Skip to content

Data model and resolvers

Almost everything in the Data model section remains valid in GraphQL integration, with a few differences.

GraphQL specific data model

Enum

Enum members are represented in the schema using their name instead of their value. This is more consistent with the way GraphQL represents enumerations.

TypedDict

TypedDict is not supported as an output type. (see FAQ)

Union

Unions are only supported between output object type, which means dataclass and NamedTuple (and conversions/dataclass model).

There are 2 exceptions which can be always be used in Union:

  • None/Optional: Types are non-null (marked with an exclamation mark ! in GraphQL schema) by default; Optional types however results in normal GraphQL types (without !).
  • apischema.UndefinedType: it is simply ignored. It is useful in resolvers, see following section

Non-null

Types are assumed to be non-null by default, as in Python typing. Nullable types are obtained using typing.Optional (or typing.Union with a None argument).

Note

There is one exception, when resolver parameter default value is not serializable (and thus cannot be included in the schema), the parameter type is then set as nullable to make the parameter non-required. For example parameters not Optional but with Undefined default value will be marked as nullable. This is only for the schema, the default value is still used at execution.

Undefined

In output, Undefined is converted to None; so in the schema, Union[T, UndefinedType] will be nullable.

In input, fields become nullable when Undefined is their default value.

Interfaces

Interfaces are simply classes marked with apischema.graphql.interface decorator. An object type implements an interface when its class inherits from an interface-marked class, or when it has flattened fields of interface-marked dataclass.

from dataclasses import dataclass

from graphql import print_schema

from apischema.graphql import graphql_schema, interface


@interface
@dataclass
class Bar:
    bar: int


@dataclass
class Foo(Bar):
    baz: str


def foo() -> Foo | None:
    ...


schema = graphql_schema(query=[foo])
schema_str = """\
type Query {
  foo: Foo
}

type Foo implements Bar {
  bar: Int!
  baz: String!
}

interface Bar {
  bar: Int!
}
"""
assert print_schema(schema) == schema_str

Resolvers

All dataclass/NamedTuple fields (excepted skipped) are resolved with their alias in the GraphQL schema.

Custom resolvers can also be added by marking methods with apischema.graphql.resolver decorator — resolvers share a common interface with apischema.serialized, with a few differences.

Methods can be synchronous or asynchronous (defined with async def or annotated with an typing.Awaitable return type).

Resolvers parameters are included in the schema with their type, and their default value.

from dataclasses import dataclass

from graphql import print_schema

from apischema.graphql import graphql_schema, resolver


@dataclass
class Bar:
    baz: int


@dataclass
class Foo:
    @resolver
    async def bar(self, arg: int = 0) -> Bar:
        ...


async def foo() -> Foo | None:
    ...


schema = graphql_schema(query=[foo])
schema_str = """\
type Query {
  foo: Foo
}

type Foo {
  bar(arg: Int! = 0): Bar!
}

type Bar {
  baz: Int!
}
"""
assert print_schema(schema) == schema_str

GraphQLResolveInfo parameter

Resolvers can have an additional parameter of type graphql.GraphQLResolveInfo (or Optional[graphql.GraphQLResolveInfo]), which is automatically injected when the resolver is executed in the context of a GraphQL request. This parameter contains the info about the current GraphQL request being executed.

Undefined parameter default — null vs. undefined

Undefined can be used as default value of resolver parameters. It can be to distinguish a null input from an absent/undefined input. In fact, null value will result in a None argument where no value will use the default value, Undefined so.

from graphql import graphql_sync

from apischema import Undefined, UndefinedType
from apischema.graphql import graphql_schema


def arg_is_absent(arg: int | UndefinedType | None = Undefined) -> bool:
    return arg is Undefined


schema = graphql_schema(query=[arg_is_absent])
assert graphql_sync(schema, "{argIsAbsent(arg: null)}").data == {"argIsAbsent": False}
assert graphql_sync(schema, "{argIsAbsent}").data == {"argIsAbsent": True}

Error handling

Errors occurring in resolvers can be caught in a dedicated error handler registered with error_handler parameter. This function takes in parameters the exception, the object, the info and the kwargs of the failing resolver; it can return a new value or raise the current or another exception — it can for example be used to log errors without throwing the complete serialization.

The resulting serialization type will be a Union of the normal type and the error handling type; if the error handler always raises, use typing.NoReturn annotation.

error_handler=None correspond to a default handler which only return None — exception is thus discarded and the resolver type becomes Optional.

The error handler is only executed by apischema serialization process, it's not added to the function, so this one can be executed normally and raise an exception in the rest of your code.

Error handler can be synchronous or asynchronous.

from dataclasses import dataclass
from logging import getLogger
from typing import Any

import graphql
from graphql.utilities import print_schema

from apischema.graphql import graphql_schema, resolver

logger = getLogger(__name__)


def log_error(
    error: Exception, obj: Any, info: graphql.GraphQLResolveInfo, **kwargs
) -> None:
    logger.error(
        "Resolve error in %s", ".".join(map(str, info.path.as_list())), exc_info=error
    )
    return None


@dataclass
class Foo:
    @resolver(error_handler=log_error)
    def bar(self) -> int:
        raise RuntimeError("Bar error")

    @resolver
    def baz(self) -> int:
        raise RuntimeError("Baz error")


def foo(info: graphql.GraphQLResolveInfo) -> Foo:
    return Foo()


schema = graphql_schema(query=[foo])
# Notice that bar is Int while baz is Int!
schema_str = """\
type Query {
  foo: Foo!
}

type Foo {
  bar: Int
  baz: Int!
}
"""
assert print_schema(schema) == schema_str
# Logs "Resolve error in foo.bar", no error raised
assert graphql.graphql_sync(schema, "{foo{bar}}").data == {"foo": {"bar": None}}
# Error is raised
assert graphql.graphql_sync(schema, "{foo{baz}}").errors[0].message == "Baz error"

Parameters metadata

Resolvers parameters can have metadata like dataclass fields. They can be passed using typing.Annotated.

from dataclasses import dataclass
from typing import Annotated

from graphql.utilities import print_schema

from apischema import alias, schema
from apischema.graphql import graphql_schema, resolver


@dataclass
class Foo:
    @resolver
    def bar(
        self, param: Annotated[int, alias("arg") | schema(description="argument")]
    ) -> int:
        return param


def foo() -> Foo:
    return Foo()


schema_ = graphql_schema(query=[foo])
# Notice that bar is Int while baz is Int!
schema_str = '''\
type Query {
  foo: Foo!
}

type Foo {
  bar(
    """argument"""
    arg: Int!
  ): Int!
}
'''
assert print_schema(schema_) == schema_str

Note

Metadata can also be passed with parameters_metadata parameter; it takes a mapping of parameter names as key and mapped metadata as value.

Parameters base schema

Following the example of type/field/method base schema, resolver parameters also support a base schema definition

import inspect
from dataclasses import dataclass
from typing import Any, Callable

import docstring_parser
from graphql.utilities import print_schema

from apischema import schema, settings
from apischema.graphql import graphql_schema, resolver
from apischema.schemas import Schema


@dataclass
class Foo:
    @resolver
    def bar(self, arg: str) -> int:
        """bar method

        :param arg: arg parameter
        """
        ...


def method_base_schema(tp: Any, method: Callable, alias: str) -> Schema | None:
    return schema(description=docstring_parser.parse(method.__doc__).short_description)


def parameter_base_schema(
    method: Callable, parameter: inspect.Parameter, alias: str
) -> Schema | None:
    for doc_param in docstring_parser.parse(method.__doc__).params:
        if doc_param.arg_name == parameter.name:
            return schema(description=doc_param.description)
    return None


settings.base_schema.method = method_base_schema
settings.base_schema.parameter = parameter_base_schema


def foo() -> Foo:
    ...


schema_ = graphql_schema(query=[foo])
schema_str = '''\
type Query {
  foo: Foo!
}

type Foo {
  """bar method"""
  bar(
    """arg parameter"""
    arg: String!
  ): Int!
}
'''
assert print_schema(schema_) == schema_str

ID type

GraphQL ID has no precise specification and is defined according API needs; it can be a UUID or and ObjectId, etc.

apischema.graphql_schema has a parameter id_types which can be used to define which types will be marked as ID in the generated schema. Parameter value can be either a collection of types (each type will then be mapped to ID scalar), or a predicate returning if the given type must be marked as ID.

from dataclasses import dataclass
from uuid import UUID

from graphql import print_schema

from apischema.graphql import graphql_schema


@dataclass
class Foo:
    bar: UUID


def foo() -> Foo | None:
    ...


# id_types={UUID} is equivalent to id_types=lambda t: t in {UUID}
schema = graphql_schema(query=[foo], id_types={UUID})
schema_str = """\
type Query {
  foo: Foo
}

type Foo {
  bar: ID!
}
"""
assert print_schema(schema) == schema_str

Note

ID type could also be identified using typing.Annotated and a predicate looking into annotations.

apischema also provides a simple ID type with apischema.graphql.ID. It is just defined as a NewType of string, so you can use it when you want to manipulate raw ID strings in your resolvers.

ID encoding

ID encoding can directly be controlled the id_encoding parameters of graphql_schema. A current practice is to use base64 encoding for ID.

from base64 import b64decode, b64encode
from dataclasses import dataclass
from uuid import UUID

from graphql import graphql_sync

from apischema.graphql import graphql_schema


@dataclass
class Foo:
    id: UUID


def foo() -> Foo | None:
    return Foo(UUID("58c88e87-5769-4723-8974-f9ec5007a38b"))


schema = graphql_schema(
    query=[foo],
    id_types={UUID},
    id_encoding=(
        lambda s: b64decode(s).decode(),
        lambda s: b64encode(s.encode()).decode(),
    ),
)

assert graphql_sync(schema, "{foo{id}}").data == {
    "foo": {"id": "NThjODhlODctNTc2OS00NzIzLTg5NzQtZjllYzUwMDdhMzhi"}
}

Note

You can also use relay.base64_encoding (see next section)

Note

ID serialization (respectively deserialization) is applied after apischema conversions (respectively before apischema conversion): in the example, uuid is already converted into string before being passed to id_serializer.

If you use base64 encodeing and an ID type which is converted by apischema to a base64 str, you will get a double encoded base64 string

Tagged unions

Important

This feature has a provisional status, as the concerned GraphQL RFC is not finalized.

apischema provides a apischema.tagged_unions.TaggedUnion base class which helps to implement the tagged union pattern. It's fields must be typed using apischema.tagged_unions.Tagged generic type.

from dataclasses import dataclass

from pytest import raises

from apischema import Undefined, ValidationError, alias, deserialize, schema, serialize
from apischema.tagged_unions import Tagged, TaggedUnion, get_tagged


@dataclass
class Bar:
    field: str


class Foo(TaggedUnion):
    bar: Tagged[Bar]
    # Tagged can have metadata like a dataclass fields
    i: Tagged[int] = Tagged(alias("baz") | schema(min=0))


# Instantiate using class fields
tagged_bar = Foo.bar(Bar("value"))
# you can also use default constructor, but it's not typed-checked
assert tagged_bar == Foo(bar=Bar("value"))

# All fields that are not tagged are Undefined
assert tagged_bar.bar is not Undefined and tagged_bar.i is Undefined
# get_tagged allows to retrieve the tag and it's value
# (but the value is not typed-checked)
assert get_tagged(tagged_bar) == ("bar", Bar("value"))

# (De)serialization works as expected
assert deserialize(Foo, {"bar": {"field": "value"}}) == tagged_bar
assert serialize(Foo, tagged_bar) == {"bar": {"field": "value"}}

with raises(ValidationError) as err:
    deserialize(Foo, {"unknown": 42})
assert err.value.errors == [{"loc": ["unknown"], "msg": "unexpected property"}]

with raises(ValidationError) as err:
    deserialize(Foo, {"bar": {"field": "value"}, "baz": 0})
assert err.value.errors == [
    {"loc": [], "msg": "property count greater than 1 (maxProperties)"}
]

JSON schema

Tagged unions JSON schema uses minProperties: 1 and maxProperties: 1.

from dataclasses import dataclass

from apischema.json_schema import deserialization_schema, serialization_schema
from apischema.tagged_unions import Tagged, TaggedUnion


@dataclass
class Bar:
    field: str


class Foo(TaggedUnion):
    bar: Tagged[Bar]
    baz: Tagged[int]


assert (
    deserialization_schema(Foo)
    == serialization_schema(Foo)
    == {
        "$schema": "http://json-schema.org/draft/2020-12/schema#",
        "type": "object",
        "properties": {
            "bar": {
                "type": "object",
                "properties": {"field": {"type": "string"}},
                "required": ["field"],
                "additionalProperties": False,
            },
            "baz": {"type": "integer"},
        },
        "additionalProperties": False,
        "minProperties": 1,
        "maxProperties": 1,
    }
)

GraphQL schema

As tagged unions are not (yet?) part of the GraphQL spec, they are just implemented as normal (input) object type with nullable fields. An error is raised if several tags are passed in input.

from dataclasses import dataclass

from graphql import graphql_sync
from graphql.utilities import print_schema

from apischema.graphql import graphql_schema
from apischema.tagged_unions import Tagged, TaggedUnion


@dataclass
class Bar:
    field: str


class Foo(TaggedUnion):
    bar: Tagged[Bar]
    baz: Tagged[int]


def query(foo: Foo) -> Foo:
    return foo


schema = graphql_schema(query=[query])
schema_str = """\
type Query {
  query(foo: FooInput!): Foo!
}

type Foo {
  bar: Bar
  baz: Int
}

type Bar {
  field: String!
}

input FooInput {
  bar: BarInput
  baz: Int
}

input BarInput {
  field: String!
}
"""
assert print_schema(schema) == schema_str

query_str = """
{
    query(foo: {bar: {field: "value"}}) {
        bar {
            field
        }
        baz
    }
}
"""
assert graphql_sync(schema, query_str).data == {
    "query": {"bar": {"field": "value"}, "baz": None}
}

FAQ

Why TypedDict is not supported as an output type?

At first, TypedDict subclasses are not real classes, so they cannot be used to check types at runtime. Runtime check is however requried to disambiguate unions/interfaces. A hack could be done to solve this issue, but there is another one which cannot be hacked: TypedDict inheritance hierarchy is lost at runtime, so they don't play nicely with the interface concept.