Aperol Config

Aperol is a featherweight framework for configuration of Python objects based on dependency injection.

Usage

Requires Python >= 3.10.

Install Aperol from PyPI:

$ pip install aperol

Define a basic configuration file dummy.yaml:

imports:
  - random

a: -1
b.type: math.pi
c.type: uniform
d:
  type: uniform()
  a: 10
  b: 11

Python objects are specified by the type keyword. In the example above, uniform is resolved to the random module specified in the imports section. On the other hand, pi is explicitly imported from the math module.

Load the config file using Aperol and use the configured objects:

>>> import random
>>> random.seed(0)  # seed set for reproducibility
>>> import aperol
>>> configured = aperol.parse_config("dummy.yaml")
>>> configured["a"]
-1
>>> configured["b"]
3.141592653589793
>>> configured["c"]
_DelayedConstructor(factory=<bound method Random.uniform of <random.Random object at ...>>, init=False, kwargs={'a': -1, 'b': 3.141592653589793})
>>> configured["d"]
10.844421851525048
>>> uniform = configured["c"]
>>> uniform()
2.1391383869735945
>>> uniform()
0.7418361694776736

Package imports can be specified in one of three ways:

1. Within the config imports section,
2. passing search_pkgs to aperol.parse_config, or
3. registering imports globally with aperol.register_imports

In all three cases, each import must either be a string module name "X.Y" corresponding to import X.Y, or a tuple ("X.Y", "Z") corresponding to import X.Y as Z.

Configuration file locations can be registered using aperol.register_config_path.

Syntax

TODO (and TBC).

Related

Aperol is inspired by Gin Config. In comparison to Gin, Aperol allows configuration to be define using YAML (or any format which can be parsed to a nested dict tree). Thus it does not require learning a completely new syntax. Aperol is also simpler and does not provide all of the functionality of Gin; however, it is powerful enough to configure any Python object enabling flexible configuration.