Marsh

(Un)marshaling library for objects in Python 3.8+.

Tested on Ubuntu, Windows and Mac OSX for Python 3.8, 3.9 and 3.10.

Relies heavily on type-hint reflection. All (un)marshaling is performed recursively which allows for support of nested types.

Influenced by the work of Omry Yadan on hydra.

The documentation is hosted on ReadTheDocs.

Getting Started

Install

marsh is available through pip

pip install marsh

Entry point

Using the unmarshaling capabilities of marsh we can create entry points for python code.

These entry points allow for arguments to be given on the command line and/or through config files which are then validated, converted to the correct types and passed to the entry point function.

Most python types are supported (primitives, type aliases, dataclasses, namedtuple e.t.c.)

Create

Creating an entry point is as simple as decorating a function and calling it without arguments.

# app.py
import marsh
from typing import Sequence, Union


@marsh.main
def run(
    a: int,
    b: Union[float, Sequence[int]],
    c: dict[str, bool],
) -> None:
    """Example of an entry point.

    Arguments:
        a: An integer argument.
        b: A floating point value or a sequence of ints.
        c: A dictionary with string keys and
            bool values. If this was python 3.8
            we would instead use typing.Dict[str, bool] as
            type hint as the builtin dict did not support
            type annotations.
    """
    print(a, type(a))
    print(b, type(b))
    print(c, type(c))


if __name__ == '__main__':
    run()

Run

When running the application we can use the positional arguments on the command line to pass values to our function.

$ python app.py a=1 b=5e-1 c.key1=true c.key2=false
1 <class 'int'>
0.5 <class 'float'>
{'key1': True, 'key2': False} <class 'dict'>

Argument validation

When giving invalid values or when required arguments are missing an error message is printed and the application exits.

$ python app.py a=1.5 b=0 c.some_key=true
failed to unmarshal config: int: could not convert: 1.5
	path: a

$ python app.py b=0 c.some_key=true
failed to unmarshal config: MissingValueError
	path: a

Help

Using --help we can also get a help message for the arguments. Here the output was piped to tail to truncate the output into displaying only the arguments of our entry point.

$ python app.py --help | tail -n 11
fields:
  a: <int>              An integer argument.

  b: <float> | [<int>, ...]
                        A floating point value or a sequence of ints.

  c: {<str>: <bool>, ...}
                        A dictionary with string keys and bool values. If this
                        was python 3.8 we would instead use typing.Dict[str,
                        bool] as type hint as the builtin dict did not support
                        type annotations.

Marshal

Marshaling values simply means taking a python object and turning it into JSON-like data.

# marshal.py
import dataclasses
import marsh


@dataclasses.dataclass
class Config:
    a: int
    b: float


config = Config(1, 5e-1)
print(marsh.marshal(config))

$ python marshal.py
{'a': 1, 'b': 0.5}

Unmarshal

Unmarshaling is the opposite of marshaling. A type is instantiated using JSON-like data.

# unmarshal.py
import dataclasses
import typing
import marsh


class Range(typing.NamedTuple):
    start: typing.Optional[int]
    stop: int


@dataclasses.dataclass
class Config:
    a: int
    b: float
    c: Range


config = marsh.unmarshal(
    Config,
    {
        'a': 1,
        'b': 1.5,
        'c': {
            'start': None,
            'stop': 5,
        },
    }
)
print(config)

$ python umarshal.py
Config(a=1, b=1.5, c=Range(start=None, stop=5))

License

MIT License.