stereotype

Stereotype is a performance-focused Python 3.8 library for providing a structure for your data and validating it. The models allow fast & easy conversion between primitive data and well-typed Python classes.

Stereotype is heavily influenced by the beauty of dataclasses and versatility of Schematics, while having much better performance - both in terms of CPU usage and memory footprint. While it wasn't an influence, it is somewhat similar to Pydantic, but also beats it in benchmarks and provides easier validation.

Stereotype supports Python 3.8 and above (future support for older versions of Python is highly unlikely) and has 100% test coverage.

Features

Support for atomic fields - bool, int, float, str
Optional for any supported type
Fields with Lists of any supported type and Dicts of atomic types to any supported type
Support for Model subclass fields, including recursive definitions
Dynamic model fields resolved using a string type key
Free-form fields using Any
Calculated serializable fields - a property present also in serialized data
Optional field defaults using atomic values or callables
Renaming or disabling fields for purposes of input/output/both
Optional hiding of None values from output
Some built-in validation helpers for most fields
Custom validation instance methods with context
Serialization roles using field blacklists or whitelists, with inheritance or overriding

Tutorial

Atomic value fields

Here is a simple flat model defined using stereotype:

from typing import Optional

from stereotype import Model, FloatField, ValidationError, Missing


class Employee(Model):
    name: str
    department: str = 'Engineering'
    female: Optional[bool] = None
    salary: float = FloatField(min_value=42., default=42.)

    human = True

    def greeting(self) -> str:
        title = {True: "Ms. ", False: "Mr. "}.get(self.female, "")
        return f'Dear {title}{self.name}'

All models need to inherit from the Model class (directly or indirectly).
All class attributes that have a type annotation (i.e. the attribute name is followed with a colon) will be converted to model fields, i.e. they can be deserialized from input, serialized to output or be validated. See [Fields] for information about what types are supported.
Fields the type of which is Optional (a.k.a. Union[None, other]) will be valid even if None is supplied.
The assigned value for a field will be used as the default if the key was not present in the input. Note None in the input will not be converted to the default. The default must be a value of the field's allowed type.
If a callable is provided as a default, it will be called with no arguments to create a new default.
If the assigned value is a stereotype ***Field, it can be used to configure more options than the type and default - like some basic validation rules, hiding None from output, overriding serialization & deserialization names, etc.
Fields that don't provide a default (either as the field's assigned value, or using the default kwarg to a custom field) will be required.
Attributes without a type annotation will not be treated as fields and are just static class members.
Instance methods of a model can use the values freely. Note they may be called even if the model doesn't pass the validation rules.

Model instances are created by passing dictionaries. Models can then be validated and serialized back to raw data:

model = Employee({'name': 'Jane Doe', 'female': True, 'salary': 150000})
model.validate()
assert model.name == 'Jane Doe'
assert model.department == 'Engineering'
assert model.female is True
assert model.salary == 150000.0
assert model.human is True
assert model.greeting() == 'Dear Ms. Jane Doe'
model.salary = 200000.0
primitive = model.to_primitive()
assert primitive == {'name': 'Jane Doe', 'department': 'Engineering', 'female': True, 'salary': 200000.0}

Some amount of conversion will be done automatically - ints to floats, floats without decimals to ints, strings to numbers and booleans, anything to strings.
Calls to validate will pass without an exception if there is no problem.
Fields can be freely accessed as attributes to retrieve their values, class attributes and all methods work normally.
Fields can also be assigned to. Validation can be repeated after changing a model's initial values.
Data can be serialized back to primitive structures using to_primitve or it's alias serialize.

Validation errors raised from the validate method can report common data issues, such as missing required fields and broken validation rules:

model = Employee({'department': None, 'salary': '10'})
try:
    model.validate()
    assert False, 'Validation fails due to errors below'  # pragma: no cover
except ValidationError as e:
    assert str(e) == 'name: This field is required'
    assert e.errors == {
        'name': ['This field is required'],
        'department': ['This field is required'],
        'salary': ['Must be at least 42.0'],
    }
assert model.name is Missing
assert model.department is None
assert model.salary == 10.0
assert model.to_primitive() == {'department': None, 'female': None, 'salary': 10.0}

The string representation of a ValidationError reports the first errors, while the errors property can be used to identify all errors in the model structure, including nested models.
Not providing a value to a required field will trigger a validation error for the field. The model's attribute can still be accessed but will contain the special Missing object.
Providing None to a field with a default will override the default. Unless the field is Optional, this also causes a validation error.
There are a few built-in validation rules for the specific Field types. Custom validation can be implemented using [validator methods].
Omitted required fields in the input will be missing in serialized data as well.

Conversion errors are raised during model initialization if the input to a model wasn't a dictionary or any of the fields contain data that cannot be interpreted by the corresponding field.

try:
    model = Employee({'name': 'Alien', 'female': 'Unknown'})
except ConversionError as e:
    assert str(e) == 'female: Value must be a boolean or a true/false/yes/no string value'

ConversionErrors also have the errors property with the same structure of errors, but will only have one error.

Compound value fields

Fields can also be lists and dictionaries of other fields:

from typing import List, Dict

from stereotype import Model, ListField, StrField, FloatField, ValidationError
from stereotype.fields.compound import DictField


class Book(Model):
    name: str
    authors: List[str] = ListField(min_length=1)
    reviews: Dict[str, float] = DictField(
        key_field=StrField(min_length=1),
        value_field=FloatField(min_value=0, max_value=5),
        default=dict,
    )


model = Book({'name': 'The Go Programming Language'})
model.authors = ['Alan A.A. Donovan', 'Brian W. Kernighan']
model.validate()
assert model.to_primitive() == {
    'name': 'The Go Programming Language',
    'authors': ['Alan A.A. Donovan', 'Brian W. Kernighan'],
    'reviews': {},
}

Lists are created using the List type annotation and any other supported annotation, including other lists.
You can specify validation rules by providing an explicit ListField, but this is not necessary. Validation rules for items may be specified by providing an explicit field instance in the item_field argument.
Dictionaries are created using the Dict type annotation, with an atomic key type and any value type.
Extra options can again be specified with an explicit DictField, such as validation rules for the keys and values.
You can use the list and dict constructors as defaults if the fields are not required.

model = Book({
    'name': "The Hitchhiker's Guide to the Galaxy",
    'authors': ['Douglas Adams', None],
    'reviews': {'dolphins': 4.2, 'whales': -2},
})
assert model.reviews['dolphins'] == 4.2
try:
    model.validate()
    assert False, 'Validation fails due to errors below'  # pragma: no cover
except ValidationError as e:
    assert e.errors == {
        'authors': {'1': ['This field is required']},
        'reviews': {'whales': ['Must be between 0 and 5']},
    }

Validation errors in compound fields will be located using nested dictionaries to the individual items.

Deeper structures with model fields

Models can be used as model fields themselves.

from __future__ import annotations
from typing import Optional, Any, Union, List
from stereotype import Model


class Variable(Model):
    name: str


class IsEqual(Model):
    type = 'equal'

    variable: Variable
    constant: Constant


class Constant(Model):
    value: Optional[Any] = None


equality = IsEqual({'variable': {'name': 'x'}, 'constant': {'value': 42}})
assert equality.variable.name == 'x'
assert equality.type == 'equal'
assert equality.to_primitive() == {'variable': {'name': 'x'}, 'constant': {'value': 42}}

Simply use the model's identifier in the annotation.
Model fields can also be Optional and can have a callable default, such as the model class itself.
Add from __future__ import annotations to have fields with models that weren't declared yet.
Note Any can be used to contain values of any type that won't be further processed.
Note IsEqual.type doesn't have a type annotation, hence isn't a field and won't be present in serialized data.

The non-field string type attribute has a special usage when using dynamic model fields:

class Negation(Model):
    type = 'not'

    formula: Union[IsEqual, Conjunction]


class Conjunction(Model):
    type = 'and'

    operands: List[Union[IsEqual, Negation, Conjunction]] = list


equality_copy = equality.copy(deep=True)
model = Conjunction({'operands': [
    {'type': 'not', 'formula': equality_copy},
    {'type': 'equal', 'variable': {'name': 'y'}, 'constant': {}},
]})
assert isinstance(model.operands[0], Negation)
assert model.operands[0].type == 'not'
assert model.to_primitive() == {'operands': [
    {'type': 'not', 'formula': {'type': 'equal', 'variable': {'name': 'x'}, 'constant': {'value': 42}}},
    {'type': 'equal', 'variable': {'name': 'y'}, 'constant': {'value': None}},
]}