我有一个Django应用程序，我想在其中将一个字段从ForeignKey更改为ManyToManyField。我想保留我的旧数据。最简单/最好的流程是什么？如果重要，我使用sqlite3作为我的数据库后端。
如果我对问题的总结不清楚，这是一个例子。说我有两个型号：
class Author(models.Model):
author = models.CharField(max_length=100)
class Book(models.Model):
author = models.ForeignKey(Author)
title = models.CharField(max_length=100)
说我的数据库中有很多数据。现在，我想按如下方式更改Book模型：
class Book(models.Model):
author = models.ManyToManyField(Author)
title = models.CharField(max_length=100)
我不想“丢失”我之前的所有数据。
实现这一目标的最佳/最简单方法是什么？
肯
答案 0 :(得分：58)
我意识到这个问题很老，当时数据迁移的最佳选择是使用South。现在Django有自己的migrate命令，过程略有不同。
我已将这些模型添加到名为books的应用中 - 如果不是您的情况，请相应调整。
首先，将字段添加到Book和related_name至少一个或两个（或者它们会发生冲突）：
class Book(models.Model):
author = models.ForeignKey(Author, related_name='book')
authors = models.ManyToManyField(Author, related_name='books')
title = models.CharField(max_length=100)
生成迁移：
$ ./manage.py makemigrations
Migrations for 'books':
0002_auto_20151222_1457.py:
- Add field authors to book
- Alter field author on book
现在，创建一个空迁移来保存数据本身的迁移：
./manage.py makemigrations books --empty
Migrations for 'books':
0003_auto_20151222_1459.py:
并添加以下内容。要准确了解其工作原理，请查看Data Migrations上的文档。注意不要覆盖迁移依赖性。
# -*- coding: utf-8 -*-
from __future__ import unicode_literals
from django.db import models, migrations
def make_many_authors(apps, schema_editor):
"""
Adds the Author object in Book.author to the
many-to-many relationship in Book.authors
"""
Book = apps.get_model('books', 'Book')
for book in Book.objects.all():
book.authors.add(book.author)
class Migration(migrations.Migration):
dependencies = [
('books', '0002_auto_20151222_1457'),
]
operations = [
migrations.RunPython(make_many_authors),
]
现在从模型中删除author字段 - 它应如下所示：
class Book(models.Model):
authors = models.ManyToManyField(Author, related_name='books')
title = models.CharField(max_length=100)
为此创建一个新的迁移，并全部运行：
$ ./manage.py makemigrations
Migrations for 'books':
0004_remove_book_author.py:
- Remove field author from book
$ ./manage.py migrate
Operations to perform:
Synchronize unmigrated apps: messages, staticfiles
Apply all migrations: admin, auth, sessions, books, contenttypes
Synchronizing apps without migrations:
Creating tables...
Running deferred SQL...
Installing custom SQL...
Running migrations:
Rendering model states... DONE
Applying books.0002_auto_20151222_1457... OK
Applying books.0003_auto_20151222_1459... OK
Applying books.0004_remove_book_author... OK
就是这样。之前在book.author处可用的作者现在应该位于book.authors.all()的查询集中。
答案 1 :(得分：1)
你应该做的最好和最简单的事情可能是：
创建具有不同名称的多对多字段
CREATE FUNCTION [dbo].[XMLTable](
@x XML
)
RETURNS TABLE
AS RETURN
/*----------------------------------------------------------------------
This INLINE TVF uses a recursive CTE that processes each element and
attribute of the XML document passed in.
----------------------------------------------------------------------*/
WITH cte AS (
/*------------------------------------------------------------------
Anchor part of the recursive query. Retrieves the root element
of the XML document
------------------------------------------------------------------*/
SELECT
1 AS lvl,
x.value('local-name(.)','NVARCHAR(MAX)') AS Name,
CAST(NULL AS NVARCHAR(MAX)) AS ParentName,
CAST(1 AS INT) AS ParentPosition,
CAST(N'Element' AS NVARCHAR(20)) AS NodeType,
x.value('local-name(.)','NVARCHAR(MAX)') AS FullPath,
x.value('local-name(.)','NVARCHAR(MAX)')
+ N'['
+ CAST(ROW_NUMBER() OVER(ORDER BY (SELECT 1)) AS NVARCHAR)
+ N']' AS XPath,
ROW_NUMBER() OVER(ORDER BY (SELECT 1)) AS Position,
x.value('local-name(.)','NVARCHAR(MAX)') AS Tree,
x.value('text()[1]','NVARCHAR(MAX)') AS Value,
x.query('.') AS this,
x.query('*') AS t,
CAST(CAST(1 AS VARBINARY(4)) AS VARBINARY(MAX)) AS Sort,
CAST(1 AS INT) AS ID
FROM @x.nodes('/*') a(x)
UNION ALL
/*------------------------------------------------------------------
Start recursion. Retrieve each child element of the parent node
------------------------------------------------------------------*/
SELECT
p.lvl + 1 AS lvl,
c.value('local-name(.)','NVARCHAR(MAX)') AS Name,
CAST(p.Name AS NVARCHAR(MAX)) AS ParentName,
CAST(p.Position AS INT) AS ParentPosition,
CAST(N'Element' AS NVARCHAR(20)) AS NodeType,
CAST(
p.FullPath
+ N'/'
+ c.value('local-name(.)','NVARCHAR(MAX)') AS NVARCHAR(MAX)
) AS FullPath,
CAST(
p.XPath
+ N'/'
+ c.value('local-name(.)','NVARCHAR(MAX)')
+ N'['
+ CAST(ROW_NUMBER() OVER(
PARTITION BY c.value('local-name(.)','NVARCHAR(MAX)')
ORDER BY (SELECT 1)) AS NVARCHAR )
+ N']' AS NVARCHAR(MAX)
) AS XPath,
ROW_NUMBER() OVER(
PARTITION BY c.value('local-name(.)','NVARCHAR(MAX)')
ORDER BY (SELECT 1)) AS Position,
CAST(
SPACE(2 * p.lvl - 1) + N'|' + REPLICATE(N'-', 1)
+ c.value('local-name(.)','NVARCHAR(MAX)') AS NVARCHAR(MAX)
) AS Tree,
CAST( c.value('text()[1]','NVARCHAR(MAX)') AS NVARCHAR(MAX) ) AS Value,
c.query('.') AS this,
c.query('*') AS t,
CAST(
p.Sort
+ CAST( (lvl + 1) * 1024
+ (ROW_NUMBER() OVER(ORDER BY (SELECT 1)) * 2) AS VARBINARY(4)
) AS VARBINARY(MAX) ) AS Sort,
CAST(
(lvl + 1) * 1024
+ (ROW_NUMBER() OVER(ORDER BY (SELECT 1)) * 2) AS INT
)
FROM cte p
CROSS APPLY p.t.nodes('*') b(c)
), cte2 AS (
SELECT
lvl AS Depth,
Name AS NodeName,
ParentName,
ParentPosition,
NodeType,
FullPath,
XPath,
Position,
Tree AS TreeView,
Value,
this AS XMLData,
Sort, ID
FROM cte
UNION ALL
/*------------------------------------------------------------------
Attributes do not need recursive calls. So add the attributes
to the query output at the end.
------------------------------------------------------------------*/
SELECT
p.lvl,
x.value('local-name(.)','NVARCHAR(MAX)'),
p.Name,
p.Position,
CAST(N'Attribute' AS NVARCHAR(20)),
p.FullPath + N'/@' + x.value('local-name(.)','NVARCHAR(MAX)'),
p.XPath + N'/@' + x.value('local-name(.)','NVARCHAR(MAX)'),
1,
SPACE(2 * p.lvl - 1) + N'|' + REPLICATE('-', 1)
+ N'@' + x.value('local-name(.)','NVARCHAR(MAX)'),
x.value('.','NVARCHAR(MAX)'),
NULL,
p.Sort,
p.ID + 1
FROM cte p
CROSS APPLY this.nodes('/*/@*') a(x)
)
SELECT
ROW_NUMBER() OVER(ORDER BY Sort, ID) AS ID,
ParentName, ParentPosition,Depth, NodeName, Position,
NodeType, FullPath, XPath, TreeView, Value, XMLData
FROM cte2;
go
SELECT * FROM dbo.XMLTable('
<employees>
<emp name="jacob"/>
<emp name="steve">
<phone>123</phone>
some text
</emp>
</employees>
')
编写一个小函数将foreignkey值转换为M2M值：
authors = models.ManyToManyField(Author)
运行后，您可以从表中删除作者字段并再次运行迁移。