scriptconfig 0.7.10

Installation

The scriptconfig. package can be installed via pip:

pip install scriptconfig

Example Script

Scriptconfig is used to define a flat configuration dictionary with values that can be specified via Python keyword arguments, command line parameters, or a yaml config file. Consider the following script that prints its config, opens a file, computes its hash, and then prints it to stdout.

import scriptconfig as scfg
import hashlib
import ubelt as ub


class FileHashConfig(scfg.Config):
    """
    The docstring will be the description in the CLI help
    """
    __default__ = {
        'fpath': scfg.Value(None, position=1, help='a path to a file to hash'),
        'hasher': scfg.Value('sha1', choices=['sha1', 'sha512'], help='a name of a hashlib hasher'),
    }


def main(**kwargs):
    config = FileHashConfig(data=kwargs, cmdline=True)
    print('config = {!r}'.format(config))
    fpath = config['fpath']
    hasher = getattr(hashlib, config['hasher'])()

    with open(fpath, 'rb') as file:
        hasher.update(file.read())

    hashstr = hasher.hexdigest()
    print('The {hasher} hash of {fpath} is {hashstr}'.format(
        hashstr=hashstr, **config))


if __name__ == '__main__':
    main()

If this script is in a module hash_demo.py (e.g. in the examples folder of this repo), it can be invoked in these following ways.

Purely from the command line:

# Get help
python hash_demo.py --help

# Using key-val pairs
python hash_demo.py --fpath=$HOME/.bashrc --hasher=sha1

# Using a positional arguments and other defaults
python hash_demo.py $HOME/.bashrc

From the command line using a yaml config:

# Write out a config file
echo '{"fpath": "hashconfig.json", "hasher": "sha512"}' > hashconfig.json

# Use the special `--config` cli arg provided by scriptconfig
python hash_demo.py --config=hashconfig.json

# You can also mix and match, this overrides the hasher in the config with sha1
python hash_demo.py --config=hashconfig.json --hasher=sha1

Lastly you can call it from good ol’ Python.

import hash_demo
hash_demo.main(fpath=hash_demo.__file__, hasher='sha512')

Example Script (New Syntax)

NEW in 0.6.2: there is now a more concise syntax available using a scriptconfig.DataConfig. The equivalent version of the above code is:

import scriptconfig as scfg
import hashlib


class FileHashConfig(scfg.DataConfig):
    """
    The docstring will be the description in the CLI help
    """
    fpath = scfg.Value(None, position=1, help='a path to a file to hash')
    hasher = scfg.Value('sha1', choices=['sha1', 'sha512'], help='a name of a hashlib hasher')


def main(**kwargs):
    config = FileHashConfig.cli(data=kwargs)
    print('config = {!r}'.format(config))
    fpath = config['fpath']
    hasher = getattr(hashlib, config['hasher'])()

    with open(fpath, 'rb') as file:
        hasher.update(file.read())

    hashstr = hasher.hexdigest()
    print('The {hasher} hash of {fpath} is {hashstr}'.format(
        hashstr=hashstr, **config))


if __name__ == '__main__':
    main()

This can be invoked from the examples folder similarly to the above script (replace hash_data.py with hash_data_datconfig.py.)

Project Design Goals

• Write Python programs that can be invoked either through the commandline or via Python itself.

• Drop in replacement for any dictionary-based configuration system.

• Intuitive parsing (currently working on this), ideally improve on argparse if possible. This means being able to easily specify simple lists, numbers, strings, and paths.

To get started lets consider some example usage:

>>> import scriptconfig as scfg
>>> # In its simplest incarnation, the config class specifies default values.
>>> # For each configuration parameter.
>>> class ExampleConfig(scfg.Config):
>>>     __default__ = {
>>>         'num': 1,
>>>         'mode': 'bar',
>>>         'ignore': ['baz', 'biz'],
>>>     }
>>> # Creating an instance, starts using the defaults
>>> config = ExampleConfig()
>>> # Typically you will want to update default from a dict or file.  By
>>> # specifying cmdline=True you denote that it is ok for the contents of
>>> # `sys.argv` to override config values. Here we pass a dict to `load`.
>>> kwargs = {'num': 2}
>>> config.load(kwargs, cmdline=False)
>>> assert config['num'] == 2
>>> # The `load` method can also be passed a json/yaml file/path.
>>> config_fpath = '/tmp/foo'
>>> open(config_fpath, 'w').write('{"num": 3}')
>>> config.load(config_fpath, cmdline=False)
>>> assert config['num'] == 3
>>> # It is possible to load only from CLI by setting cmdline=True
>>> # or by setting it to a custom sys.argv
>>> config.load(cmdline=['--num=4'])
>>> assert config['num'] == 4
>>> # Note that using `config.load(cmdline=True)` will just use the
>>> # contents of sys.argv

Notice in the above example the keys in your default dictionary are command line arguments and values are their defaults. You can augment default values by wrapping them in scriptconfig.Value objects to encapsulate information like help documentation or type information.

>>> import scriptconfig as scfg
>>> class ExampleConfig(scfg.Config):
>>>     __default__ = {
>>>         'num': scfg.Value(1, help='a number'),
>>>         'mode': scfg.Value('bar', help='mode1 help'),
>>>         'mode2': scfg.Value('bar', type=str, help='mode2 help'),
>>>         'ignore': scfg.Value(['baz', 'biz'], help='list of ignore vals'),
>>>     }
>>> config = ExampleConfig()
>>> # smartcast can handle lists as long as there are no spaces
>>> config.load(cmdline=['--ignore=spam,eggs'])
>>> assert config['ignore'] == ['spam', 'eggs']
>>> # Note that the Value type can influence how data is parsed
>>> config.load(cmdline=['--mode=spam,eggs', '--mode2=spam,eggs'])

The above examples are even more concise with the new DataConfig syntax.

>>> import scriptconfig as scfg
>>> # In its simplest incarnation, the config class specifies default values.
>>> # For each configuration parameter.
>>> class ExampleConfig(scfg.DataConfig):
>>>     num = 1
>>>     mode = 'bar'
>>>     ignore = ['baz', 'biz']
>>> # Creating an instance, starts using the defaults
>>> config = ExampleConfig()
>>> assert config['num'] == 1
>>> # Or pass in known data. (load as shown in the original example still works)
>>> kwargs = {'num': 2}
>>> config = ExampleConfig.cli(default=kwargs, cmdline=False)
>>> assert config['num'] == 2
>>> # The `load` method can also be passed a json/yaml file/path.
>>> config_fpath = '/tmp/foo'
>>> open(config_fpath, 'w').write('{"mode": "foo"}')
>>> config.load(config_fpath, cmdline=False)
>>> assert config['num'] == 2
>>> assert config['mode'] == "foo"
>>> # It is possbile to load only from CLI by setting cmdline=True
>>> # or by setting it to a custom sys.argv
>>> config = ExampleConfig.cli(argv=['--num=4'])
>>> assert config['num'] == 4
>>> # Note that using `config.load(cmdline=True)` will just use the
>>> # contents of sys.argv

Features

• Serializes to json

• Dict-like interface. By default a Config object operates independent of config files or the command line.

• Can create command line interfaces

  • Can directly create an independent argparse object

  • Can use special command line loading using self.load(cmdline=True). This extends the basic argparse interface with:

    • Can specify options as either --option value or --option=value

    • Default config options allow for “smartcasting” values like lists and paths

    • Automatically add --config, --dumps, and --dump CLI options when reading cmdline via load.

• Inheritence unions configs.

Gotchas

CLI Values with commas:

When using scriptconfig to generate a command line interface, it uses a function called smartcast to try to determine input type when it is not explicitly given. If you’ve ever used a program that tries to be “smart” you’ll know this can end up with some weird behavior. The case where that happens here is when you pass a value that contains commas on the command line. If you don’t specify the default value as a scriptconfig.Value with a specified type, if will interpret your input as a list of values. In the future we may change the behavior of smartcast, or prevent it from being used as a default.

Boolean flags and positional arguments:

scriptconfig always provides a key/value way to express arguments. However, it also recognizes that sometimes you want to just type --flag and not --flag=1. We allow for this for Values with isflag=1, but this causes a corner-case ambituity with positional arguments. For the following example:

class MyConfig(scfg.DataConfig):
    arg1 = scfg.Value(None, position=1)
    flag1 = scfg.Value(False, isflag=True, position=1)

For --flag 1 We cannot determine if you wanted {'arg1': 1, 'flag1': False} or {'arg1': None, 'flag1': True}.

This is fixable by either using strict key/value arguments, expressing all positional arguments before using flag arguments, or using the `` – `` construct and putting all positional arguments at the end. In the future we may raise an AmbiguityError when specifying arguments like this, but for now we leave the behavior undefined.

FAQ

Question: How do I override the default values for a scriptconfig object using json file?

Answer: This depends if you want to pass the path to that json file via the command line or if you have that file in memory already. There are ways to do either. In the first case you can pass --config=<path-to-your-file> (assuming you have set the cmdline=True keyword arg when creating your config object e.g.: config = MyConfig(cmdline=True). In the second case when you create an instance of the scriptconfig object pass the default=<your dict> when creating the object: e.g. config = MyConfig(default=json.load(open(fpath, 'r'))). But the special --config --dump and --dumps CLI arg is baked into script config to make this easier.

Related Software

I’ve never been completely happy with existing config / argument parser software. I prefer to not use decorators, so click and to some extend hydra are no-gos. Fire is nice when you want a really quick CLI, but is not so nice if you ever go to deploy the program in the real world.

The builtin argparse in Python is pretty good, but I with it was easier to do things like allowing arguments to be flags or key/value pairs. This library uses argparse under the hood because of its stable and standard backend, but that does mean we inherit some of its quirks.

The configargparse library - like this one - augments argparse with the ability to read defaults from config files, but it has some major usage limitations due to its implementation and there are better options (like jsonargparse). It also does not support the use case of calling the CLI as a Python function very well.

The jsonargparse library is newer than this one, and looks very compelling. I feel like the definition of CLIs in this library are complementary and I’m considering adding support in this library for jsonargparse because it solves the problem of nested configurations and I would like to inherit from that. Keep an eye out for this feature in future work.

Hydra - https://hydra.cc/docs/intro/

OmegaConf - https://omegaconf.readthedocs.io/en/latest/index.html

Argparse - https://docs.python.org/3/library/argparse.html

JsonArgparse - https://jsonargparse.readthedocs.io/en/stable/index.html

Fire - https://pypi-hypernode.com/project/fire/

Click - https://pypi-hypernode.com/project/click/

ConfigArgparse - https://pypi-hypernode.com/project/ConfigArgParse/

TODO

• [ ] Nested Modal CLI’s

• [X] Policy on nested heirachies (currently disallowed) - jsonargparse will be the solution here.

  • [ ] How to best integrate with jsonargparse

• [ ] Policy on smartcast (currently enabled)

  • [ ] Find a way to gracefully way to make smartcast do less. (e.g. no list parsing, but int is ok, we may think about accepting YAML)

• [X] Policy on positional arguments (currently experimental) - we have implemented them permissively with one undefined corner case.

  • [X] Fixed length - nope

  • [X] Variable length

  • [X] Can argparse be modified to always allow for them to appear at the beginning or end? - Probably not.

  • [x] Can we get argparse to allow a positional arg change the value of a prefixed arg and still have a sane help menu?

• [x] Policy on boolean flags - See the isflag argument of scriptconfig.Value

• [x] Improve over argparse’s default autogenerated help docs (needs exploration on what is possible with argparse and where extensions are feasible)