Skip to main content

An extension of argparse to configure Python with Python

Project description

magiconfig

Configure Python with Python.

Table of Contents

(Created by gh-md-toc)

Overview

magiconfig is an extension of argparse that stops the configuration complexity clock by enabling users to configure Python with Python. It provides all the power of Python to manipulate and compose configuration parameters, bypassing the limitations of text-based configuration languages.

Philosophy

This module treats argparse as an engine that ultimately provides a namespace of attributes ("dests") to be consumed by user applications. With magiconfig, these attributes can be provided by an imported MagiConfig Python object, in addition to the usual command-line arguments.

Features

magiconfig is compatible with both Python 2 and Python 3. It provides a custom ArgumentParser class, which is a drop-in replacement for argparse.ArgumentParser. It also provides MagiConfig, MagiConfigOptions, and MagiConfigError classes. The precedence of parameter values is: command line > config file > defaults.

ArgumentParser

The API of this class is extended with several additional functions to manage config settings, as well as to provide other useful operations.

Constructor

The constructor supports several additional options:

  • config_options: takes an instance of MagiConfigOptions; default = None (falls back to standard argparse behavior)
  • config_only_help: include config-only args in the help message (see add_config_only()); default = True

parse_args(), parse_known_args()

These function interfaces are unchanged from argparse, but they return a MagiConfig object. If an input namespace argument is provided but is not of type MagiConfig, a conversion will be attempted.

parse_config(config_name, config_obj, config_strict, namespace=None)

This is mainly an internal function used in parse_known_args(), but like that function, it could also be used standalone.

  • config_name: name of config file to import
  • config_obj: name of config object inside config file
  • config_strict: whether to reject imported config object if it has unknown attributes
  • namespace: Namespace object to append to, if any

Raises MagiConfigError if any required config-only arguments are missing or if config_strict is True and the imported config has unknown attributes.

set_config_options(**kwargs)

This function allows changing the config options after the parser is initialized. It accepts all parameters that can be used to construct an instance of MagiConfigOptions. Raises MagiConfigError if any other parameters are provided.

copy_config_options(config_options)

This function allows copying another MagiConfigOptions object to be used in this parser. Raises MagiConfigError if an object with any unknown parameters is provided.

remove_config_options()

This function allows removing all config options from the parser.

write_config(namespace, filename, obj=None)

  • namespace: MagiConfig object to be written
  • filename: name of file to write
  • obj: name of the MagiConfig object in the file (default: class member config_options.obj or "config" if no config_options specified)

This function can be used to preserve the state of the configuration after any command-line modifications (see Example 1).

add_config_only(*args, **kwargs)

This interface allows adding dests that are only provided by the config, not by command-line arguments.

  • args: no default value, not required
  • **kwargs: default value OR required (value=None)

Raises MagiConfigError if any dests have already been used by arguments (actions) added to the parser. Similarly, add_argument() now raises ArgumentError if it specifies a dest that has already been added as config-only by this function.

remove_config_only(arg)

  • arg: name of config-only arg to remove

Raises KeyError if arg is not found in the list of config-only args.

remove_argument(arg, keep=False)

This function allows removing a single argument (a missing feature in argparse).

  • arg: flag (for optional arguments) or dest (for positional arguments)
  • keep: for optional arguments, True will remove just the single specified flag arg (the entire action is removed only if it has no remaining flags); False (default) always removes the action associated with the flag
  • for positional arguments, all positional actions with the specified dest are removed (but not optional arguments with that dest)

Exits with error() if an unknown argument is provided.

MagiConfigOptions

This simple class stores options related to the use of configs in the ArgumentParser.

Constructor

  • args: the command-line arguments to indicate the config file (default: ["-C", "--config"])
  • help: custom help message for config args (optional)
  • required: if the config arg is required when parsing (default: False)
  • default: default value for the config file name (default: None)
  • dest: destination for config arg (default: "config")
  • obj: name of the MagiConfig object to be imported from the config file (default: "config")
  • obj_args: command-line arguments to indicate the name of the object to be imported (optional)
  • obj_help: custom help message for obj args (optional)
  • obj_dest: destination for obj arg (default: "obj")
  • strict: reject imported config object if it has unknown attributes (default: False)
  • strict_args: optional command-line arguments to toggle strictness
    • if strict above is set to False, providing an arg will toggle it to True; if set to True, will toggle it to False
  • strict_help: custom help message for strict args (optional)
  • strict_dest: destination for strict arg (default: "strict")

The values for args, obj_args, and strict_args can be positional arguments (rather than the optional arguments shown here).

MagiConfig

This class extends argparse.Namespace to add a few useful methods. It is used both as the input object in config files and as the output object of ArgumentParser.

write(filename, config_obj)

  • filename: name of file to write
  • config_obj: name of MagiConfig object in file

join(other_config, prefer_other=False)

  • other_config: other MagiConfig object to merge
  • prefer_other: prefer values from other config, if dest is present in both configs (default: prefer this config)

getattr(), setattr()

These class methods are extended to handle nested config objects automatically. Any nonexistent intermediate objects are initialized as MagiConfig instances. Example:

from magiconfig import MagiConfig
x = MagiConfig()
setattr(x,"y.z","test")
print(x)

returns: MagiConfig(y=MagiConfig(z='test'))

This enables obtaining dests from nested configs by using dots in the dest names.

MagiConfigError

This class derives from Exception and denotes magiconfig-specific errors.

Other

Subparser aliases

_SubParsersAction.add_parser is modified to backport the use of subparsers aliases to Python 2.

Convenience

All public classes and constants from argparse are added to the magiconfig namespace for easier drop-in usage.

A class ArgumentDefaultsRawHelpFormatter is defined to present help messages with default values and without line wrapping (from ConfigArgParse).

Examples

1) Basic setup

The simple script in examples/example1.py demonstrates the different ways to set values, as well as some of the features of magiconfig.

The help printout for the arguments defined in the script:

usage: example1.py [-h] [-C CONFIG] [-f FOO] -b BAR [-i]

optional arguments:
  -h, --help            show this help message and exit
  -C CONFIG, --config CONFIG
                        name of config file to import (w/ object: config)
  -f FOO, --foo FOO     foo arg
  -b BAR, --bar BAR     bar arg
  -i, --ipsum           ipsum arg

When the script is run as follows:

python3 examples/example1.py -C examples/config1.py --foo 'foo'

It prints the resulting namespace:

MagiConfig(bar=3.0, foo='foo', ipsum=False)

Here, the bar argument is set by the config file examples/config1.py, the foo argument is set on the command line, and the ipsum argument retains its default value.

The script also writes the final namespace into a config file examples/config1_out.py:

from magiconfig import MagiConfig

config = MagiConfig()
config.bar = 3.0
config.foo = 'foo'
config.ipsum = False

With this config file, the script can be rerun to produce the same output without the need to specify any other command-line arguments: python examples/example1.py -C examples/config1_out.py.

2) Subparsers

The script in examples/example2.py demonstrates how a common config file examples/config2.py can be used with multiple subparsers.

The parser has two modes defined, one (with an argument foo) and two (with an argument bar). Each subparser mode specifies a different config object; in this case, each of these config objects is a member of a top-level config object.

The script can be run in each mode with the same input config file:

> python examples/example2.py one -C examples/config2.py
MagiConfig(foo='foo')
> python examples/example2.py two -C examples/config2.py
MagiConfig(bar=2.0)

3) Config-driven

In a config-driven script, it may be desirable to encapsulate many parameters only in the config file, while supporting only parameters related to running the script as command-line arguments. The script in examples/example3.py is an example.

It shows how an organized schema with different categories and parameters can be defined and transmitted to the parser. This allows the parser to use strict mode to validate input configurations, rejecting any config with unknown parameters. The config file examples/config3.py can be used with the script:

> python examples/example3.py -C examples/config3.py -v
MagiConfig(dataset=MagiConfig(background='background', path='/data', signal='signal'), hyper=MagiConfig(learning_rate=0.1, loss='log'), training=MagiConfig(size=0.5, weights=[1, 1]), verbose=True)

This example also shows how config-only arguments can be given default values or marked as required. These attributes are reflected in the help message:

> python examples/example3.py --help
usage: example3.py [-h] [-C CONFIG] [-v]

optional arguments:
  -h, --help            show this help message and exit
  -C CONFIG, --config CONFIG
                        name of config file to import (w/ object: config) (default: None)
  -v, --verbose         enable verbose output (default: False)

config-only arguments:
  dataset.background    (required)
  dataset.path            (default: /data)
  dataset.signal        (required)
  hyper.learning_rate
  hyper.loss
  training.size
  training.weights

4) Scaling up

When scaling up an application to handle a large number of possible inputs, a typical pattern is that some of the parameters are common, while other parameters may be unique to each input. Rather than requiring a separate config file for each possible input, all of the config objects can be generated within a single Python file. The script in examples/example4.py allows the config object name to be specified on the command line; other config objects in the config file are just ignored.

The help message for this script is:

usage: example4.py [-h] [-C CONFIG] [-O OBJ] [-f FOO] -b BAR -i INPUT

optional arguments:
  -h, --help            show this help message and exit
  -C CONFIG, --config CONFIG
                        name of config file to import (w/ object from -O,--obj) (default: None)
  -O OBJ, --obj OBJ     name of object to import from config file (default: config)
  -f FOO, --foo FOO     foo arg (default: lorem)
  -b BAR, --bar BAR     bar arg (default: None)
  -i INPUT, --input INPUT
                        input arg (default: None)

The script can be run with different inputs all contained in examples/config4.py:

> python3 examples/example4.py -C examples/config4.py -O config.a
MagiConfig(bar=3.0, foo='foo', input='a')
> python3 examples/example4.py -C examples/config4.py -O config.b
MagiConfig(bar=3.0, foo='foo', input='b')

Inspirations

This project owes inspiration (and in some cases code) to:

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

magiconfig-2.2.0.tar.gz (18.7 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page