TL;DR

# config.yml
foo:
  bar: <% ENV[REQUIRED] %>
  baz: <% ENV[OPTIONAL, true] %>
list_of_stuff:
  - fun<% ENV[NICE, dament] %>al
  - fun<% ENV[AGH, er] %>al
  - more/<% ENV[THAN, er] %>/one/<% ENV[interpolation, er] %>!

# app.py
from configly import Config


config = Config.from_yaml('config.yml')

print(config.foo.bar)
print(config.foo['baz'])
for item in config.list_of_stuff:
    print(item)

pip install configly[yaml]

Introduction

Loading configuration is done in every (application) project, and yet it is often overlooked and condidered too easy or straightforward to bother using a library to manage doing it.

Therefore, we often see code like this:

# config.py
import os

# Maybe it's following 12factor and loading all the config from the environment.
config = {
    'log_level': os.getenv('LOG_LEVEL'),
    'database': {
        # At least here, I can nest values if I want to organize things.
        'password': os.environ['DATABASE_PASSWORD'],
        'port': int(os.environ['DATABASE_PORT']),
    }
}

or this

# config.py
import os

class Config:
    log_level = os.getenv('LOG_LEVEL')

    # Here it's not so easy to namespace
    database_password = os.environ['DATABASE_PASSWORD']
    database_port = int(os.environ['DATABASE_PORT'])


# Oh goodness!
class DevConfig(Config):
    environment = 'dev'

or this

import configparser
# ...🤢... Okay I dont even want to get into this one.

And this is all assuming that everyone is loading configuration at the outermost entrypoint! The two worst possible outcomes in configuration are:

• You are loading configuration lazily and/or deeply within your application, such that it hits a critical failure after having seemingly successfully started up.
• There is not a singular location at which you can go to see all configuration your app might possibly be reading from.

The pitch

Configly asserts configuration should:

• Be centralized
  • One should be able to look at one file to see all (env vars, files, etc) which must exist for the application to function.
• Be comprehensive
  • One should not find configuration being loaded secretly elsewhere
• Be declarative/static
  • code-execution (e.g. the class above) in the definition of the config inevitably makes it hard to interpret, as the config becomes more complex.
• Be namespacable
  • One should not have to prepend foo_ namespaces to all foo related config names
• Be loaded, once, at app startup
  • (At least the definition of the configuration you're loading)
• (Ideally) have structured output
  • If something is an int, ideally it would be read as an int.

To that end, the configly.Config class exposes a series of classmethods from which your config can be loaded. It's largely unimportant what the input format is, but we started with formats that deserialize into at least str, float, int, bool and None types.

from configly import Config


# Currently supported input formats.
config = Config.from_yaml('config.yml')
config = Config.from_json('config.json')
config = Config.from_toml('config.toml')

Given an input config.yml file:

# config.yml
foo:
  bar: <% ENV[REQUIRED] %>
  baz: <% ENV[OPTIONAL, true] %>
list_of_stuff:
  - fun<% ENV[NICE, dament] %>al
  - fun<% ENV[AGH, er] %>al
  - more/<% ENV[THAN, er] %>/one/<% ENV[interpolation, er] %>!

A number of things are exemplified in the example above:

• Each <% ... %> section indicates an interpolated value, the interpolation can be a fragment of the overall value, and multiple values can be interpolated within a single value.

• ENV is an "interpolator" which knows how to obtain environment variables

• [VAR] Will raise an error if that piece of config is not found, whereas [VAR, true] will default VAR to the value after the comma

• Whatever the final value is, it's interpreted as a literal value in the format of the file which loads it. I.E. true -> python True, 1 -> python 1, and null -> python None.

Now that you've loaded the above configuration:

# app.py
from configly import Config


config = Config.from_yaml('config.yml')

# You can access namespaced config using dot access
print(config.foo.bar)

# You have use index syntax for dynamic, or non-attribute-safe key values.
print(config.foo['baz'])

# You can iterate over lists
for item in config.list_of_stuff:
    print(item)

# You can *generally* treat key-value maps as dicts
for key, value in config.foo.items():
    print(key, value)

# You can *actually* turn key-value maps into dicts
dict(config.foo) == config.foo.to_dict()

Installing

# Basic installation
pip install configly

# To use the yaml config loader
pip install configly[yaml]

# To use the toml config loader
pip install configly[toml]

# To use the vault config loader
pip install configly[vault]