Configuration Overview

The runtime configuration of a Script is built from several layers.

  • There is a ConfigTemplate that can have default values for certain config variables. These defaults are the basis of the config dict. (See ConfigTemplates for more details on ConfigTemplate).
  • The script can define an initial_config dict that is laid on top of the ConfigTemplate defaults, so any shared config variables are overwritten by the initial_config.
  • The ConfigTemplate.get_parser() method generates an argparse.ArgumentParser. This parser parses the commandline options.
  • If the commandline options specify any files via the --config-file option, then those files are read, and the contents are overlaid on top of the config. The first file specified will be overlaid first, then the second, and so on.
  • If the commandline options specify any optional config files via the --opt-config-file option, and if those files exist, then each existing file is read and the contents are overlaid on top of the config.
  • Finally, any other commandline options are overlaid on top of the config.

After the config is built, the script logs the config, and saves it to a localconfig.json file. This file can be inspected or reused for a later script run.


It’s very powerful to be able to build a configuration dict that can hold any key value pairs, but it’s non-trivial for users to verify if their config is valid or if there are options that they’re not taking advantage of.

To make the config more well-defined, we have the ConfigTemplate. The ConfigTemplate is comprised of ConfigVariable objects, and is based on the argparse.ArgumentParser, but with these qualities:

  • The ConfigTemplate can keep track of all config variables, including ones that aren’t available as commandline options. The option-less config variables must be specified via default, config file, or initial_config.
  • The templates can be added together, via ConfigTemplate.update().
  • Each ConfigVariable self-validates, and the ConfigTemplate makes sure there are no conflicting commandline options.
  • There is a ConfigTemplate.remove_option() method to remove a commandline option from the corresponding ConfigVariable. This may be needed if you want to add two config templates together, but they both have a -f commandline option specified, for example.
  • The ConfigTemplate.validate_config() method validates the built configuration. Each ConfigVariable can define whether they’re required, whether they require or are incompatible with other variables (required_vars and incompatible_vars), and each can define their own validate_cb callback function.
  • There is a ConfigTemplate.add_argument() for those who want to maintain argparse syntax.

Parent parsers are supported, to group commandline options in the --help output. Subparsers are not currently supported, though it may be possible to replace the ConfigTemplate.parser with a subparser-enabled parser at the expense of validation and the ability to ConfigTemplate.update().

When supporting downstream scripts, it’s best to keep each ConfigTemplate modular. It’s easy to combine them via ConfigTemplate.update(), but less trivial to remove functionality. The action config template, for instance, can be added to the base config template right before running parse_args().

LoggingDict and ReadOnlyDict

Each Script has a config dict. By default, this dict is a LoggingDict, which logs any changes made to the config.

For example, if the config looked like:

    "foo": 1,
    "bar": [2, 3, 4],
    "baz": {
        "z": 5,
        "y": 6,
        "x": 7,

then updating the config might log:

00:11:22  INFO - root.config['baz'] update: y now 8

Alternatively, someone could change the script class to StrictScript, which uses ReadOnlyDict. Once the ReadOnlyDict is locked, it cannot be modified.

By either explicitly logging any changes to the config, and/or preventing any changes to the config, it’s easier to debug any unexpected behavior.