emf-codegen — Python code generator for Ecore

A Python code generator for Ecore models, built on EMFPy — the Python port of EMF/Ecore. It is the Python equivalent of the TypeScript @emfts/codegen.

Status: complete (Phases 1–4, 6–8; Phase 5 intentionally out of scope). See PLAN.md for the roadmap.

What it does

• Loads an .ecore model (via EMFPy's XMI reader) plus a .genconfig.xmi configuration.
• Builds an internal generation model (GenModel / GenPackage / GenClass / …).
• Emits Python source in one of three modes:
  • plain — dataclass-style Python classes (no EMF runtime dependency),
  • decorator — dataclasses carrying Ecore metadata (ECLASS_URI class var + field(metadata={"kind", "type", "many"})),
  • emf — concrete, typed EObject subclasses (one per EClass) whose @property accessors delegate to e_get/e_set, plus a <prefix>_package module holding the EPackage and class/feature literals, a factory, and an __init__ barrel.
• A CLI (emfpy-codegen generate … / init …) for build pipelines.

Generation modes

Mode	Output	Runtime dependency	Use when
plain	@dataclass per class + Enum	none	plain data holders, no EMF semantics
decorator	@dataclass + metadata + ECLASS_URI	none (metadata is inert)	data holders that retain Ecore provenance
emf	typed EObject subclass per class + package/factory/barrel	emf	full EMF semantics (reflection, opposites, containment) with a typed class API

Python adaptations vs. the TypeScript original

• decorator: EMFPy has no per-field decorator framework, so the TS @ModelClass / @Attribute / @Reference decorators become an ECLASS_URI class variable plus dataclasses.field(metadata=…), reachable via dataclasses.fields().
• emf: like the TS original, one concrete class per EClass. The TS impl stores values in private fields and reimplements the reflective eGet/eSet switch; the Python classes instead back every typed @property with self.e_get/self.e_set on the package's feature literals, so the EMFPy runtime maintains opposites, containment and container references for free. Generated classes are themselves mypy --strict clean.

Usage

# Generate from an existing model + config
emfpy-codegen generate -m examples/model/library.ecore \
    -c examples/model/library.genconfig.xmi -o generated -v

# Bootstrap a .genconfig.xmi for a model
emfpy-codegen init -m examples/model/library.ecore --mode emf --prefix Library

Programmatic API:

from emf_codegen import generate, generate_in_memory

result = generate("model/library.ecore", "model/library.genconfig.xmi", output_dir="out")
for diag in result.diagnostics:
    print(diag.level, diag.message)

Generated emf-mode example

from org.example import Book, Author, Library, BookCategory   # generated classes

book = Book()
book.title = "Neuromancer"
print(book.available)              # True  (default literal "true" parsed)
print(book.category)               # BookCategory.FICTION

author = Author()
book.author = author               # bidirectional opposite is maintained …
assert book in list(author.books)  # … so author.books now contains the book

library = Library()
library.books.add(book)            # containment
assert book.library is library     # readonly container reference derives from it

A ready-made example is generated under examples/generated/.

Relationship to the TypeScript original

@emfts/codegen	emf-codegen
generates TypeScript	generates Python
@emfts/core runtime	emf (EMFPy) runtime
EJS templates	Jinja2 templates
commander CLI	argparse CLI
sax for XMI	EMFPy XMIResource (stdlib xml.sax)
@ModelClass/@Attribute decorators	ECLASS_URI + field(metadata=…)
per-class BasicEObject subclass + field-backed reflective eGet/eSet	per-class EObject subclass with @property → e_get/e_set
service-interface + REST-client generators	out of scope

Development

python -m venv .venv && source .venv/bin/activate
pip install -e ../EMFPy-core   # the EMFPy runtime (provides the `emf` package)
pip install -e ".[dev]"
pytest && ruff check . && mypy

License

Eclipse Public License 2.0 (EPL-2.0).