IB force history: record each run's last step; don't overwrite on restart

docs/documentation/case.md

@@ -866,7 +866,9 @@ row    = t_step / ib_force_stride - t_step_start / ib_force_stride - 1
 offset = (row * num_ibs + ib_id - 1) * 353
 ```
 
-Rows count from the first step the run records, not from `t_step`, so row 0 is the first row of the file whether the run starts at step 0 or resumes from a restart. The first recorded step is the first multiple of `ib_force_stride` after `t_step_start`; `t_step_start` itself is skipped, because at that point the force is still the one from before the run began.
+Rows count from the first step the run records, not from `t_step`, so row 0 is the first row of the file whether the run starts at step 0 or resumes from a restart. The first recorded step is the first multiple of `ib_force_stride` after `t_step_start`; `t_step_start` itself is skipped, because at that point the force is still the one from before the run began. The last row is the run's final step if that step is a multiple of `ib_force_stride`, and otherwise the last multiple before it.
+
+A resumed run writes `D/ib_forces_<t_step_start>.dat` (`D/ib_forces_n<n_start>.dat` with `cfl_dt`) instead of overwriting `D/ib_forces.dat`. When each run resumes from the final step of the run before it, the files concatenate in order with no row repeated and none of the `ib_force_stride` sampling missing. This holds for any stride with a fixed `dt`. With `cfl_dt` the step counter restarts at 0 in each run, so only `ib_force_stride = 1` gives an unbroken history; a larger stride can leave up to `ib_force_stride - 1` unrecorded steps at the end of each run. A run killed mid-way has no final row and may hold rows past the restart it is resumed from.
 
 Rows are written in global body-id order, so the file is byte-identical however the domain is decomposed, and no merge step is needed after a parallel run.
 

src/simulation/m_data_output.fpp

@@ -1370,7 +1370,14 @@ contains
         @:PROHIBIT(len_trim(probe) /= IB_REC_BODY, &
                    & "IB force record width disagrees with IB_REC_BODY;  IB_REC_FMT and IB_REC_BODY must be changed together")
 
-        file_loc = trim(case_dir) // '/D/ib_forces.dat'
+        ! A resumed run writes its own file rather than overwriting the history of the runs before it
+        if (cfl_dt .and. n_start > 0) then
+            write (file_loc, '(A,I0,A)') trim(case_dir) // '/D/ib_forces_n', n_start, '.dat'
+        else if (t_step_start > 0) then
+            write (file_loc, '(A,I0,A)') trim(case_dir) // '/D/ib_forces_', t_step_start, '.dat'
+        else
+            file_loc = trim(case_dir) // '/D/ib_forces.dat'
+        end if
 #ifdef MFC_MPI
         ! MPI_MODE_CREATE does not truncate, so a shorter run following a longer one in the same
         ! directory would keep the old tail past its last record. Delete first, as the ib_state

src/simulation/p_main.fpp

@@ -8,6 +8,7 @@ program p_main
     use m_global_parameters
     use m_start_up
     use m_time_steppers
+    use m_data_output, only: s_write_ib_force_history
     use m_nvtx
 
     implicit none
@@ -90,6 +91,9 @@ program p_main
         call system_clock(cpu_end)
     end do
 
+    ! Step N's IB force record is written at the start of step N, which the loop never reaches for its last step
+    if (ib_state_wrt) call s_write_ib_force_history(t_step)
+
     call nvtxEndRange  ! Simulation
 
     deallocate (proc_time, io_proc_time)