inito 1.0.0rc2

A Lombok-inspired boilerplate-elimination library for Python.

inito

A Lombok-inspired boilerplate-elimination library for Python. inito generates constructors, repr, equality/hashing, accessors, and fluent builders once at class-decoration time — never at instance construction or attribute-access time — so the generated classes perform like handwritten ones. Zero runtime dependencies.

from inito import Data


@Data
class User:
    name: str
    age: int = 0


# @Data wrote __init__, __repr__, __eq__, __hash__, get_name/set_name, ...
user = User("Ada", age=30)
print(user)              # User(name='Ada', age=30)
print(user.get_name())   # Ada
user.set_age(31)
print(user == User("Ada", 31))   # True

Without inito, the class above is ~20 lines of hand-written __init__, __repr__, __eq__, __hash__, and accessor boilerplate. With inito it is the three lines you see, and the generated methods are the same Python you would have written by hand — benchmarked at parity with handwritten classes and dataclasses.

Decorators at a glance

Decorator	Generates
@Data	constructor, __repr__, __eq__, __hash__, get_x/set_x — the all-in-one
@Value	like @Data but immutable and setter-free (constructor, repr, eq/hash, getters)
@Getter / @Setter	get_x() / set_x(value) accessors only
@ToString	__repr__ only
@EqualsAndHashCode	__eq__ + __hash__ only
@NoArgsConstructor	zero-argument constructor using each field's default
@AllArgsConstructor	constructor taking every field
@RequiredArgsConstructor	constructor taking only the fields without a default
@Builder / builder	fluent Cls.builder().x(1).build(), optional .to_builder()
@Service / @Singleton / @Inject	dependency-injection: register a class + auto-wire it from a Container

Every PascalCase decorator also has a lowercase alias (data, builder, value, ...) bound to the same object — use whichever reads better.

Install

pip install inito

or

uv add inito

Quick start

from inito import Data


@Data
class User:
    name: str
    age: int = 0


user = User("Ada", age=30)
print(user)              # User(name='Ada', age=30)
print(user.get_name())   # Ada
user.set_age(31)

@Data also accepts options:

from inito import Data


@Data(frozen=True)
class Point:
    x: int
    y: int

@builder generates a fluent, chainable builder, and composes with @dataclass:

from dataclasses import dataclass
from inito import builder


@builder(to_builder=True)
@dataclass
class Request:
    prompt: str
    temperature: float = 0.7


request = Request.builder().prompt("hello").build()
revised = request.to_builder().temperature(0.9).build()

@Service/@Singleton/@Inject add lightweight dependency injection — register a class, then let a Container autowire its constructor from other registered services:

from inito import Service, Singleton, Inject, default_container


@Singleton                       # one shared instance per container
class Database:
    def __init__(self) -> None:
        self.users = {1: "Ada"}


@Service                         # autowired from the container on demand
class UserService:
    def __init__(self, db: Database) -> None:
        self.db = db

    def name(self, user_id: int) -> str:
        return self.db.users[user_id]


@Inject                          # fills in `service` from the container
def main(service: UserService) -> None:
    print(service.name(1))       # Ada


main()
print(default_container.get(UserService).name(1))   # Ada — same wiring, explicit

# @Service never rewrites the class: it's still an ordinary Python class.
plain = UserService(Database())

Type checking

inito ships a mypy plugin that synthesizes every generated member (__init__'s real signature, get_x/set_x, .builder()/Builder/ .to_builder()) so mypy --strict sees your decorated classes correctly — no # type: ignore needed. Enable it in your pyproject.toml or mypy.ini:

[tool.mypy]
plugins = ["inito.typing.mypy_plugin"]

The plugin is mypy-only — pyright has no third-party plugin mechanism. @Data and @AllArgsConstructor additionally ship a .pyi stub marked with the standard typing.dataclass_transform (PEP 681), so pyright also gets a correctly-typed __init__ for those two decorators — no plugin needed, since this is a standard both tools understand natively. get_x/set_x and @Builder's fluent chain remain pyright's gap regardless (see below), since dataclass_transform can only express a constructor signature, not arbitrary generated methods.

Status

Implemented today: @Data (constructor, __repr__, __eq__, __hash__, getters, setters), @Getter (getters only), @Setter (setters only), @NoArgsConstructor (no-argument constructor using field defaults), @AllArgsConstructor (constructor only, every field), @RequiredArgsConstructor (constructor only accepting required fields), @Builder/builder (fluent builder, to_builder=True support), @ToString (__repr__ only — pairs well with @Builder for a readable repr without pulling in @Data's constructor/eq/hash/accessors), @EqualsAndHashCode (__eq__/__hash__ only), @Value (@Data without setters — constructor, __repr__, __eq__, __hash__, getters; genuinely immutable on its own, no @dataclass(frozen=True) stacking required), and a dependency-injection subsystem: @Service/@Component (registers a class's constructor dependencies into a Container), @Singleton (sugar for singleton-scoped @Service), and @Inject (auto-wires a function's annotated parameters from a container per call). @Service never mutates the decorated class — it stays an ordinary, directly constructible Python class; container.get(cls) is the DI-aware, lazily-resolving path. See Quick start for a worked DI example.

All of inito.md's Initial Features (v1) are now implemented, plus @Value and dependency injection, both pulled forward from its Future Features list. See the performance page for benchmarks against handwritten classes, dataclasses, and attrs. See TASKS.md for what's left.

Works with your framework

InitO has zero runtime dependencies and generates plain methods on plain classes, so it drops into any project — Django, FastAPI, Sanic, Flask, or no framework. On a framework's model class (Pydantic, SQLAlchemy, Django) prefer the additive decorators (@Getter/@Setter/@ToString/@EqualsAndHashCode) and let the framework own construction; use @Builder(use_init=True) when you want the builder to run the model's own validating constructor. The dependency-injection layer is safe to resolve from async request handlers. See Using InitO with your framework. Interoperability is verified in CI against Pydantic v2, SQLAlchemy 2.0, and Django, on Python 3.9–3.14.

Known limitation: pyright doesn't see most generated members

Every generated member (get_x, set_x, .builder(), .to_builder(), the generated constructor's parameters, ...) is attached to your class via setattr at decoration time — real attributes at runtime. mypy sees all of them correctly once you enable the bundled plugin (the same approach attrs/Pydantic use). pyright has no equivalent third-party plugin mechanism, so get_x/set_x and @Builder's fluent chain still show up as unknown attributes there — your code runs correctly regardless, this is a pyright-only static-typing gap.

@Data's and @AllArgsConstructor's constructors are the exception: pyright does see these correctly, via a .pyi stub marked with the standard typing.dataclass_transform (no inito-specific plugin needed — this is a mechanism pyright supports natively). This doesn't extend to @NoArgsConstructor/@RequiredArgsConstructor: both were deliberately left unmarked, since their real signatures diverge from what dataclass_transform can express (@NoArgsConstructor truly accepts zero arguments rather than "all fields optional"; @RequiredArgsConstructor excludes defaulted fields from __init__ entirely rather than making them optional) — applying the marker there would make pyright silently accept calls the real runtime rejects, which is worse than the current honest gap. Closing the remaining gap for get_x/set_x/@Builder would need a different strategy (e.g. a companion stub generator); tracked in TASKS.md Phase 17, not required for this release.

Immutability: @Value and @Data(frozen=True)

@Value and @Data(frozen=True) are genuinely immutable, on their own — no @dataclass(frozen=True) stacking needed. Attribute assignment and deletion always raise dataclasses.FrozenInstanceError after construction:

from inito import Value


@Value
class Point:
    x: int
    y: int


point = Point(1, 2)
point.x = 5   # raises dataclasses.FrozenInstanceError
del point.x   # also raises

For an immutable class, generated constructors (and @Builder's build()) assign fields via object.__setattr__ internally, bypassing the blocking __setattr__ — the exact technique a real frozen dataclass's own __init__ uses — so construction always succeeds. A non-frozen class uses a plain self.x = x instead, which is both faster and keeps attribute reads at handwritten speed. @Data(frozen=True) also skips generating setters; @Value never generates them in the first place.

Composing with frozen dataclasses

To combine a decorator that has no frozen option of its own (@Builder, @AllArgsConstructor, ...) with @dataclass(frozen=True), stack the @dataclass(frozen=True) innermost — i.e. closest to the class:

@Data
@dataclass(frozen=True)   # innermost — correct
class Point:
    x: int
    y: int

In this order the frozen __setattr__ already exists when inito generates its constructor, so inito detects it and builds correctly. The reverse order (@dataclass(frozen=True) outermost, applied after inito) is not supported — inito can't see a decorator that hasn't run yet, so its constructor would use self.x = x and construction would raise FrozenInstanceError. For an immutable class, prefer @Value or @Data(frozen=True) (no stacking needed at all), or use the innermost form above. Post-construction mutation always raises FrozenInstanceError — only construction is exempted from the frozen check.

Self-referential fields

Self-referential type hints (e.g. a linked-list next: Node) work correctly:

from typing import Optional
from inito import Data


@Data
class Node:
    value: int
    next: Optional[Node] = None

inito resolves annotations eagerly, once, at decoration time — before the class's own name is bound in its module's globals — so naively this would fail to resolve. Instead, resolve_type_hints temporarily seeds the module's namespace with the class itself just before resolution (and removes it immediately after), which only affects resolution of the class being decorated, not any other class in its inheritance chain. This is a one-time, decoration-time-only operation with no per-instance or per-call cost. Forward references to any other, already-defined class continue to work normally, and a genuinely undefined name still correctly raises AnnotationResolutionError.

Use Optional[Node] rather than Node | None for a self-referential field: the annotation is evaluated at runtime by get_type_hints, and the | union syntax isn't valid there before Python 3.10 (inito supports 3.9+).

Contributing

See CONTRIBUTING.md.

License

MIT — see LICENSE.