Reactor DI for Python

A code generator for dependency injection (DI) in Python which is based on the mediator and factory patterns.

Features

• Two powerful decorators: @module and @law_of_demeter, plus lookup and make annotation markers
• Code generation approach: Generates DI code rather than runtime injection
• Mediator pattern: Central coordination of dependencies
• Factory pattern: Object creation abstraction
• Type-safe: Full type hint support
• Nested modules: Child modules use lookup[Type] to share dependencies from parent modules
• Lazy dependency resolution: Dependencies resolved individually on first access, supporting deferred initialization patterns (e.g., async context managers)
• TYPE_CHECKING compatible: Works with if TYPE_CHECKING: imports for circular dependency avoidance
• Pydantic compatible: Works with Pydantic BaseSettings/BaseModel annotation-only fields
• ABC compatible: Automatic resolution of abstract @property conflicts with DI annotations
• Python 3.9+ support: Tested on Python 3.9 through 3.15

Installation

pip install reactor-di

Quick Start

from reactor_di import module, law_of_demeter, CachingStrategy

class DatabaseConfig:
    host = "localhost"
    port = 5432
    timeout = 30

@law_of_demeter("_config")
class DatabaseService:
    _config: DatabaseConfig
    _host: str      # Forwarded from config.host
    _port: int      # Forwarded from config.port
    _timeout: int   # Forwarded from config.timeout
    
    def connect(self) -> str:
        return f"Connected to {self._host}:{self._port} (timeout: {self._timeout}s)"

# Module: Automatic DI: Implements annotations as @cached_property functions
@module(CachingStrategy.NOT_THREAD_SAFE)
class AppModule:
    config: DatabaseConfig      # Directly instantiated
    database: DatabaseService   # Synthesized with dependencies

# Usage
app = AppModule()
db_service = app.database
print(db_service.connect())  # → "Connected to localhost:5432 (timeout: 30s)"

# Properties are cleanly forwarded
print(db_service._host)      # → "localhost" (from config.host)
print(db_service._timeout)   # → 30 (from config.timeout)

Examples

The examples/ directory contains testable examples that demonstrate all the features shown in this README:

• quick_start.py - The complete Quick Start example above, converted to testable format
• quick_start_advanced.py - Advanced quick start with inheritance patterns
• caching_strategy.py - Demonstrates CachingStrategy.DISABLED, CachingStrategy.NOT_THREAD_SAFE, and CachingStrategy.THREAD_SAFE
• make_marker.py - Subtype factory generation with make[BaseType, ImplType]
• nested_modules.py - Nested modules with lookup for parent dependency sharing
• stacked_decorators.py - Shows using multiple @law_of_demeter decorators on the same class
• custom_prefix.py - Demonstrates custom prefix options (prefix='', prefix='cfg_', etc.)
• side_effects.py - Tests side effect isolation during decoration
• testing.py - Testing pattern: replacing module components with mocks
• typing_decorators_preserve_class_type.py - Demonstrates (and statically asserts via mypy) that @law_of_demeter and @module preserve the input class type
• typing_lookup_make_erasure.py - Demonstrates (and statically asserts via mypy) that lookup[T] and make[B, I] erase to T / B for type checkers, so attribute access type-checks without cast()

Running Examples

# Run all examples as tests
uv run pytest examples/

# Run a specific example
uv run pytest examples/quick_start.py

All examples are automatically tested as part of the CI pipeline to ensure they stay current with the codebase.

Tests

The tests/ directory contains regression and unit tests (63 tests):

• test_module_integration.py - Module + law_of_demeter integration with annotation-only configs, Pydantic compatibility, and abstract property conflict resolution
• test_lazy_resolution.py - Lazy per-attribute resolution with deferred initialization patterns
• test_forward_ref.py - TYPE_CHECKING forward reference handling in module factory
• test_dataclass_law_of_demeter.py - Dataclass + stacked @law_of_demeter regression tests (3 tests)
• test_pure_hasattr.py - Comprehensive tests for the pure_hasattr utility (15 tests)
• test_thread_safe.py - Thread-safe caching strategy with concurrent access tests (12 tests)
• test_type_utils_markers.py - Runtime invariants of the lookup / make markers, e.g. that they are callable for typing._type_check (14 tests)
• test_law_of_demeter.py - Law of Demeter decorator tests
• test_side_effects.py - Side effects isolation during decoration

# Run all tests (examples + regression tests)
uv run pytest

# Run only regression tests
uv run pytest tests/

Architecture

Reactor DI uses a code generation approach with clean separation of concerns:

• module.py - The @module decorator for dependency injection containers
• law_of_demeter.py - The @law_of_demeter decorator for property forwarding
• caching.py - Caching strategies (CachingStrategy.DISABLED, CachingStrategy.NOT_THREAD_SAFE, CachingStrategy.THREAD_SAFE)
• type_utils.py - Simplified type checking utilities and the lookup/make annotation markers (Python 3.9+ stable APIs)

The decorators work together through simple hasattr checks - @law_of_demeter creates forwarding properties that @module recognizes as already implemented, enabling clean cooperation without complex validation logic.

Advanced Usage

Caching Strategies

from reactor_di import module, CachingStrategy

# No caching - components created fresh each time
@module(CachingStrategy.DISABLED)
class DevModule:
    service: MyService

# Cached components - same instance returned (not thread-safe)
@module(CachingStrategy.NOT_THREAD_SAFE)
class ProdModule:
    service: MyService

# Cached components - same instance returned (thread-safe, uses locking)
@module(CachingStrategy.THREAD_SAFE)
class ThreadSafeProdModule:
    service: MyService

Multiple Decorator Integration

@law_of_demeter("_config")    # Creates forwarding properties
@law_of_demeter("_module")    # Auto-setup: self._config = self._module.config
class ResourceController:
    def __init__(self, module):
        self._module = module
        # Decorator automatically sets up: self._config = module.config
    
    # From _config
    _timeout: int
    _is_dry_run: bool
    
    # From _module  
    _api: object
    _namespace: str

Subtype Factory Generation

Use make[BaseType, ImplType] to program to an interface while the factory instantiates a concrete implementation:

from abc import ABC, abstractmethod
from reactor_di import module, make, CachingStrategy

class Logger(ABC):
    @abstractmethod
    def log(self, message: str) -> str: ...

class ConsoleLogger(Logger):
    def log(self, message: str) -> str:
        return f"[console] {message}"

@module(CachingStrategy.NOT_THREAD_SAFE)
class AppModule:
    logger: make[Logger, ConsoleLogger]   # factory returns ConsoleLogger()

app = AppModule()
assert isinstance(app.logger, ConsoleLogger)
assert isinstance(app.logger, Logger)

Nested Modules

Child modules can share dependencies from their parent using lookup[Type]:

from reactor_di import module, lookup, CachingStrategy

@module(CachingStrategy.NOT_THREAD_SAFE)
class CacheModule:
    db: lookup[DatabaseConnection]   # resolved from parent module
    cache_config: CacheConfig        # created locally

@module(CachingStrategy.NOT_THREAD_SAFE)
class AppModule:
    db: DatabaseConnection           # created here
    cache_module: CacheModule        # child module — db injected from here

app = AppModule()
assert app.cache_module.db is app.db  # same instance

An optional second parameter renames the lookup — the child uses a different local name than the parent's attribute:

@module(CachingStrategy.NOT_THREAD_SAFE)
class AuditModule:
    connection: lookup[DatabaseConnection, "db"]  # look up "db" on parent, bind as "connection"

On a component (non-module class), @law_of_demeter skips lookup-annotated attributes — the parent module's factory resolves them through the dependency-map mechanism instead of forwarding from the base reference.

lookup[T] and make[B, I] erase to T / B at type-check time (mirroring how Final[T] and ClassVar[T] erase to T), so attribute access works naturally in IDEs and mypy without cast().

Custom Prefixes

# No prefix - direct forwarding
@law_of_demeter('config', prefix='')
class DirectController:
    timeout: int        # → config.timeout
    is_dry_run: bool    # → config.is_dry_run

# Custom prefix
@law_of_demeter('config', prefix='cfg_')
class PrefixController:
    cfg_timeout: int        # → config.timeout
    cfg_is_dry_run: bool    # → config.is_dry_run

API Reference

Core Decorators

@module(strategy: CachingStrategy = CachingStrategy.DISABLED)

Creates a dependency injection module that automatically instantiates and provides dependencies.

Parameters:

• strategy: Caching strategy for component instances
  • CachingStrategy.DISABLED: Create new instances each time (default)
  • CachingStrategy.NOT_THREAD_SAFE: Cache instances (not thread-safe)
  • CachingStrategy.THREAD_SAFE: Cache instances with per-instance locking (thread-safe)

Usage:

@module(CachingStrategy.NOT_THREAD_SAFE)
class AppModule:
    config: Config
    service: Service  # Automatically injected with dependencies

@law_of_demeter(base_ref: str, prefix: str = "_")

Creates property forwarding from a base reference to avoid Law of Demeter violations.

Parameters:

• base_ref: Name of the base object attribute to forward from
• prefix: Prefix for forwarded property names (default: "_")

Usage:

@law_of_demeter("_config")
class Service:
    _timeout: int     # Forwards to _config.timeout
    _host: str        # Forwards to _config.host

Annotation Markers

lookup[Type]

Marks a module annotation as a dependency that should be resolved from the parent module rather than being created locally. On a module, @module skips factory generation and installs a lightweight property; the actual value is injected lazily by the parent. On a component, lookup is a no-op.

Parameters:

• First (required): The type of the dependency
• Second (optional): Name of the attribute to look up on the parent module. Defaults to the annotation's own name.

Usage:

@module(CachingStrategy.NOT_THREAD_SAFE)
class ChildModule:
    shared_db: lookup[DatabaseConnection]              # lookup "shared_db" on parent
    connection: lookup[DatabaseConnection, "db"]        # lookup "db" on parent, bind as "connection"
    local_service: MyService                           # created locally

make[BaseType, ImplType]

Marks a module annotation for subtype factory generation. The factory instantiates ImplType (a subtype of BaseType) instead of BaseType. This enables programming to interfaces while keeping concrete wiring in one place.

Parameters:

• First (required): The base type (used for the property return type)
• Second (required): The implementation type (actually instantiated by the factory)

Usage:

@module(CachingStrategy.NOT_THREAD_SAFE)
class AppModule:
    service: make[ServiceBase, HttpService]   # factory returns HttpService()
    cache: make[Cache, InMemoryCache]         # factory returns InMemoryCache()

Type Utilities

The simplified type utilities leverage Python 3.9+ stable type hint APIs:

get_alternative_names(name: str, prefix: str = "_") -> list[str]

Generates alternative names for dependency mapping (e.g., _config → config).

has_constructor_assignment(class_type: type[Any], attr_name: str) -> bool

Detects if a constructor assigns to an attribute using regex source analysis.

is_primitive_type(attr_type: type[Any]) -> bool

Identifies primitive types (int, str, bool, etc.) that shouldn't be auto-instantiated.

resolve_abstract_property_conflicts(cls: type[Any]) -> None

Replaces inherited abstract @property methods that collide with DI annotations, installing concrete dict-backed properties so lazy DI resolution via __getattr__ works correctly. Safe to call multiple times.

pure_hasattr(obj: Any, attr_name: str) -> bool

Checks if an attribute exists without side effects like triggering descriptors or properties. Used internally to avoid premature evaluation during dependency resolution.

Enums

CachingStrategy

Component caching strategies for the @module decorator.

• DISABLED = "disabled": No caching, create new instances each time
• NOT_THREAD_SAFE = "not_thread_safe": Cache instances (not thread-safe)
• THREAD_SAFE = "thread_safe": Cache instances with per-instance locking (thread-safe)

Development

This project uses modern Python tooling and best practices:

• Package manager: uv
• Testing: pytest with coverage
• Linting: ruff and black
• Type checking: mypy

Setup

1. Clone the repository
2. Install dependencies:

   uv sync
   

Running Tests

# Run all tests (116 tests: 53 examples + 63 regression/unit tests)
uv run pytest

# Run tests with coverage and HTML/terminal reports
uv run pytest --cov

# Run example tests only
uv run pytest examples/                     # Run all examples as tests (53 tests)
uv run pytest examples/caching_strategy.py  # Caching strategy examples (5 tests)
uv run pytest examples/custom_prefix.py     # Custom prefix examples (6 tests)
uv run pytest examples/quick_start.py       # Quick start examples (4 tests)

# Run regression/unit tests only
uv run pytest tests/                        # Run all regression tests (63 tests)

Debugging in PyCharm

The project is optimized for fast development and debugging:

1. Default: Tests run without coverage for fast feedback and debugging
2. Coverage analysis: Use --cov flag when you need coverage reports
3. Coverage threshold: Set to 90% to maintain high code quality (when coverage is enabled)

This configuration ensures breakpoints work reliably and tests run quickly during development.

Code Quality

# Run linting
uv run ruff check src tests examples
uv run black --check src tests examples

# Run type checking
uv run mypy src examples

# Fix formatting
uv run black src tests examples

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests for new functionality with meaningful assertions
5. Ensure all tests pass and 90% coverage is maintained
6. Verify realistic code scenarios rather than mocking impossible edge cases
7. Submit a pull request

License

This project is licensed under the MIT License - see the LICENSE file for details.