django rest framework - how do you flatten nested data?

Here's an approach based on James's answer but for a newer version of Django Rest Framework and support for reading and writing (update of the nested field only, it should be easy enough to add creation, see DRF's documentation for that.)

class ProfileSerializer(serializers.ModelSerializer):
    class Meta:
        model = Profile
        fields = ('phone', 'some', 'other', 'fields')


class UserDetailsSerializer(serializers.ModelSerializer):
    """User model with Profile. Handled as a single object, profile is flattened."""
    profile = ProfileSerializer()

    class Meta:
        model = User
        fields = ('username', 'email', 'profile')
        read_only_fields = ('email', )

    def to_representation(self, obj):
        """Move fields from profile to user representation."""
        representation = super().to_representation(obj)
        profile_representation = representation.pop('profile')
        for key in profile_representation:
            representation[key] = profile_representation[key]

        return representation

    def to_internal_value(self, data):
        """Move fields related to profile to their own profile dictionary."""
        profile_internal = {}
        for key in ProfileSerializer.Meta.fields:
            if key in data:
                profile_internal[key] = data.pop(key)

        internal = super().to_internal_value(data)
        internal['profile'] = profile_internal
        return internal

    def update(self, instance, validated_data):
        """Update user and profile. Assumes there is a profile for every user."""
        profile_data = validated_data.pop('profile')
        super().update(instance, validated_data)

        profile = instance.profile
        for attr, value in profile_data.items():
            setattr(profile, attr, value)
        profile.save()

        return instance

James' answer is what I finally used. As I had several serializers using this method, I converted it to a mixin:

class FlattenMixin(object):
    """Flatens the specified related objects in this representation"""
    def to_representation(self, obj):
        assert hasattr(self.Meta, 'flatten'), (
            'Class {serializer_class} missing "Meta.flatten" attribute'.format(
                serializer_class=self.__class__.__name__
            )
        )
        # Get the current object representation
        rep = super(FlattenMixin, self).to_representation(obj)
        # Iterate the specified related objects with their serializer
        for field, serializer_class in self.Meta.flatten:
            serializer = serializer_class(context = self.context)
            objrep = serializer.to_representation(getattr(obj, field))
            #Include their fields, prefixed, in the current   representation
            for key in objrep:
                rep[field + "__" + key] = objrep[key]
        return rep

This way, you can do something like:

class MembershipSerializer(FlattenMixin, serializers.HyperlinkedModelSerializer):
    class Meta:
        model = Membership
        fields = ('id', 'url', 'group', 'date_joined', 'invite_reason')
        flatten = [ ('person', PersonSerializer) ]

I didn't try it with HyperlinkedModelSerializer but with ModelSerializer you can make custom serializer class that supports flatten option.

class CustomModelSerializer(serializers.ModelSerializer):
    def __init__(self, *args, **kwargs):
        self.flatten = kwargs.pop('flatten', False)
        super(CustomModelSerializer, self).__init__(*args, **kwargs)

    def get_fields(self):
        fields = super(CustomModelSerializer, self).get_fields()
        for field_name, field in fields.items():
            if getattr(field, 'flatten', False):
                del fields[field_name]
                for nested_field_name, nested_field in field.fields.iteritems():
                    nested_field.source = (field_name + '.' +
                                           (nested_field.source or nested_field_name))
                    fields[nested_field_name] = nested_field
        return fields

Usage:

class MembershipSerializer(CustomModelSerializer):
    person = PersonSerializer(flatten=True)

    class Meta:
        model = Membership
        fields = ('person', ...)


class PersonSerializer(CustomModelSerializer):

    class Meta:
        model = Person
        fields = (...)

Here's another solution to add the fields dynamically thanks to meta classes!

class FlattenNestedFields(SerializerMetaclass):
    def __new__(cls, name, bases, attrs):
        klass = super().__new__(cls, name, bases, attrs)
        for related_name, fields in attrs['related_fields']:
            klass.Meta.fields += fields
            for field in fields:
                field_rhs = ReadOnlyField(source=f'{related_name}.{field}')
                klass._declared_fields[field] = field_rhs
                setattr(klass, field, field_rhs)
        return klass

Now, when defining your serializer, all you need to do is this:

class MembershipSerializer(serializers.HyperlinkedModelSerializer, metaclass=FlattenNestedFields):
    related_fields = [
        ('person', ['first_name', 'last_name', 'favourite_food']),
    ]

    class Meta:
        model = Membership
        fields = ('id', 'url', 'group', 'date_joined', 'invite_reason',)

and the attributes (as well as the Meta inner class fields) are generated by the metaclass.

combining ekuusela's answer and this example from the DRF documentatation, you can also control which fields (from the nested object) you want to display.
Your serializer would look like this

class UserDetailsSerializer(serializers.ModelSerializer):
    """User model with Profile. Handled as a single object, profile is flattened."""

    profile = ProfileSerializer()

    def __init__(self, *args, **kwargs):
        self.allow_fields = kwargs.pop('fields', None)
        super(ProfileSerializer, self).__init__(*args, **kwargs)

    class Meta:
        model = User
        fields = ('username', 'email', 'profile')

    def to_representation(self, instance):
        representation = super().to_representation(instance)
        profile_representation = representation.pop('profile')
        representation.update(profile_representation)

        if self.allow_fields is not None:
            # Drop any fields that are not specified in the `fields` argument.
            allowed = set(self.allow_fields)
            existing = set(representation)
            for field_name in existing - allowed:
                representation.pop(field_name)
        return representation

And you would instantiate your Serializer as if it was only a singe Model

serializer = UserDetailsSerializer(user, fields=('username', 'email','profile_field1', 'profile_field2'))