Django REST framework 3.5

The 3.5 release is the second in a planned series that is addressing schema generation, hypermedia support, API client libraries, and finally realtime support.


The 3.5 release would not have been possible without our collaborative funding model. If you use REST framework commercially and would like to see this work continue, we strongly encourage you to invest in its continued development by signing up for a paid plan.

Many thanks to all our sponsors, and in particular to our premium backers, Rover, Sentry, Stream, and Machinalis.

Improved schema generation

Docstrings on views are now pulled through into schema definitions, allowing you to use the schema definition to document your API.

There is now also a shortcut function, get_schema_view(), which makes it easier to adding schema views to your API.

For example, to include a swagger schema to your API, you would do the following:

  • Run pip install django-rest-swagger.

  • Add 'rest_framework_swagger' to your INSTALLED_APPS setting.

  • Include the schema view in your URL conf:

from rest_framework.schemas import get_schema_view
from rest_framework_swagger.renderers import OpenAPIRenderer, SwaggerUIRenderer

schema_view = get_schema_view(
    title='Example API',
    renderer_classes=[OpenAPIRenderer, SwaggerUIRenderer]

urlpatterns = [
    url(r'^swagger/$', schema_view),

There have been a large number of fixes to the schema generation. These should resolve issues for anyone using the latest version of the django-rest-swagger package.

Some of these changes do affect the resulting schema structure, so if you're already using schema generation you should make sure to review the deprecation notes, particularly if you're currently using a dynamic client library to interact with your API.

Finally, we're also now exposing the schema generation as a publicly documented API, allowing you to more easily override the behaviour.

Requests test client

You can now test your project using the requests library.

This exposes exactly the same interface as if you were using a standard requests session instance.

client = RequestsClient()
response = client.get('http://testserver/users/')
assert response.status_code == 200

Rather than sending any HTTP requests to the network, this interface will coerce all outgoing requests into WSGI, and call into your application directly.

Core API client

You can also now test your project by interacting with it using the coreapi client library.

# Fetch the API schema
client = CoreAPIClient()
schema = client.get('http://testserver/schema/')

# Create a new organisation
params = {'name': 'MegaCorp', 'status': 'active'}
client.action(schema, ['organisations', 'create'], params)

# Ensure that the organisation exists in the listing
data = client.action(schema, ['organisations', 'list'])
assert(len(data) == 1)
assert(data == [{'name': 'MegaCorp', 'status': 'active'}])

Again, this will call directly into the application using the WSGI interface, rather than making actual network calls.

This is a good option if you are planning for clients to mainly interact with your API using the coreapi client library, or some other auto-generated client.

Live tests

One interesting aspect of both the requests client and the coreapi client is that they allow you to write tests in such a way that they can also be made to run against a live service.

By switching the WSGI based client instances to actual instances of requests.Session or coreapi.Client you can have the test cases make actual network calls.

Being able to write test cases that can exercise your staging or production environment is a powerful tool. However in order to do this, you'll need to pay close attention to how you handle setup and teardown to ensure a strict isolation of test data from other live or staging data.

RAML support

We now have preliminary support for RAML documentation generation.

RAML Example

Further work on the encoding and documentation generation is planned, in order to make features such as the 'Try it now' support available at a later date.

This work also now means that you can use the Core API client libraries to interact with APIs that expose a RAML specification. The RAML codec gives some examples of interacting with the Spotify API in this way.

Validation codes

Exceptions raised by REST framework now include short code identifiers. When used together with our customizable error handling, this now allows you to modify the style of API error messages.

As an example, this allows for the following style of error responses:

    "message": "You do not have permission to perform this action.",
    "code": "permission_denied"

This is particularly useful with validation errors, which use appropriate codes to identify differing kinds of failure...

    "name": {"message": "This field is required.", "code": "required"},
    "age": {"message": "A valid integer is required.", "code": "invalid"}

Client upload & download support

The Python coreapi client library and the Core API command line tool both now fully support file uploads and downloads.


Generating schemas from Router

The router arguments for generating a schema view, such as schema_title, are now pending deprecation.

Instead of using DefaultRouter(schema_title='Example API'), you should use the get_schema_view() function, and include the view in your URL conf.

Make sure to include the view before your router urls. For example:

from rest_framework.schemas import get_schema_view
from my_project.routers import router

schema_view = get_schema_view(title='Example API')

urlpatterns = [
    url('^$', schema_view),
    url(r'^', include(router.urls)),

Schema path representations

The 'pk' identifier in schema paths is now mapped onto the actually model field name by default. This will typically be 'id'.

This gives a better external representation for schemas, with less implementation detail being exposed. It also reflects the behaviour of using a ModelSerializer class with fields = '__all__'.

You can revert to the previous behaviour by setting 'SCHEMA_COERCE_PATH_PK': False in the REST framework settings.

Schema action name representations

The internal retrieve() and destroy() method names are now coerced to an external representation of read and delete.

You can revert to the previous behaviour by setting 'SCHEMA_COERCE_METHOD_NAMES': {} in the REST framework settings.


The functionality of the built-in DjangoFilterBackend is now completely included by the django-filter package.

You should change your imports and REST framework filter settings as follows:

  • rest_framework.filters.DjangoFilterBackend becomes django_filters.rest_framework.DjangoFilterBackend.
  • rest_framework.filters.FilterSet becomes django_filters.rest_framework.FilterSet.

The existing imports will continue to work but are now pending deprecation.

CoreJSON media type

The media type for CoreJSON is now application/json+coreapi, rather than the previous application/vnd.json+coreapi. This brings it more into line with other custom media types, such as those used by Swagger and RAML.

The clients currently accept either media type. The old style-media type will be deprecated at a later date.

ModelSerializer 'fields' and 'exclude'

ModelSerializer and HyperlinkedModelSerializer must include either a fields option, or an exclude option. The fields = '__all__' shortcut may be used to explicitly include all fields.

Failing to set either fields or exclude raised a pending deprecation warning in version 3.3 and raised a deprecation warning in 3.4. Its usage is now mandatory.