Skip to content
/ marshy Public

An object marshaling system for python aiming for convention over configuration while still allowing for extreme flexibility

License

Notifications You must be signed in to change notification settings

tofarr/marshy

Repository files navigation

Marshy - Better Marshalling for Python.

This project is a general purpose externalizer for python objects. (Like Marshmallow or Pedantic) The guiding philosophy is convention over configuration, with the aim of still making customizations as pain free as possible, based on python type hints.

Out of the box, it supports primitives, dataclasses, and enums.

Installation

pip install marshy

General Usage

Given the following dataclass:

from typing import List, Optional
import dataclasses


@dataclasses.dataclass
class Doohickey:
  title: str
  description: Optional[str] = None
  tags: List[str] = dataclasses.field(default_factory=list)

Marshall data with:

import marshy
result = marshy.dump(Doohickey('Thingy', tags=['a','b']))
# result == dict(title='Thingy', description=None, tags=['a','b'])

Unmarshall data with:

result = marshy.load(Doohickey, dict(title='Thingy'))
# result == Doohickey('Thingy', description=None, tags=[])

Custom properties

Custom properties are also serialized by default. (If they have a setter, it is used when loading):

import marshy
from functools import reduce
import dataclasses

@dataclasses.dataclass
class Factorial:
  value: int
  
  @property
  def factorial(self) -> int:
    return reduce(lambda a, b: a*b, range(1, self.value+1))
    
factorial = Factorial(4)
dumped = marshy.dump(factorial)
# dumped == dict(value=4, factorial=24)
loaded = marshy.load(Factorial, dumped)
# loaded == factorial

Under The Hood

Internally, API defines 3 core concepts:

  • A Marshaller: Is responsible for marshalling / unmarshalling a single type of object.
  • A MarshallerFactory: has a create method used to create marshallers for types, and has a priority which controls the order in which they are run. (higher first)
  • A MarshyContext: coordinates the activities of Marshallers and Factories

Creating a Custom Marshaller Context

If you need multiple independent sets of rules for marshalling data, then you should create your own marshalling contexts and store references to them. The default works well otherwise:

# Create a new blank marshy context - this will fail
# because there are no preset types or factories.
from marshy.marshy_context import MarshyContext
my_marshaller_context = MarshyContext()
dumped = my_marshaller_context.dump(Doohickey('Thingy'))

# Create a new marshaller context which copies the default rules.
from marshy.marshy_context import create_marshy_context
my_default_context = create_marshy_context()
dumped = my_default_context.dump(Doohickey('Thingy'))

Creating a Custom Marshaller

To customize marshalling for a type, write a marshaller and then register it with your context:

from marshy.marshaller.marshaller_abc import MarshallerABC
from marshy.types import ExternalType


class MyDoohickeyMarshaller(MarshallerABC[Doohickey]):
    marshalled_type = Doohickey

    def load(self, item: ExternalType) -> Doohickey:
        return Doohickey(item[0], item[1], item[2])

    def dump(self, item: Doohickey) -> ExternalType:
        return [item.title, item.description, item.tags]

my_default_context.register_marshaller(MyDoohickeyMarshaller())
dumped = my_default_context.dump(Doohickey('Thingy'))
# dumped == ['Thingy', None, []]

loaded = my_default_context.load(Doohickey, dumped)
# dumped == Doohickey('Thingy')

Creating a Custom Marshaller Factory

Sometimes you need to create a marshaller for a while concept of object rather than a single type - In this case you need a factory, (this is how the default rules work!). Examples:

  • ListMarshallerFactory looks for typed lists (e.g.,: List[str]) and creates marshallers for them - you already saw the results in Doohickey.tags above.
  • OptionalMarshallerFactory looks for optional fields (e.g.,: Optional[str]) and creates marshallers that mean each individual other marshaller does not need to accommodate the case where a value is None - just mark it optional!
  • DataClassMarshallerFactory provides a marshaller for dataclasses assuming they have a standard constructor based on their fields.

Customizing dataclass attributes:

Taking the doohickey example:

from marshy import dump, get_default_marshy_context
from marshy.marshaller.bool_marshaller import BoolMarshaller
from marshy.marshaller.primitive_marshaller import StrMarshaller
from marshy.marshaller.obj_marshaller import attr_config, ObjMarshaller
attr_marshallers = [
    attr_config(StrMarshaller(), "title"),
    attr_config(BoolMarshaller(), "tags"),
]
get_default_marshy_context().register_marshaller(ObjMarshaller(Doohickey, attr_marshallers))
dumped = dump(Doohickey('Thingy'))
# dumped == dict(title='Thingy', tags=False)

Customizing dataclass marshalling

As an alternative to defining a custom marshaller / factory, it is possible to simply define a marshaller_factory class method. (Note: this becomes the default for all contexts) Imagine a case where you have a dataclass representing a 2D point, which you want to be marshalled in the format [x, y] (An array rather than the standard object):

from dataclasses import dataclass
from marshy.marshaller.marshaller_abc import MarshallerABC
from marshy import load, dump

@dataclass
class Point:
    x: float
    y: float
    
    @classmethod
    def __marshaller_factory__(cls, marshaller_context):
       return PointMarshaller()
       
class PointMarshaller(MarshallerABC[Point]):
    marshalled_type = Point

    def load(self, item):
        return Point(item[0], item[1])
    
    def dump(self, item):
        return [item.x, item.y]
        
dumped = dump(Point(1.2, 3.4))
loaded = load(Point, dumped)

Circular References

Due to the fact that types in the object graph can self reference, we defer resolution of most marshaller until as late as possible. DeferredMarshaller is responsible for this, and means types can self reference.

Circular references within objects will still cause an error. (Unless you decide on an error handling protocol for this an implement a custom Factory to deal with it!)

Customizing the default context

We use Injecty to manage dependency injection. Since any top level module injecty_config_* is automatically loaded, our injecty_config_marshy.py defines our default Marshaller / Factory implementations.

Adding Polymorphic Implementations

We also use injecty to find polymorphic implementations of types to marshall / unmarshall. Taking the following classes where Pet has implementations Cat and Dog:

from abc import ABC, abstractmethod
from dataclasses import dataclass

@dataclass
class PetAbc(ABC):
    name: str

    @abstractmethod
    def vocalize(self) -> str:
        """ What sound does this make? """


class Cat(PetAbc):

    def vocalize(self):
        return "Meow!"


class Dog(PetAbc):

    def vocalize(self) -> str:
        return "Woof!"

In order to deserialize a Pet, marshy needs to be informed tha the implementations exist. This can be done at any point in the configuration:

from injecty import get_default_injecty_context
import marshy
get_default_injecty_context().register_impls(PetAbc, [Cat, Dog])
pet = ['Cat', dict(name='Felix')]
loaded = marshy.load(PetAbc, pet)

Tests for this are here

Performance Tests

Basic Tests show performance is approximate with marshmallow:

python -m timeit -s "
from test.performance.marshy_performance import run
run(1000)
"
python -m timeit -s "
from test.performance.marshmallow_performance import run
run(1000)
"

Release Procedure

status

The typical process here is:

  • Create a PR with changes. Merge these to main (The Quality workflows make sure that your PR meets the styling, linting, and code coverage standards).
  • New releases created in gitHub are automatically uploaded to pypi

About

An object marshaling system for python aiming for convention over configuration while still allowing for extreme flexibility

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages