Better filtering for Django REST Framework
Clone or download
Ryan P Kilby
Ryan P Kilby Add gitignore
Latest commit c5ffc82 Jul 29, 2018

README.rst

Django Rest Framework Filters

https://travis-ci.org/philipn/django-rest-framework-filters.png?branch=master

django-rest-framework-filters is an extension to Django REST framework and Django filter that makes it easy to filter across relationships. Historically, this extension also provided a number of additional features and fixes, however the number of features has shrunk as they are merged back into django-filter.

Using django-rest-framework-filters, we can easily do stuff like:

/api/article?author__first_name__icontains=john
/api/article?is_published!=true

! These docs pertain to the upcoming 1.0 release. Current docs can be found here.


Features

  • Easy filtering across relationships.
  • Support for method filtering across relationships.
  • Automatic filter negation with a simple param!=value syntax.
  • Backend for complex operations on multiple filtered querysets. eg, q1 | q2.

Requirements

  • Python: 3.4, 3.5, 3.6
  • Django: 1.11, 2.0, 2.1b1
  • DRF: 3.8
  • django-filter: 2.0

Installation

Install with pip, or your preferred package manager:

$ pip install djangorestframework-filters

Add to your INSTALLED_APPS setting:

INSTALLED_APPS = [
    'rest_framework_filters',
    ...
]

FilterSet usage

Upgrading from django-filter to django-rest-framework-filters is straightforward:

  • Import from rest_framework_filters instead of from django_filters
  • Use the rest_framework_filters backend instead of the one provided by django_filter.
# django-filter
from django_filters.rest_framework import FilterSet, filters

class ProductFilter(FilterSet):
    manufacturer = filters.ModelChoiceFilter(queryset=Manufacturer.objects.all())
    ...


# django-rest-framework-filters
import rest_framework_filters as filters

class ProductFilter(filters.FilterSet):
    manufacturer = filters.ModelChoiceFilter(queryset=Manufacturer.objects.all())
    ...

To use the django-rest-framework-filters backend, add the following to your settings:

REST_FRAMEWORK = {
    'DEFAULT_FILTER_BACKENDS': (
        'rest_framework_filters.backends.RestFrameworkFilterBackend', ...
    ),
    ...

Once configured, you can continue to use all of the filters found in django-filter.

Filtering across relationships

You can easily traverse multiple relationships when filtering by using RelatedFilter:

from rest_framework import viewsets
import rest_framework_filters as filters


class ManagerFilter(filters.FilterSet):
    class Meta:
        model = Manager
        fields = {'name': ['exact', 'in', 'startswith']}


class DepartmentFilter(filters.FilterSet):
    manager = filters.RelatedFilter(ManagerFilter, field_name='manager', queryset=Manager.objects.all())

    class Meta:
        model = Department
        fields = {'name': ['exact', 'in', 'startswith']}


class CompanyFilter(filters.FilterSet):
    department = filters.RelatedFilter(DepartmentFilter, field_name='department', queryset=Department.objects.all())

    class Meta:
        model = Company
        fields = {'name': ['exact', 'in', 'startswith']}


# company viewset
class CompanyView(viewsets.ModelViewSet):
    filter_class = CompanyFilter
    ...

Example filter calls:

/api/companies?department__name=Accounting
/api/companies?department__manager__name__startswith=Bob

queryset callables

Since RelatedFilter is a subclass of ModelChoiceFilter, the queryset argument supports callable behavior. In the following example, the set of departments is restricted to those in the user's company.

def departments(request):
    company = request.user.company
    return company.department_set.all()

class EmployeeFilter(filters.FilterSet):
    department = filters.RelatedFilter(filterset=DepartmentFilter, queryset=departments)
    ...

Recursive & Circular relationships

Recursive relations are also supported. Provide the module path as a string in place of the filterset class.

class PersonFilter(filters.FilterSet):
    name = filters.AllLookupsFilter(field_name='name')
    best_friend = filters.RelatedFilter('people.views.PersonFilter', field_name='best_friend', queryset=Person.objects.all())

    class Meta:
        model = Person

This feature is also useful for circular relationships, where a related filterset may not yet be created. Note that you can pass the related filterset by name if it's located in the same module as the parent filterset.

class BlogFilter(filters.FilterSet):
    post = filters.RelatedFilter('PostFilter', queryset=Post.objects.all())

class PostFilter(filters.FilterSet):
    blog = filters.RelatedFilter('BlogFilter', queryset=Blog.objects.all())

Supporting Filter.method

django_filters.MethodFilter has been deprecated and reimplemented as the method argument to all filter classes. It incorporates some of the implementation details of the old rest_framework_filters.MethodFilter, but requires less boilerplate and is simpler to write.

  • It is no longer necessary to perform empty/null value checking.
  • You may use any filter class (CharFilter, BooleanFilter, etc...) which will validate input values for you.
  • The argument signature has changed from (name, qs, value) to (qs, name, value).
class PostFilter(filters.FilterSet):
    # Note the use of BooleanFilter, the original model field's name, and the method argument.
    is_published = filters.BooleanFilter(field_name='date_published', method='filter_is_published')

    class Meta:
        model = Post
        fields = ['title', 'content']

    def filter_is_published(self, qs, name, value):
        """
        `is_published` is based on the `date_published` model field.
        If the publishing date is null, then the post is not published.
        """
        # incoming value is normalized as a boolean by BooleanFilter
        isnull = not value
        lookup_expr = LOOKUP_SEP.join([name, 'isnull'])

        return qs.filter(**{lookup_expr: isnull})

class AuthorFilter(filters.FilterSet):
    posts = filters.RelatedFilter('PostFilter', queryset=Post.objects.all())

    class Meta:
        model = Author
        fields = ['name']

The above would enable the following filter calls:

/api/posts?is_published=true
/api/authors?posts__is_published=true

In the first API call, the filter method receives a queryset of posts. In the second, it receives a queryset of users. The filter method in the example modifies the lookup name to work across the relationship, allowing you to find published posts, or authors who have published posts.

Automatic Filter Negation/Exclusion

FilterSets support automatic exclusion using a simple param!=value syntax. This syntax internally sets the exclude property on the filter.

/api/page?title!=The%20Park

This syntax supports regular filtering combined with exclusion filtering. For example, the following would search for all articles containing "Hello" in the title, while excluding those containing "World".

/api/articles?title__contains=Hello&title__contains!=World

Note that most filters only accept a single query parameter. In the above, title__contains and title__contains! are interpreted as two separate query parameters. The following would probably be invalid, although it depends on the specifics of the individual filter class:

/api/articles?title__contains=Hello&title__contains!=World&title_contains!=Friend

Allowing any lookup type on a field

If you need to enable several lookups for a field, django-filter provides the dict-syntax for Meta.fields.

class ProductFilter(filters.FilterSet):
    class Meta:
        model = Product
        fields = {
            'price': ['exact', 'lt', 'gt', ...],
        }

django-rest-framework-filters also allows you to enable all possible lookups for any field. This can be achieved through the use of AllLookupsFilter or using the '__all__' value in the Meta.fields dict-style syntax. Generated filters (Meta.fields, AllLookupsFilter) will never override your declared filters.

Note that using all lookups comes with the same admonitions as enabling '__all__' fields in django forms (docs). Exposing all lookups may allow users to construct queries that inadvertently leak data. Use this feature responsibly.

class ProductFilter(filters.FilterSet):
    # Not overridden by `__all__`
    price__gt = filters.NumberFilter(field_name='price', lookup_expr='gt', label='Minimum price')

    class Meta:
        model = Product
        fields = {
            'price': '__all__',
        }

# or

class ProductFilter(filters.FilterSet):
    price = filters.AllLookupsFilter()

    # Not overridden by `AllLookupsFilter`
    price__gt = filters.NumberFilter(field_name='price', lookup_expr='gt', label='Minimum price')

    class Meta:
        model = Product

You cannot combine AllLookupsFilter with RelatedFilter as the filter names would clash.

class ProductFilter(filters.FilterSet):
    manufacturer = filters.RelatedFilter('ManufacturerFilter', queryset=Manufacturer.objects.all())
    manufacturer = filters.AllLookupsFilter()

To work around this, you have the following options:

class ProductFilter(filters.FilterSet):
    manufacturer = filters.RelatedFilter('ManufacturerFilter', queryset=Manufacturer.objects.all())

    class Meta:
        model = Product
        fields = {
            'manufacturer': '__all__',
        }

# or

class ProductFilter(filters.FilterSet):
    manufacturer = filters.RelatedFilter('ManufacturerFilter', queryset=Manufacturer.objects.all(), lookups='__all__')  # `lookups` also accepts a list

    class Meta:
        model = Product

Can I mix and match django-filter and django-rest-framework-filters?

Yes you can. django-rest-framework-filters is simply an extension of django-filter. Note that RelatedFilter and other django-rest-framework-filters features are designed to work with rest_framework_filters.FilterSet and will not function on a django_filters.FilterSet. However, the target RelatedFilter.filterset may point to a FilterSet from either package, and both FilterSet implementations are compatible with the other's DRF backend.

# valid
class VanillaFilter(django_filters.FilterSet):
    ...

class DRFFilter(rest_framework_filters.FilterSet):
    vanilla = rest_framework_filters.RelatedFilter(filterset=VanillaFilter, queryset=...)


# invalid
class DRFFilter(rest_framework_filters.FilterSet):
    ...

class VanillaFilter(django_filters.FilterSet):
    drf = rest_framework_filters.RelatedFilter(filterset=DRFFilter, queryset=...)

Caveats & Limitations

MultiWidget is incompatible

djangorestframework-filters is not compatible with form widgets that parse query names that differ from the filter's attribute name. Although this only practically applies to MultiWidget, it is a general limitation that affects custom widgets that also have this behavior. Affected filters include RangeFilter, DateTimeFromToRangeFilter, DateFromToRangeFilter, TimeRangeFilter, and NumericRangeFilter.

To demonstrate the incompatiblity, take the following filterset:

class PostFilter(FilterSet):
    publish_date = filters.DateFromToRangeFilter()

The above filter allows users to perform a range query on the publication date. The filter class internally uses MultiWidget to separately parse the upper and lower bound values. The incompatibility lies in that MultiWidget appends an index to its inner widget names. Instead of parsing publish_date, it expects publish_date_0 and publish_date_1. It is possible to fix this by including the attribute name in the querystring, although this is not recommended.

?publish_date_0=2016-01-01&publish_date_1=2016-02-01&publish_date=

MultiWidget is also discouraged since:

  • core-api field introspection fails for similar reasons
  • _0 and _1 are less API-friendly than _min and _max

The recommended solutions are to either:

  • Create separate filters for each of the sub-widgets (such as publish_date_min and publish_date_max).
  • Use a CSV-based filter such as those derived from BaseCSVFilter/BaseInFilter/BaseRangeFilter. eg,
?publish_date__range=2016-01-01,2016-02-01

Complex Operations

The ComplexFilterBackend defines a custom querystring syntax and encoding process that enables the expression of complex queries. This syntax extends standard querystrings with the ability to define multiple sets of parameters and operators for how the queries should be combined.


! Note that this feature is experimental. Bugs may be encountered, and the backend is subject to change.


To understand the backend more fully, consider a query to find all articles that contain titles starting with either "Who" or "What". The underlying query could be represented with the following:

q1 = Article.objects.filter(title__startswith='Who')
q2 = Article.objects.filter(title__startswith='What')
return q1 | q2

Now consider the query, but modified with upper and lower date bounds:

q1 = Article.objects.filter(title__startswith='Who').filter(publish_date__lte='2005-01-01')
q2 = Article.objects.filter(title__startswith='What').filter(publish_date__gte='2010-01-01')
return q1 | q2

Using just a FilterSet, it is certainly feasible to represent the former query by writing a custom filter class. However, it is less feasible with the latter query, where multiple sets of varying data types and lookups need to be validated. In contrast, the ComplexFilterBackend can create this complex query through the arbitrary combination of a simple filter. To support the above, the querystring needs to be created with minimal changes. Unencoded example:

(title__startswith=Who&publish_date__lte=2005-01-01) | (title__startswith=What&publish_date__gte=2010-01-01)

By default, the backend combines queries with both & (AND) and | (OR), and supports unary negation ~. E.g.,

(param1=value1) & (param2=value2) | ~(param3=value3)

The backend supports both standard and complex queries. To perform complex queries, the query must be encoded and set as the value of the complex_filter_param (defaults to filters). To perform standard queries, use the backend in the same manner as the RestFrameworkFilterBackend.

Configuring ComplexFilterBackend

Similar to other backends, ComplexFilterBackend must be added to a view's filter_backends atribute. Either add it to the DEFAULT_FILTER_BACKENDS setting, or set it as a backend on the view class.

REST_FRAMEWORK = {
    'DEFAULT_FILTER_BACKENDS': (
        'rest_framework_filters.backends.ComplexFilterBackend',
    ),
}

# or

class MyViewSet(generics.ListAPIView):
    filter_backends = (rest_framework_filters.backends.ComplexFilterBackend, )
    ...

You may customize how queries are combined by subclassing ComplexFilterBackend and overriding the operators attribute. operators is a map of operator symbols to functions that combine two querysets. For example, the map can be overridden to use the QuerySet.intersection() and QuerySet.union() instead of & and |.

class CustomizedBackend(ComplexFilterBackend):
    operators = {
        '&': QuerySet.intersection,
        '|': QuerySet.union,
        '-': QuerySet.difference,
    }

Unary negation relies on ORM internals and may be buggy in certain circumstances. If there are issues with this feature, it can be disabled by setting the negation attribute to False on the backend class. If you do experience bugs, please open an issue on the bug tracker.

Complex querystring encoding

Below is the procedure for encoding a complex query:

  • Convert the query paramaters into individual querystrings.
  • URL-encode the individual querystrings.
  • Wrap the encoded strings in parentheses, and join with operators.
  • URL-encode the entire querystring.
  • Set as the value to the complex filter param (default: filters).

Using the first example, these steps can be visualized as so:

  • title__startswith=Who, title__startswith=What
  • title__startswith%3DWho, title__startswith%3DWhat
  • (title__startswith%3DWho) | (title__startswith%3DWhat)
  • %28title__startswith%253DWho%29%20%7C%20%28title__startswith%253DWhat%29
  • filters=%28title__startswith%253DWho%29%20%7C%20%28title__startswith%253DWhat%29

Error handling

ComplexFilterBackend will raise any decoding errors under the complex filtering parameter name. For example,

{
    "filters": [
        "Invalid querystring operator. Matched: 'foo'."
    ]
}

When filtering the querysets, filterset validation errors will be collected and raised under the complex filtering parameter name, then under the filterset's decoded querystring. For a complex query like (a=1&b=2) | (c=3&d=4), errors would be raised like so:

{
    "filters": {
        "a=1&b=2": {
            "a": ["..."]
        },
        "c=3&d=4": {
            "c": ["..."]
        }
    }
{

Migrating to 1.0

Backend renamed, provides new templates

The backend has been renamed from DjangoFilterBackend to RestFrameworkFilterBackend and now uses its own template paths, located under rest_framework_filters instead of django_filters/rest_framework.

To load the included templates, it is necessary to add rest_framework_filters to the INSTALLED_APPS setting.

RelatedFilter.queryset now required

The related filterset's model is no longer used to provide the default value for RelatedFilter.queryset. This change reduces the chance of unintentionally exposing data in the rendered filter forms. You must now explicitly provide the queryset argument, or override the get_queryset() method (see queryset callables).

get_filters() renamed to get_request_filters()

django-filter has add a get_filters() classmethod to it's API, so this method has been renamed.

Publishing

$ pip install -U twine setuptools wheel
$ rm -rf dist/ build/
$ python setup.py sdist bdist_wheel
$ twine upload dist/*

License

Copyright (c) 2013-2015 Philip Neustrom <philipn@gmail.com>, 2016-2017 Ryan P Kilby <rpkilby@ncsu.edu>

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.