necessary the variable you want.

Error Message

You can get i > %copies(copy, i) and segfault.

Which model(s) are you working with?

MITgcm_ocean

Version of DART

Which version of DART are you using?
v10.0.1

Have you modified the DART code?

No

Build information

Cheyenne, intel but applies to any machine.