modelica

A pure-Python Modelica toolchain: parse Modelica, flatten it to a DAE, simulate it, and interoperate with the two Modelica Association exchange standards — FMI 3.0 for single components and SSP 2.0 for systems of them.

Distribution name is modelica; the git repo keeps its historical name pymodelica. (PyPI's pymodelica is held by an abandoned Modelon/JModelica.org upload.)

pip install modelica          # core: parse + model structure, zero dependencies
pip install "modelica[sim]"   # + numpy/scipy for simulation

Status

Pre-alpha. This is a full rewrite of a 2010 Python 2 experiment — none of that code survives, only the goal. Python 3.11+.

Why

The Python side of Modelica is split: FMPy and PyFMI consume FMUs that some other tool compiled, and the tools that actually compile Modelica (Dymola, OpenModelica, JModelica) are large non-Python programs. Nothing in Python takes Modelica source all the way to a running simulation, and nothing treats the model — flat equations, the incidence structure, the index — as an ordinary Python data structure you can inspect and rewrite.

That gap is the project: a readable, dependency-light implementation where every stage is a value you can hold.

Layers

Each layer is usable without the ones above it.

See the architecture notes for the data model and the order things get built in.

Development

make install-hooks   # once after cloning: pre-commit (auto version bump + ruff), pre-push (full check)
make check           # lint + typecheck + test, exactly what CI gates on
make docs            # build the mkdocs site strictly

The toolchain is pinned in .prototools (moon + uv) — proto install brings up everything. All task ordering lives in moon.yml; CI is a single moon run :ci.

Releasing

There is no manual release step. pyproject.toml's version is the single source of truth:

• the pre-commit hook auto-bumps the patch version whenever src/**/*.py changes;
• scripts/ci-version-guard fails the pipeline if shipped source moved without a bump (the non-bypassable counterpart);
• on main, moon run :release publishes to PyPI only if that version isn't there yet, via GitLab OIDC Trusted Publishing — no API token is stored anywhere.

Docs are published to GitLab Pages from the same pipeline.

License

MIT © Jorge Cardona