Renaming models prompts for content type deletions

[doepunk]

When renaming a model (from OldModel to NewModel) I noticed that makemigrations asks for confirmation if I had renamed NewModel to OldModel. Which is great that that was detected and the confirmation helps to get a correct migration file, which makes this step reproducible. The Model in question did not involve any foreign keys to or from it. However when I ran the migrate, I got this message

The following content types are stale and need to be deleted:

    <app-name> | <old-name>

Any objects related to these content types by a foreign key will also
be deleted. Are you sure you want to delete these content types?
If you're unsure, answer 'no'.

I'm unsure what's going on here. But shouldn't renaming involve the automatic deletion of the old name and replacing the foreign keys with the new database table keys? All without a need for confirmation. Or at least if there are no foreign keys as in my case, the confirmation seems unnecessary and it only makes it harder to run the migration unsupervised and it introduces a potential error source. Assuming that all data from the old table were copied to the new table and all foreign key relationships copied and updated as well, I don't see a use case for keeping the old table and "related objects" around. And as for the case that there were any changes in relationships reflected in the Models (other than the name change) this could be detected and confirmed in the makemigrations stage. (For example if there were still references to OldName in the code when OldName no longer exists as a class, this would result in an exception, or if a Foreign Key attribute to OldModel was deleted and not replaced with NewModel Foreign Key it's just akin to a field deletion). It just seems more reproducible to keep all confirmations in the makemigration stage.

[Tim Graham]

I'm not sure if automating the data migration is a good idea, but we should likely at least document these considerations if not.

[Simon Charette]

IMO the content types should be automatically updated upon RenameModel operations.

Unfortunately there's no way for the django.contrib.contenttypes.management.update_contenttypes receiver to differentiate between model renaming, addition and deletion hence the current prompt.

Would it make sense to dispatch the plan via post_migrate to allow receivers to deal with applied migrations? In the case of the flush command plan would be None.

Given this plan, update_contenttypes could automatically update the ContentType.model by following the RenameModel chain.

By the way I find it a bit odd that we document that post_migrate receivers should never perform database operation when update_contenttypes is.

[Aymeric Augustin]

I agree with charettes. I wrote data migrations to handle this in some projects.

[Claude Paroz]

+1

Having the plan in post_migrate would also allow us to easily fix #24075.

[Simon Charette]

Created #24100 to track the addition of the plan kwarg.

[Simon Charette]

​PR

[Tim Graham]

Doesn't seem to work quite right in my testing.

[Simon Charette]

The case issue should be fixed.

[Aymeric Augustin]

Good to see that this ticket is moving forwards, as I just hit this issue again.

[Simon Charette]

I should be able to deal with what's left next week.

[Simon Charette]

​updated PR.

[Simon Charette <charette.s@…>]

In f179113e:

Fixed #24067 -- Renamed content types upon model renaming.

Thanks to Tim for the extensive review.

[Simon Charette <charette.s@…>]

In 722344ee:

Refs #24067 -- Fixed contenttypes rename tests failures on Oracle.

Broke the initial migration in two to work around #25530 and added
'django.contrib.auth' to the available_apps to make sure its tables are also
flushed as Oracle doesn't implement cascade deletion in sql_flush().

Thanks Tim for the report.

[Simon Charette <charette.s@…>]

In 826ec5ee:

[1.10.x] Refs #24067 -- Fixed contenttypes rename tests failures on Oracle.

Broke the initial migration in two to work around #25530 and added
'django.contrib.auth' to the available_apps to make sure its tables are also
flushed as Oracle doesn't implement cascade deletion in sql_flush().

Thanks Tim for the report.

Backport of 722344ee59fb89ea2cd5b906d61b35f76579de4e from master