Object schemas

Declaration

Object schemas are defined with Object class by passing it a dictionary mapping field names to Type instances.

So given an object

class Person(object):
    def __init__(self, name, age):
        self.name = name
        self.age = age

You can define it’s type like this:

from lollipop.types import Object, String, Integer

PersonType = Object({
    'name': String(),
    'age': Integer(),
})

It will allow serializing Person types to Python’s basic types (that you can use to serialize to JSON) or validate that basic Python data:

PersonType.dump(Person('John', 38))
# => {"name": "John", "age": 38}

PersonType.validate({"name": "John"})
# => {"age": "Value is required"}

PersonType.load({"name": "John", "age": 38})
# => {"name": "John", "age": 38}

Yet it loads to same basic type dict instead of real object. To fix that, you need to provide a data constructor to type object:

PersonType = Object({
    'name': String(),
    'age': Integer(),
}, constructor=Person)

PersonType.load({"name": "John", "age": 38})
# => Person(name="John", age=38)

Constructor function should take field values as keyword arguments and return constructed object.

Value access

When you serialize (dump) objects, field values are expected to be object attributes. But library actually allows controlling that. This is done with Field class instances. When you define your object and pass types for it’s fields, what really happens is those types are wrapped with a Field subclass objects. The actual object fields are defined like this:

PersonType = Object({
    'name': AttributeField(String()),
    'age': AttributeField(Integer()),
})

Passing just a Type instances for field types is just a shortcut to wrap them all with a default field type which is AttributeField. You can change default field type with Object.default_field_type argument:

PersonType = Object({
    'name': String(),
    'age': Integer(),
}, default_field_type=AttributeField)

And you can actually mix fields defined with just Type with fields defined with Field. The first ones will be wrapped with default field type while the later ones will be used as is.

AttributeField is probably the one that would be used most of the time. It extracts value for serialization from object attribute with the same name as the field name. You can change the name of attribute to extract value from:

Person = namedtuple('Person', ['full_name'])

PersonType = Object({'name': AttributeField(String(), attribute='full_name')})

PersonType.dump(Person('John Doe'))  # => {'name': 'John Doe'}

Other useful instances are MethodField which calls given method on the object to get value instead of getting attribute or FunctionField which uses given function on a serialized object to get value. For last one there is another shortcut: if you provide a value for a field which is not Type and not Field then it will be wrapped with a Constant and then into default field type.

# Following lines are equivalent
Object({'answer': AttributeField(Constant(42))}).dump(object())  # => {'answer': 42}
Object({'answer': 42}).dump(object())  # => {'answer': 42}

Updating objects in-place

After you have created your initial version of your objects with data obtained from user you might want to allow user to update them. And you might want to allow your users to specify only changed attributes without sending all of them. Or after creation your object store additional information that you do not want to expose to users or allow users to modify, e.g. object ID or creation date. So you make them dump only or do not include them in schema at all. But since load() method return you a new copy of your object, that object does not contain those additional data. Luckily this library allows updating existing objects in-place:

user = User.get(user_id)
try:
    UserType.load_into(user, {'name': 'John Doe'})
    User.save(user)
except ValidationError as ve:
    # .. handle user validation error

If you do not want to alter existing object but still want your users to specify partial data on update, you can declare your object type as “immutable”. In this case it won’t modify your objects but will create new ones with data merged from existing object and data being deserialized:

UserType = Object({
    'name': String(),
    'birthdate': Date(),
    # ...
}, constructor=User, immutable=True)

user = User.get(user_id)
try:
    user1 = UserType.load_into(user, {'name': 'John Doe'})
    User.save(user1)
except ValidationError as ve:
    # .. handle user validation error

You can disable in-place update on per-invocation basis with inplace argument:

user1 = UserType.load_into(user, new_data, inplace=False)

For partial update validation there is a validate_for():

errors = UserType.validate_for(user, new_data)

When doing partial update all new data is validated during deserialization. Also, whole-object validations are also run.

How values are put back into object is controlled by Field subclasses that you use in object schema declaration (e.g. AttributeField, MethodField or FunctionField. See Value access for details).

Object Schema Inheritance

To be able to allow reusing parts of schema, you can supply a base Object:

BaseType = Object({'base': String()})
InheritedType = Object(BaseType, {'foo': Integer()})

# is the same as
InheritedType = Object({'base': String(), 'foo': Integer()})

You can actually supply multple base types which allows using them as mixins:

TimeStamped = Object({'created_at': DateTime(), 'updated_at': DateTime()})

BaseType = Object({'base': String()})
InheritedType = Object([BaseType, TimeStamped], {'foo': Integer()})

Polymorphic types

Sometimes you need a way to serialize and deserialize values of different types put in the same list. Or maybe you value can be of either one of given types. E.g. you have a graphical application which operates with objects of different shapes:

class Point(object):
    def __init__(self, x, y):
        self.x = x
        self.y = y

class Shape(object):
    pass

class Circle(Shape):
    def __init__(self, center, radius):
        self.center = center
        self.radius = radius

class Rectangle(Shape):
    def __init__(self, left_top, right_bottom):
        self.left_top = left_top
        self.right_bottom = right_bottom

PointType = Object({'x': Integer(), 'y': Integer()}, constructor=Point)

CircleType = Object({
    'center': PointType,
    'radius': Integer
}, constructor=Circle)

RectangleType = Object({
    'left_top': PointType,
    'right_bottom': PointType,
}, constructor=Rectangle)

To support that library provides a special type - OneOf:

def with_type_annotation(subject_type, type_name):
    return Object(subject_type, {'type': type_name},
                  constructor=subject_type.constructor)

AnyShapeType = OneOf(
    {
        'circle': with_type_annotation(CircleType, 'circle'),
        'rectangle': with_type_annotation(RectangleType, 'rectangle'),
    },
    dump_hint=lambda obj: obj.__class__.__name__.lower(),
    load_hint=dict_value_hint('type'),
)

dumped = List(AnyShapeType).dump([
    Circle(Point(5, 8), 4), Rectangle(Point(1, 10), Point(10, 1))
])
# => [
#   {'type': 'circle',
#    'center': {'x': 5, 'y': 8},
#    'radius': 4},
#   {'type': 'rectangle',
#    'left_top': {'x': 1, 'y': 10},
#    'right_bottom': {'x': 10, 'y': 1}}]

List(AnyShapeType).load(dumped)
# => [Circle(Point(5, 8), 4), Rectangle(Point(1, 10), Point(10, 1))]

OneOf uses user supplied functions to determine which particular type to use during serialization/deserialization. It helps returning proper error messages. If you’re not interested in providing detailed error message, you can just supply all types as a list. OneOf will try to use each of them in given order returning first successfull result. If all types return errors it will provide generic error message.

Two-way type references

Nesting object types inside another objects is very easy since object types are just another types. But sometimes you might have multiple application entities that reference each other. E.g. you model a library and inside you have Person model and Book model. Person can be author of multiple books and each book has (for simplicity lets assume only one) author. You want your Person type to have a reference to Book and Book to have reference to Person types.

For that matter library provides a storage for types which can provide you with delayed type resolving:

import lollipop.types as lt
from lollupop.type_registry import TypeRegistry

TYPES = TypeRegistry()

PersonType = TYPES.add('Person', lt.Object({
    'name': lt.String(),
    'books': lt.List(lt.Object(TYPES['Book'], exclude='author')),
}, constructor=Person))

BookType = TYPES.add('Book', lt.Object({
    'title': lt.String(),
    'author': lt.Object(TYPES['Person'], exclude='books'),
}, constructor=Book))

Here you can see that we get a types from our registry to use them as a base object types and then customize them (e.g. exclude some fields to eliminate circular dependency). The Object type is designed to not access base class’ properties and methods until is needed thus allowing to postpone actual type resolution and thus allowing forward references to types.

Type type registry is not a global instance, but instance local to whatever degree you want it to be local. If your application schemas can fit into one module, you declare registry in that module. If your schemas span multiple modules, it is better to put registry in a separate module (along with any custom type declarations that you might have) and import it where needed.

You can even do self references inside Object declarations. Here is example of type declaration for lollipop errors format:

TYPES = TypeRegistry()
ErrorsType = TYPES.add('Errors', lt.OneOf([
    lt.String,
    lt.List(lt.String()),
    lt.Dict(TYPES['Errors']),
]))