Primitize

Primitize is a library that facilitates converting dataclass instances into primitive objects. It provides facilites to massage the data, validate it and write it out to file in pretty much any format you would want.

Interface

from dataclasses import dataclass

from primitize.core import Primitizable, primitize, primitized


@dataclass
class Obj(Primitizable):
    a: int
    b: bool = primitized(modifier=lambda self, b: str(b).lower())

    def primitize(self):
        return json.dump(super().primitize(self))


primitize(Obj(1, True)) == '{"a": 1, "b": "true"}'

Primitize will convert any data class down to primitive types. When calling primitize(d) on any d dataclass, it will execute the d.primitize() method if it is defined, otherwise it will use a default implementation. The default implementation will loop over every field defined in the dataclass. Each field is serialized in a 2 step process:

1. First call the modifier(object, field_value).
2. The return of this is then passed into validator(object, return_of_the_modifier).

Example usage

Generating configuration files

Imagine we want to generate configuration files for several clusters, we have good sensible defaults but nothing is always exactly the same. In this example, we want each cluster configuration to be written in a json file.

from dataclasses import dataclass
from enum import Enum
import json

from primitize.core import primitize, primitized


class HostType(Enum):
    COMPUTE = "Compute"
    WEB = "Web"
    STORAGE = "Storage"


@dataclass
class User:
    username: str


@dataclass
class Cluster(Primizable):
    name: str
    size: int = primitized(validator=lambda v, o: v > 0)
    host_type: HostType = primitized(
        default=HostType.WEB, modifier=lambda v, o: v.value
    )
    admins: Set[User] = primitized(
        default_factory=set,
        modifier=lambda v, o: sorted(x),
        validator: lambda v, o: len(x) > 0,
    )

    def primitize(self):
        prim = super().primitize(self)
        payload = json.dumps(prim, sort_keys=True, indent=4)
        with (Path(".") / "output" / f"{self.name}.json").open("w") as fd:
            fd.write(payload)


clusters ={
    Cluster("A", 3, HostType.COMPUTE, {User("root")}),
    Cluster("B", 3, admins={User("root")}),
    Cluster("C", 3, HostType.STORAGE, {User("foo")}),
}

for cluster in clusters:
    primitize(cluster)

Upon executing this, you will find the following files under ./output/:

./output/A.json:

{
    "name": "A",
    "size": 3,
    "host_type": "Compute",
    "admins": [
        "root"
    ]
}

./output/B.json:

{
    "name": "A",
    "size": 3,
    "host_type": "Web",
    "admins": [
        "root"
    ]
}

./output/C.json:

{
    "name": "C",
    "size": 3,
    "host_type": "Storage",
    "admins": [
        "foo"
    ]
}

How is this different from X ?

	Primitize	Protobuf/Thrift	TypedDict	Jinja
Modifiers	Yes	No	No	No
Validators	Type & Custom	Type	Type	No
Language support	Python only	Multiple	Python	Custom DSL
Full Python API	Yes	No	Yes	No
Format flexibility	Yes	No	No	Yes

• Modifiers: Primitize allows you to define functions that will massage the values prior to serialization, this allows your to rename, or reformat data to make the serialization easier. A typical usecase for this is to have the values of a type that is easy to manipulate and use a modifier to rewrite it to the what your end format expects
• Validators: Primitize uses standard Python typing so mypy will ensure you have type checking. In Primitize you can also define functions that allows you to ensure that the value is correct upon serialization.
• Language support: Primitize is a python library, as such it is primarily useful in Python.
• Full Python API: Unlike a restrictive DSL, with Primitize it is really just your Python program running, so you can do whatever you want, we just help you convert dataclasses into primitive types so it is easier to write out.
• Format flexibility: Primitize doesn't write data out for you so you can write it out as anything you want. Do you want json ? json.dumps(primitize(data)) Do you want yaml ? yaml.dumps(primitize(data)). It's that simple.