Python Configuration System

Description

Library to define configurations using python files.

Installation

Recommended installation with pip:

pip install pycs

Usage

1. Define config schema:

   project/config.py

   from pycs import CL, CN
   
   class BaseClass:
       pass
   
   schema = CN()  # Basic config node
   schema.DICT = CN()  # Nested config node
   schema.DICT.FOO = "FOO"  # Config leaf with actual value
   schema.DICT.INT = 1
   schema.NAME = CL(None, str, required=True)  # Specification of config leaf to be defined with type
   schema.CLASSES = CN(BaseClass)  # Config node with type specification of its config leaves
   schema.SUBCLASSES = CN(CL(None, BaseClass, subclass=True))  # Config node with subclass specification of its config leaves
   schema.VAL = CL(1, desc="Interesting description") # Config leaf with description
   
   def transform(cfg: CN) -> None:
       cfg.NAME = cfg.NAME or "__static__"
       cfg.DICT.FOO = "BAR"
   
   def validate(cfg: CN) -> None:
       assert len(cfg.NAME) > 0
   
   def hook(cfg: CN) -> None:
       print("Loaded")
   
   # Add transform, validation & hooks function
   # Transforms are run after config is loaded and can change values in config
   # Can also be run at runtime using .transform()
   # If you plan to transform multiple times we strongly recommend to make them idempotent
   schema.add_transform(transform)
   # Validators are run after transforms and freeze, with them you can verify additional restrictions
   schema.add_validator(validate)
   # Hooks are run after validators and can perform additional actions outside of config
   schema.add_hook(hook)
   # Validators and hooks should not (and mostly cannot) modify the config

2. If you want to use configuration with default values or make changes in the program you can use .static_init():

   from project.config import schema
   
   cfg = schema.static_init() # 'Loaded'
   print(cfg.DICT.FOO) # 'BAR'

3. If you want to store changes more permanently, please create a config file:

   my_cfg.py

   from pycs import CN
   
   from project.config import schema
   
   # Use init_cfg() to separate changes from base variable and freeze schema
   cfg = schema.init_cfg()
   
   # Schema changes are not allowed here, only leaves can be altered.
   cfg.NAME = "Hello World!"
   cfg.DICT.INT = 2

   You can also create another file to inherit from first and add more changes:

   my_cfg2.py

   from ntc import CN
   
   from .my_cfg import cfg
   
   # Separate changes from parent, important when inheriting in multiple files
   cfg = cfg.clone()
   cfg.DICT.FOO = "BAR"

   There are a few restrictions on imports in configs:

   • If you are inheriting changes from another config, please import variable as cfg
   • No other import should be named cfg

4. Load actual config and use it in the code.

   from pycs import CN
   
   cfg = CN.load("my_cfg.py")
   # Access values as attributes
   assert cfg.NAME == "Hello World!"
   assert cfg.DICT.FOO == "BAR"

Other features

• Load config changes/updates from YAML or JSON

  my_changes.yaml

  DICT:
    FOO: BAR

  my_changes.json

  {
    "DICT": {
      "INT": 2
    }
  }

  from project.config import schema
  
  cfg = schema.load_from_data_file("my_changes.yaml")
  assert cfg.DICT.FOO == "BAR"
  cfg = schema.load_from_data_file("my_changes.json")
  assert cfg.DICT.INT == 2

• Save loaded config to a file for tracking:

  cfg.save("saved.py")

  Please note: There are limitations on which configs can be saved. Generally, config schema should be defined at the module-level and then config should be created with one of:

  • CN.load()
  • schema.static_init()
  • schema.load_from_data_file()
  • schema.load_or_static()

  In addition, only basic changes should be applied to config after loading. See CfgSaveable for how to permit saving changes with more complex types.

Development

• Install dev dependencies: pip install -e ".[dev]"

• For linting and basic fixes ruff is used: ruff check . --fix

• This repository follows strict formatting style which will be checked by the CI.

  • To format the code, use the black format: black .
  • To sort the imports, user isort utility: isort .

• To test code, use pytest: pytest .

• This repository follows semantic-release, which means all commit messages have to follow a style. You can use tools like commitizen to write your commits.

• You can also use pre-commit to help verify that all changes are valid. Multiple hooks are used, so use the following commands to install:

  pre-commit install
  pre-commit install --hook-type commit-msg

Acknowledgements

This library was inspired by yacs.