eon-schema 0.2.0

Shared eOn L0 Cap'n Proto SSoT, L1 job-config pydantic models, and L2 API specs for eon-akmc + pyeonclient

eon-schema

Shared schema package for both eon-akmc and pyeonclient. One home for:

Layer	Module	Role
L0	eon_schema.ssot	Cap’n Proto field graph (vendored from monorepo schema/)
L1	eon_schema.config	Full job-config pydantic models (MainConfig, Metatomic, Config, …)
L2	eon_schema.api / fields	In-process specs (DimerSpec, NebSpec, enums)

Consumers:

• eon-akmc — depends on eon-schema; eon.schema re-exports L1 for Sphinx / legacy imports.
• pyeonclient — pyeonclient[models] pulls eon-schema; pyeonclient.models re-exports L2.

Do not re-author job models under eon/ or duplicate L2 under pyeonclient/.

Fat tree vs this package

	Fat monorepo tarball	This PyPI package
Purpose	conda-forge eon, full C++/server builds	Shared Python schema for both clients
Cap’n Proto	Full monorepo including schema/	Vendored under src/eon_schema/ssot/
L1 models	Also present via eon.schema re-export	Authoritative under eon_schema.config
Feedstock	Uses fat tarball only	Not required for 0.2.x

L0 authoring is monorepo schema/eon_params.capnp. After edits:

python tools/params_ssot/codegen.py
./packages/eon-schema/scripts/sync_ssot_into_package.sh

L1 authoring is packages/eon-schema/src/eon_schema/config/models.py. Keep field parity with L0 for covered sections (tests/test_params_ssot.py).

See PUBLISHING.md for fat vs split release trains.

Install

pip install eon-schema            # L0 + L1 + L2 (pydantic is a hard dep)
# monorepo:
pip install -e packages/eon-schema

Quick use

# L0
from eon_schema.ssot import capnp_path, load_catalog

# L1 (same objects as eon.schema.*)
from eon_schema.config import MainConfig, Metatomic, Config

# L2 (same objects as pyeonclient.models.*)
from eon_schema.api import DimerSpec
from eon_schema.fields import MinModeMethod, Accelerant

print(capnp_path())
print(MainConfig().job)
print(DimerSpec(method=MinModeMethod.improved, accelerant=Accelerant.gp).core_kwargs())

eon-akmc / Sphinx still use:

from eon.schema import MainConfig, Metatomic  # re-export

pyeonclient still uses:

from pyeonclient.models import DimerSpec, NebSpec  # re-export

Layout

src/eon_schema/
  ssot/          # L0 vendored Cap'n Proto + catalog
  config/        # L1 job-config models (models.py)
  fields/        # enums
  api/           # L2 DimerSpec, NebSpec
  _deps.py

INI helpers (config write without eon-akmc)

from eon_schema.config import (
    MainConfig,
    PotentialConfig,
    DimerConfig,
    write_models_ini,
    write_ini,
    hydrate_ini,
    unknown_ini_keys,
    defaults_from_catalog,
)

# L1 models → config.ini (Dimer L1 prefixes mapped to INI keys)
write_models_ini(
    "config.ini",
    MainConfig(job="saddle_search"),
    PotentialConfig(potential="lj"),
    DimerConfig(dimer_improved=True),
    extra={"Debug": {"write_movies": True}},
    validate=True,
)

# Or raw sections (rgpycrumbs-style) with L0 hydrate + key check
user = {"Main": {"job": "minimization", "temperature": 400}}
cfg = hydrate_ini(user, base_sections=["Main", "Optimizer"])
assert not unknown_ini_keys(cfg)
write_ini("config.ini", cfg)

print(defaults_from_catalog("Main")["job"])

Downstream: rgpycrumbs should depend on eon-schema for write_eon_config / seed_dimers / MLflow log_params instead of full eon-akmc.