Configuration files for Python

config, configuration
pip install configurati==0.2.3



configurati is a configuration file library for python,

  • configuration files json, yaml, or python modules
  • command-line overrides of configuration file
  • configuration including optional variables, type coercion, and collections


configurati can be install from PyPi via pip or easy_install

$ easy_install configurati
$ pip install configurati

Quick Start

I want to configure my awesome app with details about my cats. I'll begin by writing my configuration file in Python,

from datetime import datetime

cats = [
  {'name' : 'Mittens', 'age' : 18},
  {'name' :     'Bob', 'age' :  4}

owner = {
  'name'     : 'Config Uratii',
  'address'  : '1234 Python Dr.',
  'birthday' : datetime(1970, 1, 1).isoformat()

To use this configuration, all I need to do is call configurati.configure in my application...

from pprint import pprint
from configurati import configure

if __name__ == '__main__':

...and tell it which configuration file to use from the command line,

$ python \

But that's not quite right. I misspelled my name, and today is Bob's birthday!

$ python         \
  --config            \
  "" "Config Urati" \
  "--cats[1].age" "5"

Oh, and I forgot about my dogs!

$ python         \
  --config            \
  "" "Config Urati" \
  "--cats[1].age" "5"           \
  "--dogs[0]" '{"name": "Sir Barks-a-lot", "age": 15}'

Actually, I don't like defining configuration files in Python. Let me do it in a more language-agnostic format,


- name: "Mittens"
  age: 18
- name: "Bob"
  age: 4

  name: "Config Uratii"
  address: "1234 Python Dr."
  "birthay": "`from datetime import datetime; datetime(1970, 1, 1).isoformat()`"

When I load this file with configurati, all text in backticks will be evaluated, but what if I want to use it in another language? I need to see the evaluated config file!

$ cat config.yaml | python -m configurati.interpolate > clean_config.yaml

Finally! Configuration complete. Now Let's actually use it!

from configurati import configure

if __name__ == '__main__':
  c = configure()

  # using bracket notation
  c['cats'][0]['age']   # 18

  # notation
  c.dogs[0].name        # "Sir Barks-a-lot"

  # ...fancy bracket notation
  c['cats[0].age']      # guess what...18

...and that's configurati!

Defining a Configuration

configurati is designed with flexibility in mind. You should be able to express everything you want in a config without being restricted by configuration format or data type. Furthermore, mixing, matching, and overriding parts of configs should be simple.

In configurati, you can specify a config in YAML, JSON, Python, or from the command line. Command line parameters directly override a config file's contents, so you can easily test out variations without generating a host of different files.


The most direct way to specify a config is as a Python module. Nearly all valid Python modules are also valid configurations (caveat: dict keys must be valid python identifiers). configurati does not restrict the types available in configuration, so modules, functions, and objects are all fair game.

import datetime
import os
from configurati import *

# merge multiple config files by importing them into the current space. Notice
# that you can import any configuration file that `configurati` understands.

# alternatively, load them as dicts
server = load_config("server_config.json")

# retrieve environment variables
environment = "dev" if env("ENVIRONMENT") is None else env("ENVIRONMENT")

# collection data types
collection_config = {
  'list_config': [1, 2, 3],
  'tuple_config': ('a', 'b', 'c'),
  'dict_config': {'a': 1, 'b': 2}

# python-only objects
date =

# valid config names have to start start with a letter and can only contain
# letters, numbers, and underscores. This variable will be ignored.
_ignored = True


YAML and JSON are two common language-agnostic data formats which allow one to express common atomic (numbers, strings) and collection (mapping, list) data types natively. configurati extends these formats via string interpolation to allow for the expression of arbitrary Python objects.

At this time, import_config does not work in these formats. Use from configurati import *; load_config("other_config.yaml").to_dict() instead.

  list_config: [1, 2, 3]
  tuple_config: "`('a', 'b', 'c')`"
    a: 1
    b: 2

date: "`import datetime;`"

server: "`from configurati import *; load_config(\"server_config.json\").to_dict()"

Loading and Using a Configuration

Configuration files can be loaded in one of two ways,

  1. By calling configurati.configure() in your code and passing in --config <path to config file> from the command line
  2. By explicitly using configurati.configure(config="/path/to/")

A configuration object is an instance of configurati.attrs.attrs, a dict-like object designed to make accessible attributes easier.

from configurati.attrs import attrs
from configurati import Missing

config = attrs.from_dict(
      'collection_config': {
        'list_config': [1, 2, 3],
        'tuple_config': ('a', 'b', 'c'),
        'dict_config': { 'a': 1, 'b': 2 }

# dot notation access
config.collection_config.list_config[2]           # 3

# nested dicts are also attrs instances
dict_config = config.collection_config.dict_config
dict_config.a                                     # 1

# bracket notation access
config['collection_config']['dict_config']['a']   # 1

# fancy bracket notation access
config['collection_config.tuple_config[2]']       # 'c'

# ...used in modifying values
config['collection_config.tuple_config[2]'] = 100
config.collection_config.tuple_config[2]          # 100

# ...and constructing new ones
config['x.y[2].z'] = "Hello"
config.x.y[2]                                     # { "z": "hello" }
config.x.y[0] is Missing                          # True

Defining Configuration Specifications

When writing a program to be configured by a user, it is helpful to validate said configuration and supply useful defaults. configurati allows one to define configuration specifications as a Python module.

from configurati import *

# variables in a spec are either required or optional. both take an optional
# `type` argument -- a function used to coerce the entered value to the desired
# type. `optional` variables also take an optional `default` argument -- the
# value to use if no value is entered.
name = required(type=str)
description = optional(type=str, default="")

# tuples are used to convey a fixed-length list of values. If all of its
# contents are optional, a tuple is as well; otherwise, it is required.
version = (
    required(type=int, help="major version"),
    optional(type=str, default="SNAPSHOT")

# valid config keys must start with a letter and can only contain letters,
# numbers, and underscores. keys starting with an underscore are a good way of
# defining reusable, partial specs.
_server = {
    'host': optional(type=str, default="localhost"),
    "port": optional(type=int, default=8888)

# like tuples, dicts are optional if all their contents are optional.
server = _server

# as the `password` field is required, the config must specificy a database
# object.
database = {
    'username': optional(type=str, default=env("USER")),
    'password': required(type=str),

# lists represent an unknown number of objects of the same type. specs for
# lists are specified by entering a single spec in its contents.
caches = [
      "host"    : optional(type=str, default="localhost"),
      "backend" : optional(type=one_of("redis", "memcached"), default="redis")

# as with configs, you can import or load other specs. If
# `relative_to_caller=True` (the default), then the path is assumed to be
# relative to this file, rather than the caller
import_spec("", relative_to_caller=True)
other_spec = load_spec("", relative_to_caller=True)

Below is an (incomplete) sample configuration

from configurati import *

# `ignored` isn't specified in the specification, so it won't show up in the
# loaded config
ignored = True

# since description is unspecified, its value will be the default -- ""
name = 'sample'

# since database.password is unspecified, attempts at loading this config
# without additional command line arguments will throw a ValidationError
# database = { 'password': 'PASSWORD' }

# to use default values in a tuple, use `Missing`
version = (1, Missing)

# is unspecified, so its default value will be used -- 8888
server = {
  'host': ''

caches = [
  {},                       # {"host": "localhost", "backend":     "redis"}
  {"backend": "memcached"}, # {"host": "localhost", "backend": "memcached"}

As with configuration files, you may choose a spec in one of two ways,

  1. via configurati.configure(spec="/path/to/")
  2. via command line arguments, "--spec "

from pprint import pprint
from configurati import configure

pprint(configure(config='', spec=''))

To complete the config, we'll specify the database password from the command line,

$ python '--database.password' 'PASSWORD'
{'caches': [{'backend': 'redis', 'host': 'localhost'},
            {'backend': 'memcached', 'host': 'localhost'}],
 'database': {'host': 'localhost',
              'password': 'PASSWORD',
              'port': 8888,
              'username': 'duckworthd'},
 'description': '',
 'name': 'sample',
 'server': {'host': '', 'port': 8888},
 'version': (1, 'SNAPSHOT')}