Append to probe and CoM files only when continuing a run

[sbryngelson]

Written with Claude Code.

The bug

s_open_probe_files appends whenever D/probe*_prim.dat already exists:

if (file_exist) then
    open (i + 30, FILE=trim(file_path), form='formatted', STATUS='old', POSITION='append')

Correct when a run is being continued. Wrong when one is being started over — which is what happens every time a case is re-run in place after a parameter change. The second run's rows land on top of the first's, nothing in the file marks the join, and the time column simply resets partway down. A reader sees one monotonic series and is silently wrong.

The two runs need not even share a grid: a case whose resolution changed between runs produces a file whose first half was recorded at different probe locations.

s_open_com_files has the same pattern.

The fix

Append only when continuing: t_step_start > 0, or n_start > 0 under cfl_dt — the same test m_start_up already uses to decide whether a run is fresh. A fresh start replaces the file, which is what every other output MFC writes already does.

How the validation is obtained

examples/2D_probe_rerun/ — a 64×64 case with two probes, 20 steps. Run it twice in the same directory:

./mfc.sh run examples/2D_probe_rerun/case.py -n 1
./mfc.sh run examples/2D_probe_rerun/case.py -n 1
wc -l D/probe1_prim.dat

	rows in D/probe1_prim.dat
before	40 — two runs of 20, spliced
after	20

The time column read straight through, before:

0.028977
0.030682
0.032386      <- end of run 1
0.000000      <- run 2 starts, time goes backwards
0.001705
0.003409

Measured with the same binary either side of the change; both runs complete, no failures. The README carries the table and the reasoning.

Why it is worth fixing rather than documenting

This produced three separate wrong numbers in one project before it was noticed. The worst: a jet whose probe files held a t = 15 run of 4,962 rows followed by a t = 40 run of 13,233 — on different grids. Read together they manufactured a velocity drop of 0.99 U_j in a single sample, which was time running backwards at the seam, and which was diagnosed as a physical instability first.

A related symptom is louder: if the probe output format changes between runs, the column count changes partway down and numpy.loadtxt refuses the file outright ("the number of columns changed from 11 to 18"). That one announces itself. The time reset does not.

Scope

Only affects re-running a case in a directory that already holds probe or CoM output. Restarts are unchanged — they still append, which is the behaviour they need. No golden file moves; the test suite runs each case in a fresh directory.

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Pull request overview

Fixes probe and center-of-mass (CoM) output files being unintentionally appended when re-running a case “from scratch” in the same directory, which can silently splice multiple runs into one misleading time series.

Changes:

• Determine whether a run is a “fresh start” vs “continuation” and only append for continuations.
• Replace (truncate) existing probe/CoM files on fresh starts to avoid splicing data across separate runs.
• Add a reproducible example (examples/2D_probe_rerun/) documenting the bug and validation.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/simulation/m_data_output.fpp	Add fresh_start gating so probe/CoM files are replaced on fresh runs and appended only on continuations.
examples/2D_probe_rerun/case.py	New example case to reproduce the probe append/splice issue by running twice in-place.
examples/2D_probe_rerun/README.md	Documentation demonstrating the incorrect behavior before and correct behavior after the fix.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

Lines of Code

File	Lines	Diff
src/simulation/m_data_output.fpp	1375	+8

Directory	Lines	Diff
simulation	28094	+8
total	46348	+8

[github-actions]

Claude Code Review

Head SHA: de78f37

Files changed:

• 3
• examples/2D_probe_rerun/README.md
• examples/2D_probe_rerun/case.py
• src/simulation/m_data_output.fpp

Findings:

• src/simulation/m_data_output.fpp: in both s_open_com_files and s_open_probe_files, fresh_start = (t_step_start == 0) .or. (cfl_dt .and. n_start == 0) OR's the two restart checks instead of selecting between them by mode. When cfl_dt is enabled, restarts are indicated by n_start > 0 while t_step_start stays at its default 0; the t_step_start == 0 term is true regardless of n_start, so fresh_start always evaluates .true. for any cfl_dt run. That routes every cfl_dt restart into the new STATUS='replace' branch, silently truncating the existing probe/CoM data files that the restart should have appended to — the opposite of this patch's intent. The condition should pick the check based on cfl_dt (e.g. merge(n_start == 0, t_step_start == 0, cfl_dt)) rather than OR-ing them.

[codecov]

Codecov Report

❌ Patch coverage is 28.57143% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.23%. Comparing base (dc0aec1) to head (4d6ac9b).
⚠️ Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
src/simulation/m_data_output.fpp	28.57%	4 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1867      +/-   ##
==========================================
- Coverage   61.26%   61.23%   -0.03%     
==========================================
  Files          84       84              
  Lines       22330    22340      +10     
  Branches     3265     3267       +2     
==========================================
+ Hits        13680    13681       +1     
- Misses       6207     6215       +8     
- Partials     2443     2444       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.