Fixes an integer-kind mismatch in MOM_random, seed_from_time()

[adcroft]

• gcc/8.3.0 issued Error: Integer too big for its kind reported in
  feedback on PR dev/master candidate 2020-05-15 #1111. The intent was to assume kind=4 in these
  calculations but apparently our compilers were promoting
  mod(dy + 32*(mo + 13*yr), 2147483648) to kind=8. There were two
  mistakes in the expression:
  • the use of 2147483648 in the mod is not representable with kind=4;
  • the mod produces negative values and should have been a modulo.
• This commit reduces the range of the results by one number on the
  positive side and removes all the negatives.

@gustavo-marques Could you test this patch for me? It passes everything at our end but so did the original PR.

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (dev-master-candidate-2020-05-15@8f57abf). Click here to learn what that means.
The diff coverage is n/a.

@@                        Coverage Diff                         @@
##             dev-master-candidate-2020-05-15    #1113   +/-   ##
==================================================================
  Coverage                                   ?   46.07%           
==================================================================
  Files                                      ?      214           
  Lines                                      ?    69379           
  Branches                                   ?        0           
==================================================================
  Hits                                       ?    31964           
  Misses                                     ?    37415           
  Partials                                   ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8f57abf...fab21a7. Read the comment docs.

[marshallward]

I know this isn't the only place where this is being violated, but we really ought to stop assuming that kind is equal to the number of bytes. There is at least one major compiler (NAG) which does not make this assumption.

I'd recommend using iso_fortran_env, only : int4.

[gustavo-marques]

I can confirm it builds with gnu/8.3.0.

[klindsay28]

@marshallward , it looks like int4 was introduced in Fortran 2008.

If you want to support earlier Fortran versions, you can use i4 = selected_int_kind(6). This is what CIME and CESM components use. selected_int_kind was introduced in Fortran 90.

Does the MOM6 development team have a position on minimal Fortran version required for model development? I don't see this at https://mom6.readthedocs.io/en/dev-gfdl/index.html or https://github.com/NOAA-GFDL/MOM6/wiki.

[marshallward]

@klindsay28 I recall last we discussed this, we found FMS had already moved onto 2008, and so we've opted to do the same. (We're already using iso_fortran_env types elsewhere in the code).

I agree we need to clarify our target standard somewhere in the docs.