Everything here needs the pyforefire module: either a build configured with -DFOREFIRE_BUILD_PYTHON=ON, or pip install forefire.

../run.bash runs only the first two, through $PYTHONEXE.

Four fires side by side, each in an 80×60 band of fuel filled at random to a different density (k_coeffs = [.10, .3, .4, .5]), so the bands sit either side of the percolation threshold. Writes percolation.nc. A demonstration of building a heterogeneous fuel map in numpy and handing it to ForeFire with addIndexLayer.

One fire under a wind rotating from 0° to 360°, which should trace a circular front. Writes 360wind.png through ForeFire's own plot[] command — the one artefact ../run.bash checks for.

Reproduces a FARSITE benchmark case: flat terrain, a 3 mph north wind, seven hours of spread, compared against the FARSITE result.

It cannot run as checked out. It reads flatland.lcp, a landscape file that is not in this repository. Download it first:

curl -LO https://github.com/mbedward/farsite/raw/refs/heads/master/examples/flatland/Inputs/a_lcpFiles/flatland.lcp

The weather it uses, flatland_3mph0deg7hr.raws, is in this directory.

These three are tests rather than examples, and ../run.bash runs none of them. They are plain scripts, not pytest modules. TESTING.md covers each in detail, including how to build the module they import.

Asserts, for every propagation model that consumes dead fuel moisture, that rate of spread stays finite, falls as moisture rises, reaches zero at the moisture of extinction, and responds to a dynamic moisture layer. It holds no reference data — every assertion follows from the published spread equations — so unlike runff it can tell a physics fix from a physics regression.

Run in CI by invariants.yml. Manually:

python tests/python/test_moisture_invariants.py        # -v for every probe

--model NAME and --test NAME, both repeatable, narrow it down.

Runs eight simulations in threads and requires each to reproduce the node count it produces alone.

It does not pass, by design, and is not run in CI. It is the failing reproduction for the shared-state problem in #175. It needs a free-threaded CPython and PYTHON_GIL=0; on an ordinary interpreter the GIL serialises every call and the test skips, reporting a green result that proves nothing.

PYTHON_GIL=0 python3.14t tests/python/test_threading.py

The smoke test cibuildwheel runs against a built wheel: the module imports, its vendored NetCDF resolves, the propagation models registered, and a trivial simulation advances. Run it against an installed forefire, never from the source tree:

python tests/python/test_wheel.py