hydr8 0.1.0

Decorator-based config injection for Hydra

hydr8

Decorator-based config injection for Hydra.

hydr8 lets you push Hydra config (or any config as long as it's a dict) values into function parameters automatically, so your functions stay clean and testable without manually threading cfg everywhere.

Installation

pip install hydr8
# or
uv add hydr8

Quick start

import hydra
from omegaconf import DictConfig
import hydr8

@hydr8.use("db")
def connect(host: str, port: int):
    print(f"Connecting to {host}:{port}")

@hydra.main(config_path="conf", config_name="config", version_base=None)
def main(cfg: DictConfig):
    hydr8.init(cfg)
    connect()  # host and port injected from cfg.db

if __name__ == "__main__":
    main()

With a config like:

db:
  host: localhost
  port: 5432

Usage

hydr8.use() is the core function. It can be used as a decorator to inject config into function parameters, or called directly to access a config sub-tree as a dict.

As a decorator

Explicit path

Pass a dot-separated path to resolve a specific config node. Config keys are matched to function parameter names. Extra config keys that don't match any parameter are silently ignored.

@hydr8.use("db.postgres")
def connect(host: str, port: int, user: str):
    ...

connect()              # all three injected from cfg.db.postgres
connect(host="remote") # host overridden, port and user from config

List indexing is supported:

@hydr8.use("db.replicas[0]")
def connect(host: str, port: int):
    ...

Implicit path (auto-resolve)

When no path is given, hydr8 derives it from the function's __module__. If the first segment of the module isn't a top-level config key, it's treated as the project name and stripped — so auto-resolve works whether you run with python -m or python file.py:

# In myproject/data/loaders.py
@hydr8.use()
def build_loader(batch_size: int, shuffle: bool):
    ...
# Resolves to cfg.data.loaders

# config.yaml
data:
  loaders:
    batch_size: 32
    shuffle: true

By default, scope="module" — the path resolves to the module's config node, and config keys are matched to function parameters. Multiple functions in the same module share the same config node.

With scope="fn", the function's qualname is appended to the path:

# In myproject/data/loaders.py
@hydr8.use(scope="fn")
def build_loader(batch_size: int, shuffle: bool):
    ...
# Resolves to cfg.data.loaders.build_loader

# config.yaml
data:
  loaders:
    build_loader:
      batch_size: 32
      shuffle: true

This works with methods too:

# In myproject/db/client.py
class Client:
    @hydr8.use(scope="fn")
    def __init__(self, host: str, port: int):
        self.host = host
        self.port = port
# Resolves to cfg.db.client.Client.__init__

as_dict mode

Pass the entire resolved sub-config as a single dict argument instead of matching individual keys:

@hydr8.use("db.postgres", as_dict="config")
def connect(config: dict):
    host = config["host"]
    port = config["port"]
    ...

Caller overrides

Caller-provided arguments always take precedence over injected config. If every required parameter is supplied by the caller, config is never accessed at all:

@hydr8.use("db")
def connect(host: str, port: int):
    ...

connect(host="remote")       # port from config, host = "remote"
connect("localhost", 5432)   # config not accessed

As a direct call

hydr8.use("path") returns a lazy, dict-like proxy. The config is resolved on first access, not at call time, so you can call use() before init().

import hydr8

db = hydr8.use("db")

hydr8.init(cfg)
db["host"]        # "localhost"
db["port"]        # 5432

This is useful when you want to read config values without decorating a function:

def connect():
    db = hydr8.use("db")
    engine = create_engine(f"postgresql://{db['host']}:{db['port']}")
    ...

An explicit path is required when using use() as a direct call. Calling use() without a path and accessing it raises TypeError, since there is no function to derive the path from.

Testing

Option A: Supply all arguments directly

When every required parameter is provided by the caller, config injection is skipped entirely — no init needed:

def test_connect():
    assert connect("localhost", 5432) == expected

Option B: override context manager

Temporarily replace the global config for a test:

from hydr8 import override

def test_connect():
    with override({"db": {"host": "test-host", "port": 9999}}):
        result = connect()
        assert result == expected

API reference

Function	Description
init(cfg)	Store the config globally (accepts any dict or OmegaConf DictConfig)
get()	Retrieve the stored config (raises RuntimeError if uninitialized)
override(overrides)	Context manager that temporarily replaces the config
use(path, *, as_dict, scope)	Decorator or direct config accessor for a config sub-tree