Simple and flexible conversions between dataclasses and jsonable dictionaries.
It maps dataclasses to jsonable dictionaries but not json strings.
- Easy to use.
- Supports common type annotations.
- Supports recursive conversions.
- Supports field-level and dataclass-level overriding.
Requirements: Python >= 3.7
Install via pip
:
pip install dataclass-jsonable
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()
.
-
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 tostr
.@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
andIntEnum
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 itstype
.@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, butUnion[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 tolist
.@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 todict
.@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
(orJ
) 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.
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 adefault
ordefault_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 nodefault
ordefault_factory
declared, and has nodefault_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
anddecoder
better alias names since 0.1.1:to_json
andfrom_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"}
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}
BSD.