Use of uninitialized parameter can lead to use of uninitialized array in MOM_thickness_diffuse (possible segfault)

[pittwolfe]

The parameter thickness_diffuse_CS@Use_KH_in_MEKE is given a value only if use_meke is true:

Lines 2385-2388 in MOM_thickness_diffuse.F90

  if (use_meke) then
    call get_param(param_file, mdl, "USE_KH_IN_MEKE", CS%Use_KH_in_MEKE, &
                   "If true, uses the thickness diffusivity calculated here to diffuse MEKE.", &
                   default=.false.)

However, Use_KH_in_MEKE is later tested in a way that doens't depend on the value of use_meke:

Line 553

  if (CS%id_KH_t > 0 .or. CS%id_KH_t1 > 0 .or. CS%Use_KH_in_MEKE) then

and lines 581-582

      if (CS%Use_KH_in_MEKE) then
        MEKE%Kh_diff(:,:) = 0.0

This last instance can lead to a segmentation fault since MEKE%Kh_diff is only allocated if use_meke is true. Note that lines 581-582 are only reached if the user requests diffusivity diagnostics (so line 553 evaluates to true), but I don't see any reason why diffusivity diagnostics should require MEKE.

The actual outcome seems to depend on the compiler, which is probably due to how they give uninitialized logical variables values. I get a crash due to segfault with the GNU compiler, but not with the Intel compiler.

The simplest solution would be to move the get_param call above the if (use_meke) then test. I could do a pull request with this change, but it'd probably be faster to have a more active developer do this than for me to try to figure out the respository & pull request policies.

[Hallberg-NOAA]

Thank you for drawing attention to this bug. It will be addressed when NOAA-GFDL#1099 is merged into the main branch of MOM6.