您可能会发现这个mixin很有用，它覆盖了get_serializer_class方法，并允许您声明一个dict，该dict将动作和序列化器类或回滚映射到通常的行为。

class MultiSerializerViewSetMixin(object):
    def get_serializer_class(self):
        """
        Look for serializer class in self.serializer_action_classes, which
        should be a dict mapping action name (key) to serializer class (value),
        i.e.:

        class MyViewSet(MultiSerializerViewSetMixin, ViewSet):
            serializer_class = MyDefaultSerializer
            serializer_action_classes = {
               'list': MyListSerializer,
               'my_action': MyActionSerializer,
            }

            @action
            def my_action:
                ...

        If there's no entry for that action then just fallback to the regular
        get_serializer_class lookup: self.serializer_class, DefaultSerializer.

        """
        try:
            return self.serializer_action_classes[self.action]
        except (KeyError, AttributeError):
            return super(MultiSerializerViewSetMixin, self).get_serializer_class()

只是想添加到现有的解决方案。如果你想为你的视图集的额外动作(即使用@action装饰器)使用不同的序列化器，你可以像这样在装饰器中添加kwargs:

@action(methods=['POST'], serializer_class=YourSpecialSerializer)
def your_extra_action(self, request):
    serializer = self.get_serializer(data=request.data)
    ...

对于所有提到的其他解决方案，我无法找到如何使用get_serializer_class函数实例化类，也无法找到自定义验证函数。对于那些像我一样仍然迷失并希望完全实现的人，请检查下面的答案。

views.py

from rest_framework.response import Response

from project.models import Project
from project.serializers import ProjectCreateSerializer, ProjectIDGeneratorSerializer


class ProjectViewSet(viewsets.ModelViewSet):
    action_serializers = {
        'generate_id': ProjectIDGeneratorSerializer,
        'create': ProjectCreateSerializer,
    }
    permission_classes = [IsAuthenticated]

    def get_serializer_class(self):
        if hasattr(self, 'action_serializers'):
            return self.action_serializers.get(self.action, self.serializer_class)

        return super(ProjectViewSet, self).get_serializer_class()

    # You can create custom function
    def generate_id(self, request):
        serializer = self.get_serializer_class()(data=request.GET)
        serializer.context['user'] = request.user
        serializer.is_valid(raise_exception=True)
        return Response(serializer.validated_data, status=status.HTTP_200_OK)

    def create(self, request, **kwargs):
        serializer = self.get_serializer_class()(data=request.data)
        serializer.context['user'] = request.user
        serializer.is_valid(raise_exception=True)
        return Response(serializer.validated_data, status=status.HTTP_200_OK)

serializers.py

import random
from rest_framework import serializers
from project.models import Project


class ProjectIDGeneratorSerializer(serializers.Serializer):
    def update(self, instance, validated_data):
        pass

    def create(self, validated_data):
        pass

    projectName = serializers.CharField(write_only=True)

    class Meta:
        fields = ['projectName']

    def validate(self, attrs):
        project_name = attrs.get('projectName')
        project_id = project_name.replace(' ', '-')

        return {'projectID': project_id}


class ProjectCreateSerializer(serializers.Serializer):
    def update(self, instance, validated_data):
        pass

    def create(self, validated_data):
        pass

    projectName = serializers.CharField(write_only=True)
    projectID = serializers.CharField(write_only=True)

    class Meta:
        model = Project
        fields = ['projectName', 'projectID']

    def to_representation(self, instance: Project):
        data = dict()
        data['projectName'] = instance.name
        data['projectID'] = instance.projectID
        data['createdAt'] = instance.createdAt
        data['updatedAt'] = instance.updatedAt

        representation = {
            'message': f'Project {instance.name} has been created.',
        }

        return representation

    def validate(self, attrs):
        print('attrs', dict(attrs))
        project_name = attrs.get('projectName')
        project_id = attrs.get('projectID')

        if Project.objects.filter(projectID=project_id).first():
            raise serializers.ValidationError(f'Project with ID {project_id} already exist')

        project = Project.objects.create(projectID=project_id,
                                         name=project_name)

        print('user', self.context['user'])
        project.user.add(self.context["user"])

        project.save()

        return self.to_representation(project)

urls . py

from django.urls import path

from .views import ProjectViewSet

urlpatterns = [
    path('project/generateID', ProjectViewSet.as_view({'get': 'generate_id'})),
    path('project/create', ProjectViewSet.as_view({'post': 'create'})),
]

models.py

# Create your models here.
from django.db import models

from authentication.models import User


class Project(models.Model):
    id = models.AutoField(primary_key=True)
    projectID = models.CharField(max_length=255, blank=False, db_index=True, null=False)
    user = models.ManyToManyField(User)
    name = models.CharField(max_length=255, blank=False)
    createdAt = models.DateTimeField(auto_now_add=True)
    updatedAt = models.DateTimeField(auto_now=True)

    def __str__(self):
        return self.name

尽管以某种方式预先定义多个序列化器似乎是最明显的文档化方法，但FWIW还有另一种方法，它利用其他文档化的代码，并允许在实例化序列化器时将参数传递给它。我认为，如果需要基于各种因素生成逻辑，比如用户管理级别、被调用的操作，甚至实例的属性，那么可能更有价值。

这个难题的第一部分是关于在实例化时动态修改序列化器的文档。该文档没有解释如何从视图集调用这段代码，也没有解释如何在字段被初始化后修改字段的只读状态——但这并不难。

第二部分——get_serializer方法也被记录了——(在get_serializer_class的“其他方法”下面)，因此它应该是安全的(源代码非常简单，希望这意味着由于修改而导致的意外副作用的可能性更小)。检查GenericAPIView下的源代码(ModelViewSet和所有其他内建的视图集类似乎都继承自GenericAPIView，它定义了get_serializer。

把两者放在一起，你可以这样做:

在一个序列化器文件中(对我来说是base_serialzers .py):

class DynamicFieldsModelSerializer(serializers.ModelSerializer):
"""
A ModelSerializer that takes an additional `fields` argument that
controls which fields should be displayed.
"""

def __init__(self, *args, **kwargs):
    # Don't pass the 'fields' arg up to the superclass
    fields = kwargs.pop('fields', None)

    # Adding this next line to the documented example
    read_only_fields = kwargs.pop('read_only_fields', None)

    # Instantiate the superclass normally
    super(DynamicFieldsModelSerializer, self).__init__(*args, **kwargs)

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

    # another bit we're adding to documented example, to take care of readonly fields 
    if read_only_fields is not None:
        for f in read_only_fields:
            try:
                self.fields[f].read_only = True
            exceptKeyError:
                #not in fields anyway
                pass

然后在你的视图集中，你可以这样做:

class MyViewSet(viewsets.ModelViewSet):
    # ...permissions and all that stuff

    def get_serializer(self, *args, **kwargs):

        # the next line is taken from the source
        kwargs['context'] = self.get_serializer_context()

        # ... then whatever logic you want for this class e.g:
        if self.action == "list":
            rofs = ('field_a', 'field_b')
            fs = ('field_a', 'field_c')
        if self.action == “retrieve”:
            rofs = ('field_a', 'field_c’, ‘field_d’)
            fs = ('field_a', 'field_b’)
        #  add all your further elses, elifs, drawing on info re the actions, 
        # the user, the instance, anything passed to the method to define your read only fields and fields ...
        #  and finally instantiate the specific class you want (or you could just
        # use get_serializer_class if you've defined it).  
        # Either way the class you're instantiating should inherit from your DynamicFieldsModelSerializer
        kwargs['read_only_fields'] = rofs
        kwargs['fields'] = fs
        return MyDynamicSerializer(*args, **kwargs)

应该就是这样了!使用MyViewSet现在应该用你想要的参数实例化你的MyDynamicSerializer——并且假设你的序列化器继承了你的DynamicFieldsModelSerializer，它应该知道该做什么。

Perhaps its worth mentioning that it can makes special sense if you want to adapt the serializer in some other ways …e.g. to do things like take in a read_only_exceptions list and use it to whitelist rather than blacklist fields (which I tend to do). I also find it useful to set the fields to an empty tuple if its not passed and then just remove the check for None ... and I set my fields definitions on my inheriting Serializers to 'all'. This means no fields that aren't passed when instantiating the serializer survive by accident and I also don't have to compare the serializer invocation with the inheriting serializer class definition to know what's been included...e.g within the init of the DynamicFieldsModelSerializer:

# ....
fields = kwargs.pop('fields', ())
# ...
allowed = set(fields)
existing = set(self.fields)
for field_name in existing - allowed:
self.fields.pop(field_name)
# ....

注意:如果我只是想要两个或三个类映射到不同的操作和/或我不想要任何特别的动态序列化器行为，我可能会使用其他人在这里提到的方法之一，但我认为这值得作为一种替代方案，特别是考虑到它的其他用途。

基于@gonz和@user2734679 answers，我创建了这个小的python包，它以ModelViewset的子类的形式提供了这个功能。下面是它的工作原理。

from drf_custom_viewsets.viewsets.CustomSerializerViewSet
from myapp.serializers import DefaltSerializer, CustomSerializer1, CustomSerializer2

class MyViewSet(CustomSerializerViewSet):
    serializer_class = DefaultSerializer
    custom_serializer_classes = {
        'create':  CustomSerializer1,
        'update': CustomSerializer2,
    }

您可能会发现这个mixin很有用，它覆盖了get_serializer_class方法，并允许您声明一个dict，该dict将动作和序列化器类或回滚映射到通常的行为。

class MultiSerializerViewSetMixin(object):
    def get_serializer_class(self):
        """
        Look for serializer class in self.serializer_action_classes, which
        should be a dict mapping action name (key) to serializer class (value),
        i.e.:

        class MyViewSet(MultiSerializerViewSetMixin, ViewSet):
            serializer_class = MyDefaultSerializer
            serializer_action_classes = {
               'list': MyListSerializer,
               'my_action': MyActionSerializer,
            }

            @action
            def my_action:
                ...

        If there's no entry for that action then just fallback to the regular
        get_serializer_class lookup: self.serializer_class, DefaultSerializer.

        """
        try:
            return self.serializer_action_classes[self.action]
        except (KeyError, AttributeError):
            return super(MultiSerializerViewSetMixin, self).get_serializer_class()