Fix history/restart frequency bugs and update scripts

- Fix bugs in history/restart frequency associated with new calendar (CICE-Consortium#589) - Define frequency in absolute terms relative to 0000-01-01-00000 and document (CICE-Consortium#589) - Update set_nml.histall to include hourly output (CICE-Consortium#589) - Update test scripts to cleanly abort if run fails where possible (CICE-Consortium#608) - Update decomp test so it's rerunable, remove restart at start of run (CICE-Consortium#601) - Add ability to do bfbcomp tests with additional options set on command line (CICE-Consortium#569) - Update documentation of calendar frequency computation, calendar types, and closed boundaries (CICE-Consortium#541) - Add optional doabort flag to abort_ice to control whether the method aborts. This is useful for testing and code coverage statistics, although doabort=.false. will not call the actual abort method, but we can test the interfaces and rest of the code.
apcraig · Jun 18, 2021 · 92fa331 · 92fa331
1 parent ff88891
commit 92fa331
Show file tree

Hide file tree

Showing 12 changed files with 190 additions and 56 deletions.
diff --git a/cice.setup b/cice.setup
@@ -644,11 +644,6 @@ EOF
       set bfbcomp = "$bfbcomp_tmp"
     endif
 
-    set fbfbcomp = ${spval}
-    if ($bfbcomp != ${spval}) then
-      set fbfbcomp = ${machcomp}_${bfbcomp}
-    endif
-
     #------------------------------------------------------------
     # Parse pesx with strict checking, limit pes for machine
 
@@ -768,6 +763,29 @@ EOF
       set testname_base = "${machcomp}_${test}_${grid}_${pesx}${soptions}.${testid}"
       set testname = "${tsdir}/${testname_base}"
       set case = ${testname}
+
+      # Add -s flags in cice.setup to bfbcomp name
+      # Parse bfbcomp test_grid_pes and sets
+      # Add sets_base and sort unique
+      # Create fbfbcomp string that should be consistent with base casename
+      set bfbcomp_regex="\(.*_[0-9x]*\)_\(.*\)"
+      set bfbcomp_test_grid_pes=`echo ${bfbcomp} | sed "s/${bfbcomp_regex}/\1/"`
+      set bfbcomp_sets=`echo ${bfbcomp} | sed "s/${bfbcomp_regex}/\2/" | sed 's/_/,/g' `
+      set bfbcomp_sets="${bfbcomp_sets},${sets_base}"
+      set bfbcomp_soptions = ""
+      # Create sorted array and remove duplicates and "none"
+      set bfbcomp_setsarray = `echo ${bfbcomp_sets} | sed 's/,/ /g' | fmt -1 | sort -u`
+      if ("${bfbcomp_setsarray}" != "") then
+        foreach field (${bfbcomp_setsarray})
+          if (${field} != "none") then
+            set bfbcomp_soptions = ${bfbcomp_soptions}"_"${field}
+          endif
+        end
+      endif
+      set fbfbcomp = ${spval}
+      if ($bfbcomp != ${spval}) then
+        set fbfbcomp = ${machcomp}_${bfbcomp_test_grid_pes}${bfbcomp_soptions}
+      endif
     endif
 
     if (-d ${case}) then

diff --git a/cicecore/cicedynB/infrastructure/comm/mpi/ice_exit.F90 b/cicecore/cicedynB/infrastructure/comm/mpi/ice_exit.F90
@@ -19,7 +19,7 @@ module ice_exit
 
 !=======================================================================
 
-      subroutine abort_ice(error_message, file, line)
+      subroutine abort_ice(error_message, file, line, doabort)
 
 !  This routine aborts the ice model and prints an error message.
 
@@ -31,10 +31,10 @@ subroutine abort_ice(error_message, file, line)
       use mpi   ! MPI Fortran module
 #endif
 
-      character (len=*), intent(in),optional :: error_message
-      character (len=*), intent(in),optional :: file
-      integer (kind=int_kind), intent(in), optional :: &
-         line       ! line number
+      character (len=*), intent(in),optional :: error_message  ! error message
+      character (len=*), intent(in),optional :: file           ! file
+      integer (kind=int_kind), intent(in), optional :: line    ! line number
+      logical (kind=log_kind), intent(in), optional :: doabort ! abort flag
 
       ! local variables
 
@@ -43,8 +43,12 @@ subroutine abort_ice(error_message, file, line)
          ierr,       & ! MPI error flag
          error_code    ! return code
 #endif
+      logical (log_kind) :: ldoabort   ! local doabort flag
       character(len=*), parameter :: subname='(abort_ice)'
 
+      ldoabort = .true.
+      if (present(doabort)) ldoabort = doabort
+
 #if (defined CESMCOUPLED)
       call flush_fileunit(nu_diag)
       call icepack_warnings_flush(nu_diag)
@@ -54,7 +58,7 @@ subroutine abort_ice(error_message, file, line)
       if (present(line))   write (nu_diag,*) subname,' line number ',line
       if (present(error_message)) write (nu_diag,*) subname,' error = ',trim(error_message)
       call flush_fileunit(nu_diag)
-      call shr_sys_abort(subname//trim(error_message))
+      if (ldoabort) call shr_sys_abort(subname//trim(error_message))
 #else
       call flush_fileunit(nu_diag)
       call icepack_warnings_flush(nu_diag)
@@ -65,8 +69,10 @@ subroutine abort_ice(error_message, file, line)
       if (present(error_message)) write (ice_stderr,*) subname,' error = ',trim(error_message)
       call flush_fileunit(ice_stderr)
       error_code = 128
-      call MPI_ABORT(MPI_COMM_WORLD, error_code, ierr)
-      stop
+      if (ldoabort) then
+         call MPI_ABORT(MPI_COMM_WORLD, error_code, ierr)
+         stop
+      endif
 #endif
 
       end subroutine abort_ice

diff --git a/cicecore/cicedynB/infrastructure/comm/serial/ice_exit.F90 b/cicecore/cicedynB/infrastructure/comm/serial/ice_exit.F90
@@ -24,25 +24,29 @@ module ice_exit
 
 !=======================================================================
 
-      subroutine abort_ice(error_message,file,line)
+      subroutine abort_ice(error_message,file,line,doabort)
 
 !  This routine aborts the ice model and prints an error message.
 
-      character (len=*), intent(in),optional :: error_message
-      character (len=*), intent(in),optional :: file
-      integer (kind=int_kind), intent(in), optional :: &
-         line       ! line number
+      character (len=*), intent(in),optional :: error_message  ! error message
+      character (len=*), intent(in),optional :: file           ! file
+      integer (kind=int_kind), intent(in), optional :: line    ! line number
+      logical (kind=log_kind), intent(in), optional :: doabort ! abort flag
 
+      logical (kind=log_kind) :: ldoabort   ! local doabort
       character(len=*), parameter :: subname='(abort_ice)'
 
+      ldoabort = .true.
+      if (present(doabort)) ldoabort = doabort
+
 #ifdef CESMCOUPLED
       call icepack_warnings_flush(nu_diag)
       write(nu_diag,*) ' '
       write(nu_diag,*) subname, 'ABORTED: '
       if (present(file))   write (nu_diag,*) subname,' called from ',trim(file)
       if (present(line))   write (nu_diag,*) subname,' line number ',line
       if (present(error_message)) write (nu_diag,*) subname,' error = ',trim(error_message)
-      call shr_sys_abort(subname//trim(error_message))
+      if (ldoabort) call shr_sys_abort(subname//trim(error_message))
 #else
       call icepack_warnings_flush(nu_diag)
       write(nu_diag,*) ' '
@@ -51,7 +55,7 @@ subroutine abort_ice(error_message,file,line)
       if (present(line))   write (nu_diag,*) subname,' line number ',line
       if (present(error_message)) write (nu_diag,*) subname,' error = ',trim(error_message)
       call flush_fileunit(nu_diag)
-      stop
+      if (ldoabort) stop
 #endif
 
       end subroutine abort_ice

diff --git a/cicecore/drivers/unittest/bcstchk/bcstchk.F90 b/cicecore/drivers/unittest/bcstchk/bcstchk.F90
@@ -272,11 +272,20 @@ program bcstchk
       endif
 
       ! Test abort_ice, regardless of test outcome
+      ! Set doabort to false to support code coverage stats, aborted runs don't seem to generate
+      ! gcov statistics
+
+      call flush_fileunit(6)
+      call ice_barrier()
+      call abort_ice(subname//' Test abort ',file=__FILE__,line=__LINE__, doabort=.false.)
       call flush_fileunit(6)
       call ice_barrier()
-      call abort_ice(subname//' Test abort ',file=__FILE__,line=__LINE__)
 
-      if (my_task == master_task) write(6,*) subname,'This line should not be written'
+      if (my_task == master_task) then
+         write(6,*) ' '
+         write(6,*) 'BCSTCHK abort test done'
+         write(6,*) ' '
+      endif
 
       call end_run()
 

diff --git a/cicecore/shared/ice_calendar.F90 b/cicecore/shared/ice_calendar.F90
@@ -11,6 +11,10 @@
 ! 2010 CM : Fixed support for Gregorian calendar: subroutines
 !           sec2time, time2sec and set_calendar added.
 ! 2020 TC : Significant refactor to move away from time as prognostic
+!           Note that the reference date is arbitrarily set to
+!           0000-01-01-00000 and dates cannot be less than that.
+!           The implementation is also limited by some integer
+!           math to myear_max which is a parameter in this module.
 
       module ice_calendar
 
@@ -130,6 +134,9 @@ module ice_calendar
       integer (kind=int_kind) :: &
          hour         ! hour of the day
 
+      integer (kind=int_kind), parameter :: &
+         myear_max = 200000           ! maximum year, limited by integer overflow in elapsed_horus
+
       ! 360-day year data
       integer (kind=int_kind) :: &
          daymo360(months_per_year)    ! number of days in each month
@@ -320,6 +327,20 @@ end subroutine advance_timestep
 
       subroutine calendar()
 
+! This sets a bunch of internal calendar stuff including history and
+! restart frequencies.  These frequencies are relative to the start
+! of time which is arbitrarily set to year=0, month=1, day=1, sec=0.
+! That means that the frequencies are repeatable between runs
+! regardless of the initial model date.
+! One thing to watch for is the size of elapsed hours.  This will
+! become a large integer and will overflow if the year is ever
+! greater than about 200,000 years.  A check has been added just
+! to make sure this doesn't happen.
+! The largest integer*4 is about 2^31 = 2*2^10^3 =~ 2*1000^3 = 2.e9
+! 2.e9 (hours) / (365*24 hours/year) =~ 228,000 years
+! The elapsed hours will overflow integers at some point after that.
+
+
 !      real (kind=dbl_kind), intent(in), optional :: &
 !         ttime                          ! time variable
 
@@ -328,9 +349,10 @@ subroutine calendar()
       integer (kind=int_kind) :: &
          ns                         , & ! loop index
          yearp,monthp,dayp,hourp    , & ! previous year, month, day, hour
-         elapsed_days               , & ! since beginning this run
-         elapsed_months             , & ! since beginning this run
-         elapsed_hours                  ! since beginning this run
+         elapsed_years              , & ! since beginning of time
+         elapsed_months             , & ! since beginning of time
+         elapsed_days               , & ! since beginning of time
+         elapsed_hours                  ! since beginning of time
       character(len=*),parameter :: subname='(calendar)'
 
       yearp=myear
@@ -347,12 +369,18 @@ subroutine calendar()
       call update_date(myear,mmonth,mday,msec)
       call set_calendar(myear)
 
+      if (myear > myear_max) then
+         write(nu_diag,*) trim(subname),' ERROR year too large, ',myear,myear_max
+         call abort_ice(subname//'ERROR: model year too large')
+      endif
+
       idate = (myear)*10000 + mmonth*100 + mday ! date (yyyymmdd) 
       yday = daycal(mmonth) + mday            ! day of the year
-      hour = (msec+1)/(seconds_per_hour)
-      elapsed_months = (myear - year_init)*months_per_year + mmonth - month_init
-      elapsed_days = compute_days_between(year_init,month_init,day_init,myear,mmonth,mday)
-      elapsed_hours = elapsed_days * hours_per_day
+      hour = int(msec/seconds_per_hour)
+      elapsed_years = myear
+      elapsed_months =  myear * months_per_year + mmonth - 1
+      elapsed_days = compute_days_between(0,1,1,myear,mmonth,mday)
+      elapsed_hours = elapsed_days * hours_per_day + hour
       call calendar_date2time(myear,mmonth,mday,msec,timesecs)
 
       !--- compute other stuff
@@ -368,12 +396,19 @@ subroutine calendar()
 
       ! History writing flags
 
+!      if (my_task == master_task) then
+!         write(nu_diag,*) subname,' elap_y',elapsed_years
+!         write(nu_diag,*) subname,' elap_m',elapsed_months
+!         write(nu_diag,*) subname,' elap_d',elapsed_days
+!         write(nu_diag,*) subname,' elap_h',elapsed_hours
+!      endif
+
       do ns = 1, nstreams
 
          select case (histfreq(ns))
          case ("y", "Y")
             if (new_year  .and. histfreq_n(ns)/=0) then
-               if (mod(myear, histfreq_n(ns))==0) &
+               if (mod(elapsed_years, histfreq_n(ns))==0) &
                    write_history(ns) = .true.
             endif
          case ("m", "M")
@@ -404,7 +439,7 @@ subroutine calendar()
 
       select case (dumpfreq)
       case ("y", "Y")
-         if (new_year  .and. mod(myear, dumpfreq_n)==0) &
+         if (new_year  .and. mod(elapsed_years, dumpfreq_n)==0) &
             write_restart = 1
       case ("m", "M")
          if (new_month .and. mod(elapsed_months,dumpfreq_n)==0) &

diff --git a/configuration/scripts/options/set_nml.histall b/configuration/scripts/options/set_nml.histall
@@ -1,5 +1,5 @@
-    histfreq       = 'm','d','1','x','x'
-    histfreq_n     =  1,1,6,1,1
+    histfreq       = 'm','d','1','h','x'
+    histfreq_n     =  1,2,6,4,1
     write_ic       = .true.
     f_tmask        = .true.
     f_blkmask      = .true.
@@ -20,10 +20,10 @@
     f_VGRDa        = .true.
     f_bounds       = .true.
     f_CMIP         = 'm'
-    f_aice         = 'md1' 
-    f_hi           = '1dm'
+    f_aice         = 'md1h' 
+    f_hi           = 'h1dm'
     f_hs           = 'd1m' 
-    f_Tsfc         = 'md' 
+    f_Tsfc         = 'mdh' 
     f_sice         = 'md' 
     f_uvel         = 'md' 
     f_vvel         = 'dm' 

diff --git a/configuration/scripts/tests/test_decomp.script b/configuration/scripts/tests/test_decomp.script
@@ -19,6 +19,9 @@ end
 cp ice_in ice_in.0
 set base_dir = ""
 
+# clean up old restart dirs to support rerun
+rm -rf ${ICE_RUNDIR}/restart.*
+
 foreach decomp (${decomps})
   ${ICE_CASEDIR}/casescripts/parse_namelist.sh ice_in ${ICE_CASEDIR}/casescripts/set_nml.d${decomp}
   cp ice_in ice_in.${decomp}

diff --git a/configuration/scripts/tests/test_logbfb.script b/configuration/scripts/tests/test_logbfb.script
@@ -1,7 +1,6 @@
-
+# This is identical to a smoke test, but triggers bfbcompare with log files instead of restarts
 #----------------------------------------------------
 # Run the CICE model
-# This is identical to a smoke test, but triggers bfbcompare with log files instead of restarts
 # cice.run returns -1 if run did not complete successfully
 
 ./cice.run
@@ -21,9 +20,12 @@ mv -f ${ICE_CASEDIR}/test_output ${ICE_CASEDIR}/test_output.prev
 cat ${ICE_CASEDIR}/test_output.prev | grep -iv "${ICE_TESTNAME} test" >! ${ICE_CASEDIR}/test_output
 rm -f ${ICE_CASEDIR}/test_output.prev
 
-set grade = FAIL
-if ( $res == 0 ) then
-  set grade = PASS
+set grade = PASS
+if ( $res != 0 ) then
+  set grade = FAIL
+  echo "$grade ${ICE_TESTNAME} run ${ttimeloop} ${tdynamics} ${tcolumn}" >> ${ICE_CASEDIR}/test_output
+  echo "$grade ${ICE_TESTNAME} test " >> ${ICE_CASEDIR}/test_output
+  exit 99
 endif
 
 echo "$grade ${ICE_TESTNAME} run ${ttimeloop} ${tdynamics} ${tcolumn}" >> ${ICE_CASEDIR}/test_output

diff --git a/configuration/scripts/tests/test_smoke.script b/configuration/scripts/tests/test_smoke.script
@@ -20,9 +20,12 @@ mv -f ${ICE_CASEDIR}/test_output ${ICE_CASEDIR}/test_output.prev
 cat ${ICE_CASEDIR}/test_output.prev | grep -iv "${ICE_TESTNAME} test" >! ${ICE_CASEDIR}/test_output
 rm -f ${ICE_CASEDIR}/test_output.prev
 
-set grade = FAIL
-if ( $res == 0 ) then
-  set grade = PASS
+set grade = PASS
+if ( $res != 0 ) then
+  set grade = FAIL
+  echo "$grade ${ICE_TESTNAME} run ${ttimeloop} ${tdynamics} ${tcolumn}" >> ${ICE_CASEDIR}/test_output
+  echo "$grade ${ICE_TESTNAME} test " >> ${ICE_CASEDIR}/test_output
+  exit 99
 endif
 
 echo "$grade ${ICE_TESTNAME} run ${ttimeloop} ${tdynamics} ${tcolumn}" >> ${ICE_CASEDIR}/test_output

diff --git a/configuration/scripts/tests/test_unittest.script b/configuration/scripts/tests/test_unittest.script
@@ -14,9 +14,12 @@ mv -f ${ICE_CASEDIR}/test_output ${ICE_CASEDIR}/test_output.prev
 cat ${ICE_CASEDIR}/test_output.prev | grep -iv "${ICE_TESTNAME} test" >! ${ICE_CASEDIR}/test_output
 rm -f ${ICE_CASEDIR}/test_output.prev
 
-set grade = FAIL
-if ( $res == 0 ) then
-  set grade = PASS
+set grade = PASS
+if ( $res != 0 ) then
+  set grade = FAIL
+  echo "$grade ${ICE_TESTNAME} run " >> ${ICE_CASEDIR}/test_output
+  echo "$grade ${ICE_TESTNAME} test " >> ${ICE_CASEDIR}/test_output
+  exit 99
 endif
 
 echo "$grade ${ICE_TESTNAME} run " >> ${ICE_CASEDIR}/test_output

diff --git a/doc/source/cice_index.rst b/doc/source/cice_index.rst
@@ -401,6 +401,7 @@ either Celsius or Kelvin units).
    "mtask", "local processor number that writes debugging data", ""
    "mu_rdg", ":math:`\bullet` e-folding scale of ridged ice", ""
    "myear", "model year", ""
+   "myear_max", "maximum allowed model year", ""
    "my_task", "task ID for the current processor", ""
    "**N**", "", ""
    "n_aero", "number of aerosol species", ""