Object serialization/deserialization tools for python.

kodemore kodemore Last update: Mar 02, 2024

🌶️ Chili

PyPI version codecov CI Release License: MIT

Chili is an extensible library which provides a simple and efficient way to encode and decode complex Python objects to and from their dictionary representation.

It offers complete coverage for the typing package; including generics, and supports custom types, allowing you to extend the library to handle your specific needs.

With support for nested data structures, default values, forward references, and data mapping and transformation, Chili is designed to be both easy to use and powerful enough to handle complex data structures.

Note: Chili is not a validation library, although it ensures the type integrity.

Installation

To install the library, simply use pip:

pip install chili

or poetry:

poetry add chili

Usage

The library provides three main classes for encoding and decoding objects, chili.Encoder and chili.Decoder, and chili.Serializer, which combines both functionalities. Functional interface is also provided through chili.encode and chili.decode functions.

Additionally, library by default supports json serialization and deserialization, so you can use chili.JsonDecoder, and chili.JsonDecoder, and chili.JsonSerializer or its functional replacement chili.json_encode and chili.json_decode to serialize and deserialize objects to and from json.

Defining encodable/decodable properties

To define the properties of a class that should be encoded and decoded, you need to define them with type annotations. The @encodable, @decodable, or @serializable decorator should also be used to mark the class as encodable/decodable or serializable.

Note: Dataclasses are supported automatically, so you don't need to use the decorator.

from chili import encodable

@encodable
class Pet:
    name: str
    age: int
    breed: str

    def __init__(self, name: str, age: int, breed: str):
        self.name = name
        self.age = age
        self.breed = breed

Encoding

To encode an object, you need to create an instance of the chili.Encoder class, and then call the encode() method, passing the object to be encoded as an argument.

Note: The chili.Encoder class is a generic class, and you need to pass the type of the object to be encoded as a type argument.

from chili import Encoder

encoder = Encoder[Pet]()

my_pet = Pet("Max", 3, "Golden Retriever")
encoded = encoder.encode(my_pet)

assert encoded == {"name": "Max", "age": 3, "breed": "Golden Retriever"}

Alternatevely you can use chili.encode function:

from chili import encode

my_pet = Pet("Max", 3, "Golden Retriever")
encoded = encode(my_pet, Pet)

assert encoded == {"name": "Max", "age": 3, "breed": "Golden Retriever"}

chili.encode function by default only encodes @encodable objects, this behavior might be amended with the force flag.

Decoding

To decode an object, you need to create an instance of the chili.Decoder class, and then call the decode() method, passing the dictionary to be decoded as an argument.

Note: The chili.Decoder class is a generic class, and you need to pass the type of the object to be decoded as a type argument.

from chili import Decoder

decoder = Decoder[Pet]()

data = {"name": "Max", "age": 3, "breed": "Golden Retriever"}
decoded = decoder.decode(data)

assert isinstance(decoded, Pet)

Alternatevely you can use chili.decode function:

from chili import decode

data = {"name": "Max", "age": 3, "breed": "Golden Retriever"}
decoded = decode(data, Pet)

assert isinstance(decoded, Pet)

chili.decode function by default only decodes @decodable objects, this behavior might be amended with the force flag.

Missing Properties

If a property is not present in the dictionary when decoding, the chili.Decoder class will not fill in the property value, unless there is a default value defined in the type annotation. Similarly, if a property is not defined on the class, the chili.Encoder class will hide the property in the resulting dictionary.

Using Default Values

To provide default values for class properties that are not present in the encoded dictionary, you can define the properties with an equal sign and the default value. For example:

from typing import List
from chili import Decoder, decodable

@decodable
class Book:
    name: str
    author: str
    isbn: str = "1234567890"
    tags: List[str] = []

book_data = {"name": "The Hobbit", "author": "J.R.R. Tolkien"}
decoder = Decoder[Book]()

book = decoder.decode(book_data)

assert book.tags == []
assert book.isbn == "1234567890"

Note: When using default values with mutable objects, such as lists or dictionaries, be aware that the default value is shared among all instances of the class that do not have that property defined in the encoded dictionary. However, if the default value is empty (e.g. [] for a list, {} for a dictionary), it is not shared among instances.

Custom Type Encoders

You can also specify custom type encoders by defining a class that implements the chili.TypeEncoder protocol and passing it as a dictionary to the encoders argument of the Encoder constructor.

from chili import Encoder, TypeEncoder

class MyCustomEncoder(TypeEncoder):
    def encode(self, value: MyCustomType) -> str:
        return value.encode()

    
type_encoders = {MyCustomType: MyCustomEncoder()}
encoder = Encoder[Pet](encoders=type_encoders)

Custom Type Decoders

You can also specify custom type decoders by defining a class that implements the chili.TypeDecoder protocol and passing it as a dictionary to the decoders argument of the Decoder constructor.

from chili import Decoder, TypeDecoder

class MyCustomDecoder(TypeDecoder):
    def decode(self, value: str) -> MyCustomType:
        return MyCustomType.decode(value)

type_decoders = {MyCustomType: MyCustomDecoder()}
decoder = Decoder[Pet](decoders=type_decoders)

Convenient Functions

The library also provides convenient functions for encoding and decoding objects.

The chili.encode function takes an object and an optional type hint and returns a dictionary.

The chili.decode function takes a dictionary, a type hint, and returns an object.

from chili import encode, decode

my_pet = Pet("Max", 3, "Golden Retriever")

encoded = encode(my_pet)
decoded = decode(encoded, Pet)

To specify custom type encoders and decoders, you can pass them as keyword arguments to the chili.encode and chili.decode functions.

Serialization

If your object is both encodable and decodable, you can use the @serializable decorator to mark it as such. You can then use the chili.Serializer class to encode and decode objects.

from chili import Serializer, serializable

@serializable
class Pet:
    name: str
    age: int
    breed: str

    def __init__(self, name: str, age: int, breed: str):
        self.name = name
        self.age = age
        self.breed = breed

my_pet = Pet("Max", 3, "Golden Retriever")
serializer = Serializer[Pet]()

encoded = serializer.encode(my_pet)
decoded = serializer.decode(encoded)

Note: that you should only use the @serializable decorator for objects that are both @encodable and @decodable.

JSON Serialization

The library also provides classes for encoding and decoding objects to and from JSON formats. The chili.JsonEncoder and chili.JsonDecoder classes provide JSON serialization.

from chili import JsonEncoder, JsonDecoder, JsonSerializer

# JSON Serialization
encoder = JsonEncoder[Pet]()
decoder = JsonDecoder[Pet]()
serializer = JsonSerializer[Pet]()

my_pet = Pet("Max", 3, "Golden Retriever")
encoded = encoder.encode(my_pet)
decoded = decoder.decode(encoded)

The encoded value will be a json string:

{"name": "Max", "age": 3, "breed": "Golden Retriever"}

The decoded value will be an instance of a Pet object.

Functional interface is also available through the chili.json_encode, chili.json_decode functions.

Private properties

Chili recognizes private attributes within a class, enabling it to serialize these attributes when a class specifies a getter for an attribute and an associated private storage (must be denoted with a _ prefix).

Here is an example:

from chili import encodable, encode

@encodable
class Pet:
    name: str

    def __init__(self, name: str) -> None:
        self._name = name

    @property
    def name(self) -> str:
        return self._name

pet = Pet("Bobik")
data = encode(pet)
assert data == {
    "name": "Bobik",
}

Mapping

Mapping allows you to remap keys, apply functions to the values, and even change the structure of the input dictionary. This is particularly useful when you need to convert data from one format to another, such as when interacting with different APIs or data sources that use different naming conventions.

Simple mapping

Here's an example of how to use the chili.Mapper class from the library with a Pet class:

from chili import Mapper

# Create a Mapper instance with the specified scheme
mapper = Mapper({
    "pet_name": "name",
    "pet_age": "age",
    "pet_tags": {
        "tag_name": "tag",
        "tag_type": "type",
    },
})

data = {
    "pet_name": "Max",
    "pet_age": 3,
    "pet_tags": [
        {"tag_name": "cute", "tag_type": "description"},
        {"tag_name": "furry", "tag_type": "description"},
    ],
}

# Apply the mapping to your input data
mapped_data = mapper.map(data)

print(mapped_data)

The mapped_data output would be:

{
    "name": "Max",
    "age": 3,
    "pet_tags": [
        {"tag": "cute", "type": "description"},
        {"tag": "furry", "type": "description"},
    ],
}

Using KeyScheme

KeyScheme can be used to define mapping rules for nested structures more explicitly. It allows you to specify both the old key and the nested mapping scheme in a single, concise object. This can be particularly useful when you want to map a nested structure but need to maintain clarity in your mapping scheme.

Here's an example of how to use chili.KeyScheme with the chili.Mapper class:

from chili import Mapper, KeyScheme

# Create a Mapper instance with the specified scheme
mapper = Mapper({
    "pet_name": "name",
    "pet_age": "age",
    "pet_tags": KeyScheme("tags", {
        "tag_name": "tag",
        "tag_type": "type",
    }),
})

pet_dict = {
    "pet_name": "Max",
    "pet_age": 3,
    "pet_tags": [
        {"tag_name": "cute", "tag_type": "description"},
        {"tag_name": "furry", "tag_type": "description"},
    ],
}

# Apply the mapping to your input data
mapped_data = mapper.map(pet_dict)

print(mapped_data)

The mapped_data output would be:

{
    "name": "Max",
    "age": 3,
    "tags": [
        {"tag": "cute", "type": "description"},
        {"tag": "furry", "type": "description"},
    ],
}

Using wildcards in mapping

The chili.Mapper supports using ... (Ellipsis) as a wildcard for keys that you want to include in the mapping but do not want to explicitly define. This can be useful when you want to map all keys in the input data, or when you want to map specific keys and leave the remaining keys unchanged.

You can use a lambda function with the ... wildcard to apply a transformation to the keys or values that match the wildcard.

Here's an example of how to use the ... wildcard with the chili.Mapper class:

from chili import Mapper

# Create a Mapper instance with the specified scheme containing a wildcard ...
mapper = Mapper({
    "pet_name": "name",
    "pet_age": "age",
    ...: lambda k, v: (f"extra_{k}", v.upper() if isinstance(v, str) else v),
})

pet_dict = {
    "pet_name": "Max",
    "pet_age": 3,
    "pet_color": "white",
    "pet_breed": "Golden Retriever",
    "pet_tags": [
        {"tag": "cute", "type": "description"},
        {"tag": "furry", "type": "description"},
    ],
}

# Apply the mapping to your input data
mapped_data = mapper.map(pet_dict)

print(mapped_data)

The mapped_data output would be:

{
    "pet_name": "Fluffy",
    "pet_age": 3,
    "extra_color": "WHITE",
    "extra_breed": "POODLE",
    "extra_tags": [
        {
            "tag": "cute",
            "type": "description",
        },
        {
            "tag": "furry",
            "type": "description",
        },
    ],
}

Using mapping in decodable/encodable objects

You can also use mapping by setting mapper parameter in @chili.encodable and @chili.decodable decorators.

from typing import List

from chili import encodable, Mapper, encode

mapper = Mapper({
    "pet_name": "name",
    "pet_age": "age",
    "pet_tags": {
        "tag_name": "tag",
        "tag_type": "type",
    },
})


@encodable(mapper=mapper)
class Pet:
    name: str
    age: int
    tags: List[str]

    def __init__(self, name: str, age: int, tags: List[str]):
        self.name = name
        self.age = age
        self.tags = tags


pet = Pet("Max", 3, ["cute", "furry"])
encoded = encode(pet)

assert encoded == {
    "pet_name": "Max",
    "pet_age": 3,
    "pet_tags": [
        {"tag_name": "cute", "tag_type": "description"},
        {"tag_name": "furry", "tag_type": "description"},
    ],
}

Alternatively you can set mapper in Encoder and Decoder classes:

encoder = Encoder[Pet](mapper=mapper)

pet = Pet("Max", 3, ["cute", "furry"])
encoded = encoder.encode(pet)

Error handling

The library raises errors if an invalid type is passed to the Encoder or Decoder, or if an invalid dictionary is passed to the Decoder.

from chili import Encoder, Decoder
from chili.error import EncoderError, DecoderError

# Invalid Type
encoder = Encoder[MyInvalidType]()  # Raises EncoderError.invalid_type

decoder = Decoder[MyInvalidType]()  # Raises DecoderError.invalid_type

# Invalid Dictionary
decoder = Decoder[Pet]()
invalid_data = {"name": "Max", "age": "three", "breed": "Golden Retriever"}
decoded = decoder.decode(invalid_data)  # Raises DecoderError.invalid_input

Performance

The table below shows the results of the benchmarks for encoding and decoding objects with the library, Pydantic, and attrs.

Command Mean [ms] Min [ms] Max [ms] Relative
poetry run python benchmarks/chili_decode.py 249.4 ± 4.1 245.5 258.8 1.01 ± 0.02
poetry run python benchmarks/pydantic_decode.py 295.5 ± 12.5 287.6 327.1 1.19 ± 0.05
poetry run python benchmarks/attrs_decode.py 260.9 ± 8.6 253.2 283.5 1.05 ± 0.04
poetry run python benchmarks/chili_encode.py 247.8 ± 2.3 245.4 253.0 1.00
poetry run python benchmarks/pydantic_encode.py 292.4 ± 4.7 287.1 302.5 1.18 ± 0.02
poetry run python benchmarks/attrs_encode.py 258.2 ± 2.1 254.4 261.4 1.04 ± 0.01

Supported types

The following section lists all the data types supported by the library and explains how they are decoded and encoded. The supported data types include built-in Python types like bool, dict, float, int, list, set, str, and tuple, as well as more complex types like collections.namedtuple, collections.deque, collections.OrderedDict, datetime.date, datetime.datetime, datetime.time, datetime.timedelta, decimal.Decimal, enum.Enum, enum.IntEnum, pathlib.Path, and various types defined in the typing module.

Simple types

Simple type are handled by a ProxyEncoder and ProxyDecoder. These types are decoded and encoded by casting the value to the specified type.

For more details please refer to chili.encoder.ProxyEncoder and chili.decoder.ProxyDecoder.

bool

Passed value is automatically cast to a boolean with python's built-in bool type during decoding and encoding process.

int

Passed value is automatically cast to an int with python's built-in int type during decoding and encoding process.

float

Passed value is automatically cast to float with python's built-in float type during decoding and encoding process.

str

Passed value is automatically cast to string with python's built-in str during encoding and decoding process.

set

Passed value is automatically cast to either set during decoding process or list during encoding process.

frozenset

Passed value is automatically cast to either frozenset during decoding process or list during encoding process.

list

Passed value is automatically cast to list with python's built-in list during encoding and decoding process.

tuple

Passed value is automatically cast either to tuple during decoding process or to list during encoding process.

dict

Passed value is automatically cast to dict with python's built-in dict during encoding and decoding process.

Complex types

Complex types are handled by corresponding Encoder and Decoder classes.

collections.namedtuple

Passed value is automatically cast to either collections.namedtuple during decoding process or list during encoding process.

collections.deque

Passed value is automatically cast to either collections.deque during decoding process or list during encoding process.

collections.OrderedDict

Passed value is automatically cast to either collections.OrderedDict during decoding process or list where each item is a list of two elements corresponding to key and value, during encoding process.

datetime.date

Passed value is automatically cast to either datetime.date during decoding process or str (valid ISO-8601 date string) during encoding process.

datetime.datetime

Passed value must be valid ISO-8601 date time string, then it is automatically hydrated to an instance of datetime.datetime class and extracted to ISO-8601 format compatible string.

datetime.time

Passed value must be valid ISO-8601 time string, then it is automatically hydrated to an instance of datetime.time class and extracted to ISO-8601 format compatible string.

datetime.timedelta

Passed value must be valid ISO-8601 duration string, then it is automatically hydrated to an instance of datetime.timedelta class and extracted to ISO-8601 format compatible string.

decimal.Decimal

Passed value must be a string containing valid decimal number representation, for more please read python's manual about decimal.Decimal, on extraction value is extracted back to string.

enum.Enum

Supports hydration of all instances of enum.Enum subclasses as long as value can be assigned to one of the members defined in the specified enum.Enum subclass. During extraction the value is extracted to value of the enum member.

enum.IntEnum

Same as enum.Enum.

pathlib.Path

Supported hydration for all instances of pathlib.Path class, during extraction value is extracted to string.

Typing module support

typing.Any

Passed value is unchanged during hydration and extraction process.

typing.AnyStr

Same as str

typing.Deque

Same as collection.dequeue with one exception, if subtype is defined, eg typing.Deque[int] each item inside queue is hydrated accordingly to subtype.

typing.Dict

Same as dict with exception that keys and values are respectively hydrated and extracted to match annotated type.

typing.FrozenSet

Same as frozenset with exception that values of a frozen set are respectively hydrated and extracted to match annotated type.

typing.List

Same as list with exception that values of a list are respectively hydrated and extracted to match annotated type.

typing.NamedTuple

Same as namedtuple.

typing.Set

Same as set with exception that values of a set are respectively hydrated and extracted to match annotated type.

typing.Tuple

Same as tuple with exception that values of a set are respectively hydrated and extracted to match annotated types. Ellipsis operator (...) is also supported.

typing.TypedDict

Same as dict but values of a dict are respectively hydrated and extracted to match annotated types.

typing.Generic

Only parametrised generic classes are supported, dataclasses that extends other Generic classes without parametrisation will fail.

typing.Optional

Optional types can carry additional None value which chili's hydration process will respect, so for example if your type is typing.Optional[int] None value is not hydrated to int.

typing.Union

Limited support for Unions.

typing.Pattern

Passed value must be a valid regex pattern, if contains flags regex should start with / and flags should be passed after / only ismx flags are supported.

Subscribe to our newsletter