model-parser

Convert process-model definitions to and from a canonical intermediate representation (IR).

model-parser is the Advanced Process Control toolbox component that owns the model scaffold contract and the transformations around it. It parses an authoring format (today: the ExprTk-style INI used by the MPC / simulation toolchain) into a normalized, backend-independent canonical IR, and lowers that IR into generated Julia and Python model views.

authoring (ExprTk INI)  --parse-->  AST  --normalize-->  canonical IR (JSON)
                                                          |
                                          emit julia      --> ModelingToolkit .jl
                                          emit julia-rhs  --> numerical f!/outputs! .jl
                                          emit python     --> numerical functions .py
                                          emit cpp        --> (planned) realtime C++

The IR is the single semantic contract. Adding a backend means writing one lower + one export, not an N×N mesh of view-to-view translators. See docs/design/model-parser.md for the authoritative product specification and docs/decisions/ for the design decision records.

Why two languages?

The Python CLI builds and validates the IR and generates Julia code; the Julia package can additionally load an IR directly into an MTK System for in-memory, dynamic workflows. Both consume the same IR. See ADR 0001.

Install

CLI from PyPI (end users)

The package on PyPI is apc-model-parser; the installed command is model-parser.

pipx install apc-model-parser
# or: uv tool install apc-model-parser
model-parser --help

Use pipx or uv tool when you only need the CLI in an isolated environment.

From source (development)

This repository uses uv for environments and tasks.

uv sync --all-groups    # include dev tools (ruff, pytest, mkdocs)
uv run model-parser --help

CLI

# 1. ExprTk INI  ->  canonical IR JSON
uv run model-parser parse examples/models/model_monod_simple.ini -o monod.ir.json

# 2. canonical IR  ->  ModelingToolkit (v11) Julia model
uv run model-parser emit julia monod.ir.json -o monod.jl

# 2a. canonical IR  ->  plain numerical ODE RHS (SciML-style f!)
uv run model-parser emit julia-rhs monod.ir.json -o monod_rhs.jl

# 2b. canonical IR  ->  Python ODE RHS and output functions
uv run model-parser emit python monod.ir.json -o monod.py

# Supporting commands
uv run model-parser validate monod.ir.json --profile julia-analysis
uv run model-parser inspect  monod.ir.json
uv run model-parser diff     monod.ir.json other.ir.json   # semantic IR diff
uv run model-parser bump     monod.ir.json other.ir.json   # suggested SemVer bump
uv run model-parser ast      examples/models/model_monod_simple.ini   # debug tree
uv run model-parser schema   -o schemas/canonical-ir.schema.json      # export schema

parse accepts --from exprtk-ini (default). validate accepts either an IR .json file or an INI file (parsed on the fly). Exit codes: 0 success, 1 validation errors, 2 usage / load failure.

How "stored MTK models" work

The persisted, version-controlled form of a model is the IR JSON plus the generated .jl script — not a serialized System object. ModelingToolkit v11 changed System internals significantly (precompilation, removal of defaults, deprecation of @mtkmodel), so serializing the live object is brittle across versions. Regenerating from the IR is the durable path. See ADR 0002 and docs/design/storing-mtk-models.md.

Development

uv sync --all-groups
uv run ruff check .
uv run ruff format --check .
uv run pytest
uv run mkdocs build --strict   # same as CI

Documentation site (after Pages is set to GitHub Actions): https://advanced-process-control.github.io/model-parser/

Scope

In scope: authoring-format parsing, the canonical IR data model + JSON Schema, semantic & profile validation, IR↔backend lowering/codegen.

Out of scope: parameter identification, scenario execution, simulation result storage, controller synthesis, deployment. Those are sibling tools that consume the IR. The parser stays small (see AGENTS.md scope guardrails).