dataclass-jsonable

中文说明

Simple and flexible conversions between dataclasses and jsonable dictionaries.

It maps dataclasses to jsonable dictionaries but not json strings.

Features

• Easy to use.
• Supports common type annotations.
• Supports recursive conversions.
• Supports field-level and dataclass-level overriding.

Installation

Requirements: Python >= 3.7

Install via pip:

pip install dataclass-jsonable

Quick Example

from dataclasses import dataclass
from datetime import datetime
from decimal import Decimal
from enum import IntEnum
from typing import List
from dataclass_jsonable import J

class Color(IntEnum):
    BLACK = 0
    BLUE = 1
    RED = 2

@dataclass
class Pen(J):
    color: Color
    price: Decimal
    produced_at: datetime

@dataclass
class Box(J):
    pens: List[Pen]

box = Box(pens=[Pen(color=Color.BLUE, price=Decimal("20.1"), produced_at=datetime.now())])

# Encode to a jsonable dictionary.
d = box.json()
print(d)  # {'pens': [{'color': 1, 'price': '20.1', 'produced_at': 1660023062}]}

# Construct dataclass from a jsonable dictionary.
print(Box.from_json(d))

APIs are only the two: .json() and .from_json().

Built-in Supported Types

• bool, int, float, str, None encoded as it is.

  @dataclass
  class Obj(J):
      a: int
      b: str
      c: bool
      d: None
  
  Obj(a=1, b="b", c=True, d=None).json()
  # => {'a': 1, 'b': 'b', 'c': True, 'd': None}

• Decimal encoded to str.

  @dataclass
  class Obj(J):
      a: Decimal
  
  Obj(a=Decimal("3.1")).json()  # => {'a': '3.1'}

• datetime encoded to timestamp integer via .timestamp() method. timedelta encoded to integer via .total_seconds() method.

  @dataclass
  class Obj(J):
      a: datetime
      b: timedelta
  
  Obj(a=datetime.now(), b=timedelta(minutes=1)).json()
  # => {'a': 1660062019, 'b': 60}

• Enum and IntEnum encoded to their values via .value attribute.

  @dataclass
  class Obj(J):
      status: Status
  
  Obj(status=Status.DONE).json()  # => {'status': 1}

• Any is encoded according to its type.

  @dataclass
  class Obj(J):
      a: Any
  
  Obj(1).json()  # {'a': 1}
  Obj("a").json()  # {'a': 'a'}
  Obj.from_json({"a": 1})  # Obj(a=1)

• Optional[X] is supported, but Union[X, Y, ...] is not.

  @dataclass
  class Obj(J):
      a: Optional[int] = None
  
  Obj(a=1).json()  # => {'a': 1}

• List[X], Tuple[X], Set[X] are all encoded to list.

  @dataclass
  class Obj(J):
      a: List[int]
      b: Set[int]
      c: Tuple[int, str]
      d: Tuple[int, ...]
  
  Obj(a=[1], b={2, 3}, c=(4, "5"), d=(7, 8, 9)).json())
  # => {'a': [1], 'b': [2, 3], 'c': [4, '5'], 'd': [7, 8, 9]}
  
  Obj.from_json({"a": [1], "b": [2, 3], "c": [4, "5"], "d": [7, 8, 9]}))
  # => Obj(a=[1], b={2, 3}, c=(4, '5'), d=(7, 8, 9))

• Dict[str, X] encoded to dict.

  @dataclass
  class Obj(J):
      a: Dict[str, int]
  Obj(a={"x": 1}).json()  # => {'a': {'x': 1}}
  Obj.from_json({"a": {"x": 1}}) # => Obj(a={'x': 1})

• Nested or recursively JSONAble (or J) dataclasses.

  @dataclass
  class Elem(J):
      k: str
  
  @dataclass
  class Obj(J):
      a: List[Elem]
  
  Obj([Elem("v")]).json()  # => {'a': [{'k': 'v'}]}
  Obj.from_json({"a": [{"k": "v"}]})  # Obj(a=[Elem(k='v')])

• Postponed annotations (the ForwardRef in PEP 563).

  @dataclass
  class Node(J):
      name: str
      left: Optional["Node"] = None
      right: Optional["Node"] = None
  
  root = Node("root", left=Node("left"), right=Node("right"))
  root.json()
  # {'name': 'root', 'left': {'name': 'left', 'left': None, 'right': None}, 'right': {'name': 'right', 'left': None, 'right': None}}

If these built-in default conversion behaviors do not meet your needs, or your type is not on the list, you can use json_options introduced below to customize it.

Customization / Overriding Examples

We can override the default conversion behaviors with json_options, which uses the dataclass field's metadata for field-level customization purpose, and the namespace is j.

The following pseudo code gives the pattern:

from dataclasses import field
from dataclass_jsonable import json_options

@dataclass
class Struct(J):
    attr: T = field(metadata={"j": json_options(**kwds)})

An example list about json_options:

• Specific a custom dictionary key over the default field's name:

  @dataclass
  class Person(J):
      attr: str = field(metadata={"j": json_options(name="new_attr")})
  Person(attr="value").json() # => {"new_attr": "value"}

  And more, we can use a function to specific a custom dictionary key. This may be convenient to work with class-level __default_json_options__ attribute (check it below).

  @dataclass
  class Obj(J):
      simple_value: int = field(metadata={"j": json_options(name_converter=to_camel_case)})
  Obj(simple_value=1).json()  # => {"simpleValue": 1}

  And we may specific a custom field name converter when converts dictionary to dataclass:

  @dataclass
  def Person(J):
    name: str = field(
          metadata={
              "j": json_options(
                  name_converter=lambda x: x.capitalize(),
                  name_inverter=lambda x: "nickname",
            )
        }
    )

  As the Person defined above, it will convert to dictionary like {"Name": "Jack"} and can be loaded from {"nickname": "Jack"}.

• Omit a field if its value is empty:

  @dataclass
  class Book(J):
      name: str = field(metadata={"j": json_options(omitempty=True)})
  Book(name="").json() # => {}

  Further, we can specify what is 'empty' via option omitempty_tester:

  @dataclass
  class Book(J):
      attr: Optional[str] = field(
          default=None,
          metadata={
              # By default, we test `empty` using `not x`.
              "j": json_options(omitempty=True, omitempty_tester=lambda x: x is None)
          },
      )
  
  Book(attr="").json()  # => {'attr': ''}
  Book(attr=None).json()  # => {}

• Always skip a field. So we can stop some "private" fields from exporting:

  @dataclass
  class Obj(J):
      attr: str = field(metadata={"j": json_options(skip=True)})
  
  Obj(attr="private").json() # => {}

• Always keep a field without encoding nor decoding, this prevents the default encoding/decoding behavior:

  @dataclass
  class Obj(J):
      timestamp: datetime = field(metadata={"j": json_options(keep=True)})
  
  Obj(timestamp=datetime.now()).json() # =>  {'timestamp': datetime.datetime(2023, 9, 5, 14, 54, 24, 679103)}

• dataclasses's field allows us to pass a default or default_factory argument to set a default value:

  @dataclass
  class Obj(J):
      attr: List[str] = field(default_factory=list, metadata={"j": json_options(**kwds)})

  There's also an option default_before_decoding in dataclass-jsonable, which specifics a default value before decoding if the key is missing in the dictionary. Sometimes this way is more concise:

  @dataclass
  class Obj(J):
      updated_at: datetime = field(metadata={"j": json_options(default_before_decoding=0)})
  
  Obj.from_json({})  # => Obj(updated_at=datetime.datetime(1970, 1, 1, 8, 0))

  dataclass-jsonable also introduces a class-level similar option __default_factory__. If a field has no default or default_factory declared, and has no default_before_decoding option used, this function will generate a default value according to its type, to prevent a "missing positional arguments" TypeError from raising.

  from dataclass_jsonable import J, zero
  
  @dataclass
  class Obj(J):
      __default_factory__ = zero
  
      n: int
      s: str
      k: List[str]
  
  Obj.from_json({})  # => Obj(n=0, s='', k=[])

• Override the default encoders and decoders.

  This way, you have complete control over how to encode and decode at field level.

  @dataclass
  class Obj(J):
      elems: List[str] = field(
          metadata={
              "j": json_options(
                  encoder=lambda x: ",".join(x),
                  decoder=lambda x: x.split(","),
              )
          }
      )
  
  Obj(elems=["a", "b", "c"]).json()  # => {'elems': 'a,b,c'}
  Obj.from_json({"elems": "a,b,c"})  # => Obj(elems=['a', 'b', 'c'])

  The following code snippet about datetime is a very common example, you might want ISO format datetime conversion over timestamp integers.

  @dataclass
  class Record(J):
      created_at: datetime = field(
          default_factory=datetime.now,
          metadata={
              "j": json_options(
                  encoder=datetime.isoformat,
                  decoder=datetime.fromisoformat,
              )
          },
      )
  
  Record().json()  # => {'created_at': '2022-08-09T23:23:02.543007'}

  dataclass-jsonable gives encoder and decoder better alias names since 0.1.1: to_json and from_json.

  @dataclass
  class Obj(J):
      elems: List[str] = field(
          metadata={
              "j": json_options(
                  to_json=lambda x: ",".join(x),  # Alias for encoder
                  from_json=lambda x: x.split(","),  # Alias for decoder
              )
          }
      )
  
  Obj(elems=["a", "b", "c"]).json()  # => {'elems': 'a,b,c'}
  Obj.from_json({"elems": "a,b,c"})  # => Obj(elems=['a', 'b', 'c'])

• For some very narrow scenarios, we may need to execute a hook function before decoding, for example, the data to be decoded is a serialized json string, and but we still want to use the built-in decoder functions instead of making a new decoder.

  import json
  
  @dataclass
  class Obj(J):
      data: Dict[str, Any] = field(metadata={"j": json_options(before_decoder=json.loads)})
  
  Obj.from_json({"data": '{"k": "v"}'})
  # => Obj(data={'k': 'v'})

• Customize default behaviors at the class level.

  If an option is not explicitly set at the field level, the __default_json_options__ provided at the class level will be attempted.

  @dataclass
  class Obj(J):
      __default_json_options__ = json_options(omitempty=True)
  
      a: Optional[int] = None
      b: Optional[str] = None
  
  Obj(b="b").json() # => {'b': 'b'}

  @dataclass
  class Obj(J):
      __default_json_options__ = json_options(name_converter=to_camel_case)
  
      status_code: int
      simple_value: str
  
  Obj2(status_code=1, simple_value="simple").json()
  # => {"statusCode": 1, "simpleValue": "simple"}

Debuging

It provides a method obj._get_origin_json(), it returns the original json dictionary which constructs instance obj via from_json().

d = {"a": 1}
obj = Obj.from_json(d)
obj._get_origin_json()
# => {"a": 1}

License

BSD.