`BaseInlineFormSet` should attempt to get it's queryset from it's instance related manager before falling back to it's model's default manager

[Simon Charette]

The newly introduced prefetch_related method can be quite handy to avoid unnecessary queries (thanks Luke :), however it's quite useless when used with BaseInlineFormSet since ​it won't even try to get its needed `queryset` from the related manager.

i.e.

from django.db import models

class Author(models.Model):
    name = models.CharField(max_length=255)

class Book(models.Model):
    name = models.CharField(max_length=255)
    author = models.ForeignKey(Author, related_name='books')
    
    class Meta:
        # Needed because `BaseModelFormSet.get_query_set` wants an ordered set or issue another query
        ordering = ('pk',)

In [1]: from django.conf import settings

In [2]: from django.db import connection

In [3]: from library.models import *

In [4]: from django.forms.models import inlineformset_factory

In [5]: a = Author.objects.create(name='David Abram')

In [6]: b1 = Book.objects.create(name='Becoming Animal', author=a)

In [7]: b2 = Book.objects.create(name='The Spell of the Sensuous', author=a)

In [8]: BookInlineFormSet = inlineformset_factory(Author, Book)

In [9]: settings.DEBUG = True

In [10]: instance = Author.objects.prefetch_related('books').get()

In [11]: BookInlineFormSet(instance=instance)
Out[11]: <django.forms.formsets.BookFormFormSet at 0x3c68d50>

In [12]: print connection.queries
[{u'time': u'0.000', u'sql': u'SELECT "library_author"."id", "library_author"."name" FROM "library_author"'},
 {u'time': u'0.000', u'sql': u'SELECT "library_book"."id", "library_book"."name", "library_book"."author_id" FROM "library_book" WHERE "library_book"."author_id" IN (1) ORDER BY "library_book"."id" ASC'},
 {u'time': u'0.000', u'sql': u'SELECT "library_book"."id", "library_book"."name", "library_book"."author_id" FROM "library_book" WHERE "library_book"."author_id" = 1  ORDER BY "library_book"."id" ASC'}]

I guess it's only a matter of time before people start trying to optimize their formsets (or their ModelAdmin by overriding get_queryset) and stumble on this limitation.

I've written a patch (I'll create a pull request) for which I'll write tests if this optimization is Accepted.

[Simon Charette]

ticket-18597.0.diff

[Simon Charette]

All tests pass on Python 2.7.3rc1 Postgres and Sqlite.

[Aymeric Augustin]

I'm not sold on the patch -- it looks suspiciously complex -- but the problem is real.

[Andreas Galazis]

Transfering my concerns from the duplicate issue. I there is some code cleanup needed before attempting to fix this.
This will at least allow people to extend existing base classes to create custom inlines that can do ( facilitating the feature in the core framework is secondary)
This would help support prefetching configuration on parent admin's queryset.
This is not sufficient to support prefetching since there are multiple (I have found at least 3) instances in inline admin and inline formset where self.get_queryset()[i] paradigm is used.

At least perform some code cleanup?

Even if this issue a won't fix for any reason at least abstract fetching instance by in index in a ge_instance_by_index method so that we can implement admins that do what we want of top of current base classes. If this is done then we will be able to overwrite ge_instance_by_index to return list(self.get_queryset())[i] instead of self.get_queryset()[i] on inlines that support prefetching

Also I believe that the logic about preparing queryset in BaseInlineFormSet should exist in a separate function called prepare_queryset:

def prepare_queryset(queryset):
        if queryset is None:
            queryset = self.model._default_manager
        if self.instance.pk is not None:
            qs = queryset.filter(**{self.fk.name: self.instance})
        else:
            qs = queryset.none()
        return qs

instead if throwing it __init__. Init could just call the above to get the value . This way filtering wouldn't be enforced in the case of related manager querysets since we could overwrite the method.

[Andreas Galazis]

added a PR related to the code cleanup needed to support custom inlines :
​https://github.com/django/django/pull/14188

[Andreas Galazis]

Can somebody review my PR I literally didn't change anything apart from moving code around so that we can overwrite functionality

[Tony Narlock]

For context, this is a blocker for a downstream plugin, see ​https://github.com/theatlantic/django-nested-admin/issues/76

[Steven Johnson]

Added new pull request with docs/tests:
​https://github.com/django/django/pull/17818