系統講一下多對多關係：

第一個例子是官網的不需要定義附加表的多對多關係，其實也需要關係表，只不過關係表很簡單，只有兩列，即源物件id（from_*_id）和目標物件id（to_*_id），由系統自動新增：

操作的例子，官網介紹似乎很長，但仔細閱讀發現內容不多，都是程式碼例子。總結起來一共幾點：

1、為物件新增多對多關係前，需要保持物件，即寫入資料庫，產生id列，然後才能利用id列建立多對多關係表。

2、通過add（）函式新增多對多關係，如a1.publications.add(p1)，其中publications定義是：

publications = models.ManyToManyField(Publication)

3、多次新增同一個關係不會重複。add（）的引數的型別不能錯。

4、可以用create（）函式，連建立to_*_id代表的目標物件，帶add（）新建物件到多對多關係表一起，一步完成兩個操作。

5、可以從源物件查詢到目標物件，也可以反向查詢。

6、可用filter查詢，並支援反向查詢

7、.distinct().count()組合使用(不懂原文說的as well是指哪裡也這樣)

8、可以用exclude查詢

9、delete刪除一側物件，則關係也隨之刪除。

10、從目標側建立多對多關係。

11、remove（）只刪除關係，保留源側和目標側物件。

12、set（）函式批量設關係

13、clear（）函式清除所有關係

等等，詳見原文。

Many-to-many relationships¶

To define a many-to-many relationship, use ManyToManyField.

In this example, an Article can be published in multiple Publication objects, and a Publication has multiple Article objects:

from django.db import models

class Publication(models.Model):
    title = models.CharField(max_length=30)

    def __str__(self):
        return self.title

    class Meta:
        ordering = ('title',)

class Article(models.Model):
    headline = models.CharField(max_length=100)
    publications = models.ManyToManyField(Publication)

    def __str__(self):
        return self.headline

    class Meta:
        ordering = ('headline',)

What follows are examples of operations that can be performed using the Python API facilities. Note that if you are using an intermediate model for a many-to-many relationship, some of the related manager's methods are disabled, so some of these examples won't work with such models.

Create a few Publications:

>>> p1 = Publication(title='The Python Journal')
>>> p1.save()
>>> p2 = Publication(title='Science News')
>>> p2.save()
>>> p3 = Publication(title='Science Weekly')
>>> p3.save()

Create an Article:

>>> a1 = Article(headline='Django lets you build Web apps easily')

You can't associate it with a Publication until it's been saved:

>>> a1.publications.add(p1)
Traceback (most recent call last):
...
ValueError: 'Article' instance needs to have a primary key value before a many-to-many relationship can be used.

Save it!

>>> a1.save()

Associate the Article with a Publication:

>>> a1.publications.add(p1)

Create another Article, and set it to appear in the Publications:

>>> a2 = Article(headline='NASA uses Python')
>>> a2.save()
>>> a2.publications.add(p1, p2)
>>> a2.publications.add(p3)

Adding a second time is OK, it will not duplicate the relation:

>>> a2.publications.add(p3)

Adding an object of the wrong type raises TypeError:

>>> a2.publications.add(a1)
Traceback (most recent call last):
...
TypeError: 'Publication' instance expected

Create and add a Publication to an Article in one step using create():

>>> new_publication = a2.publications.create(title='Highlights for Children')

Article objects have access to their related Publication objects:

>>> a1.publications.all()
<QuerySet [<Publication: The Python Journal>]>
>>> a2.publications.all()
<QuerySet [<Publication: Highlights for Children>, <Publication: Science News>, <Publication: Science Weekly>, <Publication: The Python Journal>]>

Publication objects have access to their related Article objects:

>>> p2.article_set.all()
<QuerySet [<Article: NASA uses Python>]>
>>> p1.article_set.all()
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>
>>> Publication.objects.get(id=4).article_set.all()
<QuerySet [<Article: NASA uses Python>]>

Many-to-many relationships can be queried using lookups across relationships:

>>> Article.objects.filter(publications__id=1)
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>
>>> Article.objects.filter(publications__pk=1)
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>
>>> Article.objects.filter(publications=1)
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>
>>> Article.objects.filter(publications=p1)
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>

>>> Article.objects.filter(publications__title__startswith="Science")
<QuerySet [<Article: NASA uses Python>, <Article: NASA uses Python>]>

>>> Article.objects.filter(publications__title__startswith="Science").distinct()
<QuerySet [<Article: NASA uses Python>]>

The count() function respects distinct() as well:

>>> Article.objects.filter(publications__title__startswith="Science").count()
2

>>> Article.objects.filter(publications__title__startswith="Science").distinct().count()
1

>>> Article.objects.filter(publications__in=[1,2]).distinct()
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>
>>> Article.objects.filter(publications__in=[p1,p2]).distinct()
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA uses Python>]>

Reverse m2m queries are supported (i.e., starting at the table that doesn't have a ManyToManyField):

>>> Publication.objects.filter(id=1)
<QuerySet [<Publication: The Python Journal>]>
>>> Publication.objects.filter(pk=1)
<QuerySet [<Publication: The Python Journal>]>

>>> Publication.objects.filter(article__headline__startswith="NASA")
<QuerySet [<Publication: Highlights for Children>, <Publication: Science News>, <Publication: Science Weekly>, <Publication: The Python Journal>]>

>>> Publication.objects.filter(article__id=1)
<QuerySet [<Publication: The Python Journal>]>
>>> Publication.objects.filter(article__pk=1)
<QuerySet [<Publication: The Python Journal>]>
>>> Publication.objects.filter(article=1)
<QuerySet [<Publication: The Python Journal>]>
>>> Publication.objects.filter(article=a1)
<QuerySet [<Publication: The Python Journal>]>

>>> Publication.objects.filter(article__in=[1,2]).distinct()
<QuerySet [<Publication: Highlights for Children>, <Publication: Science News>, <Publication: Science Weekly>, <Publication: The Python Journal>]>
>>> Publication.objects.filter(article__in=[a1,a2]).distinct()
<QuerySet [<Publication: Highlights for Children>, <Publication: Science News>, <Publication: Science Weekly>, <Publication: The Python Journal>]>

Excluding a related item works as you would expect, too (although the SQL involved is a little complex):

>>> Article.objects.exclude(publications=p2)
<QuerySet [<Article: Django lets you build Web apps easily>]>

If we delete a Publication, its Articles won't be able to access it:

>>> p1.delete()
>>> Publication.objects.all()
<QuerySet [<Publication: Highlights for Children>, <Publication: Science News>, <Publication: Science Weekly>]>
>>> a1 = Article.objects.get(pk=1)
>>> a1.publications.all()
<QuerySet []>

If we delete an Article, its Publications won't be able to access it:

>>> a2.delete()
>>> Article.objects.all()
<QuerySet [<Article: Django lets you build Web apps easily>]>
>>> p2.article_set.all()
<QuerySet []>

Adding via the 'other' end of an m2m:

>>> a4 = Article(headline='NASA finds intelligent life on Earth')
>>> a4.save()
>>> p2.article_set.add(a4)
>>> p2.article_set.all()
<QuerySet [<Article: NASA finds intelligent life on Earth>]>
>>> a4.publications.all()
<QuerySet [<Publication: Science News>]>

Adding via the other end using keywords:

>>> new_article = p2.article_set.create(headline='Oxygen-free diet works wonders')
>>> p2.article_set.all()
<QuerySet [<Article: NASA finds intelligent life on Earth>, <Article: Oxygen-free diet works wonders>]>
>>> a5 = p2.article_set.all()[1]
>>> a5.publications.all()
<QuerySet [<Publication: Science News>]>

Removing Publication from an Article:

>>> a4.publications.remove(p2)
>>> p2.article_set.all()
<QuerySet [<Article: Oxygen-free diet works wonders>]>
>>> a4.publications.all()
<QuerySet []>

And from the other end:

>>> p2.article_set.remove(a5)
>>> p2.article_set.all()
<QuerySet []>
>>> a5.publications.all()
<QuerySet []>

Relation sets can be set:

>>> a4.publications.all()
<QuerySet [<Publication: Science News>]>
>>> a4.publications.set([p3])
>>> a4.publications.all()
<QuerySet [<Publication: Science Weekly>]>

Relation sets can be cleared:

>>> p2.article_set.clear()
>>> p2.article_set.all()
<QuerySet []>

And you can clear from the other end:

>>> p2.article_set.add(a4, a5)
>>> p2.article_set.all()
<QuerySet [<Article: NASA finds intelligent life on Earth>, <Article: Oxygen-free diet works wonders>]>
>>> a4.publications.all()
<QuerySet [<Publication: Science News>, <Publication: Science Weekly>]>
>>> a4.publications.clear()
>>> a4.publications.all()
<QuerySet []>
>>> p2.article_set.all()
<QuerySet [<Article: Oxygen-free diet works wonders>]>

Recreate the Article and Publication we have deleted:

>>> p1 = Publication(title='The Python Journal')
>>> p1.save()
>>> a2 = Article(headline='NASA uses Python')
>>> a2.save()
>>> a2.publications.add(p1, p2, p3)

Bulk delete some Publications - references to deleted publications should go:

>>> Publication.objects.filter(title__startswith='Science').delete()
>>> Publication.objects.all()
<QuerySet [<Publication: Highlights for Children>, <Publication: The Python Journal>]>
>>> Article.objects.all()
<QuerySet [<Article: Django lets you build Web apps easily>, <Article: NASA finds intelligent life on Earth>, <Article: NASA uses Python>, <Article: Oxygen-free diet works wonders>]>
>>> a2.publications.all()
<QuerySet [<Publication: The Python Journal>]>

Bulk delete some articles - references to deleted objects should go:

>>> q = Article.objects.filter(headline__startswith='Django')
>>> print(q)
<QuerySet [<Article: Django lets you build Web apps easily>]>
>>> q.delete()

After the delete(), the QuerySet cache needs to be cleared, and the referenced objects should be gone:

>>> print(q)
<QuerySet []>
>>> p1.article_set.all()
<QuerySet [<Article: NASA uses Python>]>

除了簡單的多對多關係，還有可自定義欄位的多對多關係表：

其實相當於自定義一個class，有兩個外來鍵，分別是源物件和目標物件，其餘欄位可以自定義，如：

class Membership(models.Model):

     person = models.ForeignKey(Person, on_delete=models.CASCADE)

     group = models.ForeignKey(Group, on_delete=models.CASCADE)

     date_joined = models.DateField()

     invite_reason = models.CharField(max_length=64)

注意的是這種自定義的關係表不能用add(), create(), or set()來建立關係，因為自定義的多對多關係還有其他欄位需要補充。

這種多對多關係的建立和其他物件建立一樣，用普通方式：

Membership.objects.create(）

Extra fields on many-to-many relationships¶

When you’re only dealing with simple many-to-many relationships such as mixing and matching pizzas and toppings, a standard ManyToManyField is all you need. However, sometimes you may need to associate data with the relationship between two models.

For example, consider the case of an application tracking the musical groups which musicians belong to. There is a many-to-many relationship between a person and the groups of which they are a member, so you could use a ManyToManyField to represent this relationship. However, there is a lot of detail about the membership that you might want to collect, such as the date at which the person joined the group.

For these situations, Django allows you to specify the model that will be used to govern the many-to-many relationship. You can then put extra fields on the intermediate model. The intermediate model is associated with the ManyToManyField using the through argument to point to the model that will act as an intermediary. For our musician example, the code would look something like this:

from django.db import models

class Person(models.Model):
    name = models.CharField(max_length=128)

    def __str__(self):
        return self.name

class Group(models.Model):
    name = models.CharField(max_length=128)
    members = models.ManyToManyField(Person, through='Membership')

    def __str__(self):
        return self.name

class Membership(models.Model):
    person = models.ForeignKey(Person, on_delete=models.CASCADE)
    group = models.ForeignKey(Group, on_delete=models.CASCADE)
    date_joined = models.DateField()
    invite_reason = models.CharField(max_length=64)

When you set up the intermediary model, you explicitly specify foreign keys to the models that are involved in the many-to-many relationship. This explicit declaration defines how the two models are related.

There are a few restrictions on the intermediate model:

• Your intermediate model must contain one - and only one - foreign key to the source model (this would be Group in our example), or you must explicitly specify the foreign keys Django should use for the relationship using ManyToManyField.through_fields. If you have more than one foreign key and through_fields is not specified, a validation error will be raised. A similar restriction applies to the foreign key to the target model (this would be Person in our example).
• For a model which has a many-to-many relationship to itself through an intermediary model, two foreign keys to the same model are permitted, but they will be treated as the two (different) sides of the many-to-many relationship. If there are more than two foreign keys though, you must also specify through_fields as above, or a validation error will be raised.
• When defining a many-to-many relationship from a model to itself, using an intermediary model, you must use symmetrical=False (see the model field reference).

Now that you have set up your ManyToManyField to use your intermediary model (Membership, in this case), you’re ready to start creating some many-to-many relationships. You do this by creating instances of the intermediate model:

>>> ringo = Person.objects.create(name="Ringo Starr")
>>> paul = Person.objects.create(name="Paul McCartney")
>>> beatles = Group.objects.create(name="The Beatles")
>>> m1 = Membership(person=ringo, group=beatles,
...     date_joined=date(1962, 8, 16),
...     invite_reason="Needed a new drummer.")
>>> m1.save()
>>> beatles.members.all()
<QuerySet [<Person: Ringo Starr>]>
>>> ringo.group_set.all()
<QuerySet [<Group: The Beatles>]>
>>> m2 = Membership.objects.create(person=paul, group=beatles,
...     date_joined=date(1960, 8, 1),
...     invite_reason="Wanted to form a band.")
>>> beatles.members.all()
<QuerySet [<Person: Ringo Starr>, <Person: Paul McCartney>]>

Unlike normal many-to-many fields, you can’t use add(), create(), or set() to create relationships:

>>> # The following statements will not work
>>> beatles.members.add(john)
>>> beatles.members.create(name="George Harrison")
>>> beatles.members.set([john, paul, ringo, george])

Why? You can’t just create a relationship between a Person and a Group - you need to specify all the detail for the relationship required by the Membership model. The simple add, create and assignment calls don’t provide a way to specify this extra detail. As a result, they are disabled for many-to-many relationships that use an intermediate model. The only way to create this type of relationship is to create instances of the intermediate model.

The remove() method is disabled for similar reasons. For example, if the custom through table defined by the intermediate model does not enforce uniqueness on the (model1, model2) pair, a remove() call would not provide enough information as to which intermediate model instance should be deleted:

>>> Membership.objects.create(person=ringo, group=beatles,
...     date_joined=date(1968, 9, 4),
...     invite_reason="You've been gone for a month and we miss you.")
>>> beatles.members.all()
<QuerySet [<Person: Ringo Starr>, <Person: Paul McCartney>, <Person: Ringo Starr>]>
>>> # This will not work because it cannot tell which membership to remove
>>> beatles.members.remove(ringo)

However, the clear() method can be used to remove all many-to-many relationships for an instance:

>>> # Beatles have broken up
>>> beatles.members.clear()
>>> # Note that this deletes the intermediate model instances
>>> Membership.objects.all()
<QuerySet []>

Once you have established the many-to-many relationships by creating instances of your intermediate model, you can issue queries. Just as with normal many-to-many relationships, you can query using the attributes of the many-to-many-related model:

# Find all the groups with a member whose name starts with 'Paul'
>>> Group.objects.filter(members__name__startswith='Paul')
<QuerySet [<Group: The Beatles>]>

As you are using an intermediate model, you can also query on its attributes:

# Find all the members of the Beatles that joined after 1 Jan 1961
>>> Person.objects.filter(
...     group__name='The Beatles',
...     membership__date_joined__gt=date(1961,1,1))
<QuerySet [<Person: Ringo Starr]>

If you need to access a membership’s information you may do so by directly querying the Membership model:

>>> ringos_membership = Membership.objects.get(group=beatles, person=ringo)
>>> ringos_membership.date_joined
datetime.date(1962, 8, 16)
>>> ringos_membership.invite_reason
'Needed a new drummer.'

Another way to access the same information is by querying the many-to-many reverse relationship from a Person object:

>>> ringos_membership = ringo.membership_set.get(group=beatles)
>>> ringos_membership.date_joined
datetime.date(1962, 8, 16)
>>> ringos_membership.invite_reason
'Needed a new drummer.'