childsish / dynamic-json

Use self-referential JSON files as python configuration files

dynamic-json

Dynamic-json allows self-referential entries in a json file. It requires no dependencies, but lacks all the nice features of the yaml specification. If you prefer yaml then see [dynamic-yaml][dynamic-yaml].

Usage

The key feature that was introduced is the ability for a string scalar to reference other parts of the configuration tree. This is done using the Python string formatting syntax. The characters '{' and '}' enclose a reference to another entry in the configuration structure. The reference takes the form key1.key2 where key1 maps to another mapping object and can be found in the root mapping, and key2 can be found in key1's mapping object. Multiple levels of nesting can be used (eg. key1.key2.key3 etc...).

An example yaml configuration:

{
  "project_name": "hello-world",
  "dirs": {
    "home": "/home/user",
    "venv": "{dirs.home}/venvs/{project_name}",
    "data": "{dirs.venv}/data",
    "output": "{dirs.data}/output-{parameters.parameter1}-{parameters.parameter2}"
  },
  "parameters": {
    "parameter1": "a",
    "parameter2": "b"
  }
}

Reading in a json file:

import dynamic_json

with open('/path/to/file.json') as fileobj:
    cfg = dynamic_json.load(fileobj)
    assert cfg.dirs.venv == '/home/user/venvs/hello-world'
    assert cfg.dirs.output == '/home/user/venvs/hello-world/data/output-a-b'

As the variables are dynamically resolved, it is also possible to combine this with argparse:

import dynamic_json

from argparse import ArgumentParser

with open('/path/to/file.yaml') as fileobj:
    cfg = dynamic_json.load(fileobj)
    parser = ArgumentParser()
    parser.add_argument('--parameter1')
    parser.add_argument('--parameter2')
    parser.parse_args('--parameter1 c --parameter2 d'.split(), namespace=cfg.parameters)
    assert cfg.dirs.output == '/home/user/venvs/hello-world/data/output-c-d'

Installation

To install, simply run:

pip install dynamic-json

Restrictions

Due to the short amount of time I was willing to spend on working upon this, there are a few restrictions that I could not overcome.

• Certain keys can only be used via __getitem__ and not __getattr__. Because dict comes with it's own set of attributes that are always resolved first, the values for the following keys must be gotten using the item getter rather than the attribute getter (eg. config['items'] vs. config.items):
  • append
  • extend
  • insert
  • remove
  • pop
  • clear
  • index
  • count
  • sort
  • reverse
  • copy