schema-salad-plus-pydantic 0.1.8

Generate pydantic v2 BaseModel classes from schema-salad definitions

schema-salad-plus-pydantic

Generate pydantic v2 BaseModel classes, TypeScript interfaces, and Effect Schema modules from schema-salad definitions.

What it does

Schema-salad defines record types, enums, inheritance, and unions in YAML. This tool reads those definitions and emits a Python module of pydantic BaseModel classes, a TypeScript module of interfaces, or an Effect Schema TypeScript module with runtime validation -- all conforming to the schema.

Key features:

• Proper pydantic inheritance -- abstract bases declare fields, children inherit them, multiple inheritance works naturally.
• Schema annotations for types schema-salad can't express natively:
  • pydantic:type -- override the generated type annotation (e.g. dict[str, NativeStep])
  • pydantic:alias -- set a Field alias for JSON keys that differ from the Python name
  • pydantic:discriminator_field / pydantic:discriminator_map -- discriminated unions
• Enums -- multi-symbol enums become str, Enum classes; single-symbol enums become Literal["value"] with auto-defaults.
• Forward references -- model_rebuild() for all classes, from __future__ import annotations.
• Permissive by default -- extra="allow", populate_by_name=True.
• TypeScript output -- --format=typescript emits interfaces, string union enums, and type guard functions for discriminated unions.
• Effect Schema output -- --format=effect-schema emits Schema.Struct definitions with runtime validation via Schema.decodeUnknownSync(), discriminated unions, and type guards.

Installation

pip install schema-salad-plus-pydantic

Or with uv:

uv pip install schema-salad-plus-pydantic

Usage

CLI

Generate pydantic models from a schema-salad YAML file:

schema-salad-plus-pydantic generate schema.yml -o models.py

Pass --strict to emit models with extra="forbid" (reject unknown JSON keys); the default is permissive extra="allow".

Generate TypeScript interfaces:

schema-salad-plus-pydantic generate schema.yml --format typescript -o models.ts

Generate Effect Schema TypeScript (with runtime validation):

schema-salad-plus-pydantic generate schema.yml --format effect-schema -o models.ts

Or write to stdout:

schema-salad-plus-pydantic generate schema.yml > models.py

Python API

from io import StringIO
from schema_salad_plus_pydantic.orchestrate import generate_from_schema

buf = StringIO()
generate_from_schema("path/to/schema.yml", buf)
code = buf.getvalue()

# Or write directly to a file
with open("models.py", "w") as f:
    generate_from_schema("path/to/schema.yml", f)

# Optional: strict=True emits models with extra="forbid" (unknown keys rejected)
with open("models_strict.py", "w") as f:
    generate_from_schema("path/to/schema.yml", f, strict=True)

# Generate TypeScript interfaces
with open("models.ts", "w") as f:
    generate_from_schema("path/to/schema.yml", f, output_format="typescript")

# Generate Effect Schema TypeScript (runtime validation)
with open("models.ts", "w") as f:
    generate_from_schema("path/to/schema.yml", f, output_format="effect-schema")

Using the generated models

import json
from generated_models import MyRecord  # the module you generated

# Validate a dict
obj = MyRecord.model_validate({"field": "value", "count": 42})

# Validate from JSON
with open("data.json") as f:
    obj = MyRecord.model_validate(json.load(f))

# Access fields
print(obj.field)
print(obj.count)

# Serialize back to dict/JSON
print(obj.model_dump())
print(obj.model_dump_json(indent=2))

Schema annotations

Add pydantic:* keys to schema-salad field definitions to control the generated type annotations. Requires a pydantic namespace declaration:

$namespaces:
  pydantic: "https://example.org/pydantic#"

$graph:
- name: MyRecord
  type: record
  fields:
    - name: steps
      type: Any?
      pydantic:type: "dict[str, Step]"

    - name: format_version
      type: string
      pydantic:alias: "format-version"

    - name: creator
      type: Any?
      pydantic:type: "list[Person | Organization] | None"
      pydantic:discriminator_field: "class"
      pydantic:discriminator_map: '{"Person": "Person", "Organization": "Organization"}'

TypeScript output

With --format=typescript, the same schema annotations produce TypeScript interfaces with mapped types:

Python / pydantic	TypeScript
dict[str, Step]	Record<string, Step>
list[Person | Organization]	Array<Person | Organization>
Literal["value"]	"value" (string literal)
int, float	number
str	string

Discriminated unions emit type guard functions:

export function isPerson(v: Person | Organization): v is Person {
  return v?.class === "Person";
}

Effect Schema output

With --format=effect-schema, the tool generates TypeScript using Effect Schema which provides both compile-time types and runtime validation:

Python / pydantic	Effect Schema
dict[str, Step]	Schema.Record({ key: Schema.String, value: StepSchema })
list[Person | Organization]	Schema.Array(Schema.Union(PersonSchema, OrganizationSchema))
Literal["value"]	Schema.Literal("value")
int, float	Schema.Number
str	Schema.String
enum (multi-symbol)	Schema.Literal("a", "b", "c")
optional field	Schema.optional(T)
inheritance	...ParentSchema.fields spread

Records become Schema.Struct definitions with derived type aliases:

import { Schema } from "effect"

export const MyRecordSchema = Schema.Struct({
  name: Schema.optional(Schema.Union(Schema.Null, Schema.String)),
  status: Schema.optional(Schema.Union(Schema.Null, StatusEnumSchema)),
})
export type MyRecord = typeof MyRecordSchema.Type

// Runtime validation
const record = Schema.decodeUnknownSync(MyRecordSchema)({
  name: "example",
  status: "active",
})

Circular references (e.g. recursive schemas) are handled automatically via Schema.suspend.

Development

Setup with uv:

uv sync --group test --group lint --group mypy

Run checks:

make test         # pytest
make lint         # ruff + black
make mypy         # type checking

Releasing

See RELEASE_CHECKLIST.md. Quick version:

make add-history  # generate PR acknowledgements in HISTORY.rst
make release      # tag, build, push (triggers PyPI publish via GitHub Actions)

License

MIT -- see LICENSE.