CFD-craft for Python

A native Rust research solver for incompressible flow and passive-gas dispersion, called in-process from Python 3.10 or newer. No NumPy, OpenFOAM or subprocess is needed.

pip install cfd-craft

The example loads the small point-source case of the user manual (prescribed flow, one tracer, one downstream receptor) directly from GitHub:

import json
import urllib.request
import cfd_craft as cfd

URL = (
    "https://raw.githubusercontent.com/dietmarwo/CFD-craft/main/"
    "examples/manual/point-source.json"
)
with urllib.request.urlopen(URL, timeout=30) as response:
    case = json.load(response)
report = cfd.check(case)
if not report["valid"]:
    raise ValueError(report["diagnostics"])
result = cfd.simulate(case)
print(result["termination"], result["assessment"])
tracer = result["constituents"][0]
print(tracer["receptors_kg_m3"]["R_down"] * 1e9, "micrograms/m3")

To keep a local copy, download it once (for example curl -O https://raw.githubusercontent.com/dietmarwo/CFD-craft/main/examples/manual/point-source.json) and load it with case = json.loads(Path("point-source.json").read_text()) after from pathlib import Path. check and simulate also accept a JSON string or bytes.

check runs the native CLI's strict parsing and preflight and returns coded diagnostics with JSON paths. simulate releases the GIL, enforces the case's work, wall-time and thread limits and returns the same summary schema as the Rust CLI; incomplete calculations keep their termination and assessment. Use cancel_file=PATH for cooperative cancellation; Ctrl-C stops the solver and raises KeyboardInterrupt. run.memory_limit_mib is a preflight estimate, and the host-memory refusal applies on Linux only.

python -m cfd_craft describe, check CASE.json and simulate CASE.json print JSON; the cfd-craft-python console script does the same. Fields, checksummed archives, restarts and profiles belong to the Rust CLI (cargo install cfd-craft-cli --locked).

Version 0.1.0 is a research prototype: experimental canyon validation and a successful fine-grid resolution study remain open, and a numerical acceptance is not physical validation.

• Python API, examples and limitations
• Repository, user manual and evidence
• Changelog

MIT licensed.