This package provides a field type for Django models that allows joins to a related model without a foreign key.
Install via pip:
pip install django-joinfield
joinfield in INSTALLED_APPS in your settings file. Then you can import
JoinField and use it when defining your models:
from joinfield.joinfield import JoinField class Parent(models.Model): column = models.CharField(max_length=64) class Child(models.Model): parent = JoinField(Parent, to_field='column')
The database column created in the child table will have the type defined by
to_field parent column. In this case, a CharField. It will not be a
foreign key and there will be no database constraints between these
Now you can do joins between child and parent using the orm:
parent = Parent.objects.first() children = Child.objects.filter(parent=parent) # Or children = Child.objects.filter(parent__column='some value')
Let's imagine a database of people and family information. One table might contain general family information keyed by Surname:
class Surname(models.Model): name = models.CharField(max_length=32, primary_key=True) crest = models.ImageField(...) references = models.ManyToManyField('FamilyDocuments', ...) origin = models.ForeignKeyField('Country')
A separate table contains the individual people. Each person has a last name and if that last name corresponds to a Surname in the Surname table, we want to be able to join from Person to Surname. However, not every person's last name will correspond to a Surname that we have detailed information for. So we don't want to require a record in the Surname table in order to create a Person. The person class will use the JoinField:
class Person(models.Model): first_name = models.CharField(max_length=32) last_name = JoinField(Surname, on_delete=models.DO_NOTHING)
This will result in the database column for Person.last_name being a CharField just like it is defined on Surname.name.
A Person object can be created by assigning either a Surname object or a literal value to the last_name attribute:
# Create a Person object with a literal value for last name Person.objects.create(first_name='Jon', last_name='Doe') # A Surname object can be created after the fact doe = Surname.objects.create(name='Doe') # A person can be created by passing a Surname object as the last_name Person.objects.create(first_name='Jane', last_name=doe)
The ORM can be used for both forward and reverse relationships:
# These two queries are equivalent Person.objects.filter(last_name='Doe') # literal filter Person.objects.filter(last_name=doe) # filter on forward relationship # The reverse relationship can also be used for filtering Surname.objects.filter(person__first_name='Jon') # And annotations Surname.objects.annotate(count=Count('person'))
Other than the lack of database constraints (which provides the ability to assign a literal in addition to an instance), the JoinField is very much like a ForeignKey field. The value that is stored in the database is the literal value assigned or the value of the field on the instance we join to. When an instance is retrieved, the attribute defined as a JoinField will have the instance we join to as it's value. An additional attribute with "_id" appended stores the literal value.
From the example above, when a Person instance is retrieved, the last_name attribute will be a Surname instance and the last_name_id attribute will be the actual value of the last_name. For Person instances that don't join to any value in the Surname table (not valid for ForeignKey), the last_name attribute will be None and the last_name_id attribute will still have the literal last_name value.
jon = Person.objects.get(first_name='Jon', last_name='Doe') print (jon.first_name, jon.last_name, jon.last_name_id)
Will print the following:
(u'Jon', <Surname: Surname object>, u'Doe')
And if we create a Person with no Surname:
james = Person.objects.create(first_name='James', last_name='Smith') print (james.first_name, james.last_name, james.last_name_id)
prints the following:
(u'James', None, u'Smith')