Downsampled parallel output may write full-resolution data when bubbles_euler is enabled

[sbryngelson]

Summary

In s_write_parallel_data_files (src/simulation/m_data_output.fpp), the file_per_process + down_sample path appears to write full-resolution field data using a downsampled element count when bubbles_euler is enabled. The non-bubbles path in the same if handles downsampling correctly, so the two branches disagree.

This is pre-existing on master (not introduced by any open PR). Filing to investigate rather than asserting a confirmed data-corruption bug, since I have not been able to execute the affected combination.

Detail

Inside if (file_per_process):

1. data_size is set from the downsampled extents when down_sample is on:

if (down_sample) then
    data_size = (m_ds + 3)*(n_ds + 3)*(p_ds + 3)
else
    data_size = (m + 1)*(n + 1)*(p + 1)
end if

2. MPI_IO_DATA%var is bound unconditionally to the full-size conservative fields:

call s_initialize_mpi_data(q_cons_vf, qbmm_pb=pb_ts(1), qbmm_mv=mv_ts(1))
! binds MPI_IO_DATA%var(i)%sf => q_cons_vf(i)%sf(0:m, 0:n, 0:p)

3. The write then diverges by branch:

if (bubbles_euler) then
    do i = 1, sys_size
        ! no down_sample case: full-size source, downsampled count
        call MPI_FILE_WRITE_ALL(ifile, MPI_IO_DATA%var(i)%sf, data_size*mpi_io_type, mpi_io_p, status, ierr)
    end do
    ...
else
    if (down_sample) then
        do i = 1, sys_size
            ! correct: writes the downsampled buffer
            call MPI_FILE_WRITE_ALL(ifile, q_cons_temp_ds(i)%sf, data_size*mpi_io_type, mpi_io_p, status, ierr)
        end do
    else
        ...
    end if
end if

The else branch switches its source array to q_cons_temp_ds when downsampling. The bubbles_euler branch does not, so with down_sample = T it writes the first (m_ds+3)*(n_ds+3)*(p_ds+3) elements of the full-resolution array — a contiguous prefix of the fine grid rather than the coarsened field.

The same concern applies to the qbmm .and. .not. polytropic sub-loop immediately below, which also writes MPI_IO_DATA%var with the downsampled data_size.

Reachability

The combination does not appear to be exercised:

• examples/3D_IGR_33jet sets file_per_process = T and down_sample = T, but is IGR (no bubbles_euler), so it takes the correct else branch.
• examples/3D_IGR_jet_1fluid sets file_per_process = T, down_sample = F.

I did not find a case combining bubbles_euler = T with down_sample = T and file_per_process = T.

Questions

1. Is bubbles_euler + down_sample + file_per_process a supported combination? If not, a @:PROHIBIT in the checker would be clearer than the current silent divergence.
2. If supported, should the bubbles_euler branch select q_cons_temp_ds under down_sample, mirroring the else branch?
3. The two ! TODO: check if sys_size is correct comments on the surrounding loops suggest prior uncertainty about this routine and may be related.

Notes

Found while reviewing #1679; the region is byte-identical to master, so it is independent of that PR.