testCases/README.md

Execution checks

This directory holds observer unit tests, a serial smoke test and a serial/MPI integration check.

The serial smoke test checks that simulationCases/bretherton.c compiles and integrates a few dozen time steps at coarse resolution without producing a truncated log or blowing up. It compares against nothing, so it supports no claim about the discretisation or about the physics. Its value is that it fails loudly and in seconds when a change breaks the build or the parameter path.

File	Role
`smoke.params`	coarse, short-run parameters (`MAXlevel = 9`, `tmax = 0.05`)
`runSmokeTests.sh`	builds and runs the case, then checks the log
`runParallelSmokeTests.py`	compares serial and two-rank MPI runs and restarts
`centralFilmObserver.c`	tests convergence windows, quality gates and history resets
`runCentralFilmTests.sh`	compiles and runs the observer tests with a C99 compiler
`regridTube.c`	measures transfer and projection changes when refining a saved tube state

Cases that compare against an exact solution live in ../verificationCases/ and are driven by ../verificationCases/runVerification.sh. The campaign that compares against independent experimental data — Taylor (1961) and Aussillous & Quéré (2000) — is configured by sweep.params and sweep-drop.params at the repository root and is meant for production hardware. Those three evidence sources support three different claims and are deliberately kept apart.

Running

bash testCases/runSmokeTests.sh   # this directory only
bash runTests.sh --unit          # observer tests without Basilisk
bash runTests.sh                 # observer, verification and smoke tests

The build goes to simulationCases/9999/, which is gitignored.

The parallel integration check needs Linux with Python pidfd support, qcc, mpicc, and an Open MPI rankfile naming two allocated cores. It builds serial and MPI modes once, compares short fresh and restarted runs, checks their logs and snapshots, and exercises collective early termination. It also compares refined restarts and checks that rejecting a changed domain length leaves the seed checkpoint intact. Short runs must report the expected incomplete terminal condition; a nonzero exit alone does not pass. Run it only inside the corresponding reserved CPU set with a healthy MPI toolchain:

python3 testCases/runParallelSmokeTests.py --rankfile /absolute/path/to/rankfile \
  --work-root /absolute/path/to/empty-test-directory

The check retains command logs and stops on failed execution, VOF CFL warnings or disagreement beyond its stated tolerances. It checks parallel execution and restart consistency; performance and mesh convergence require separate tests.