Make QuerySet.bulk_create() populate fields on related models

[Jarek Glowacki]

As of dj110 bulk_create returns pks for the objects it has created. Great! I want more.

Going to hop right into the use case:

class A(models.Model):
    pass


class B(models.Model):
    a = models.ForeignKey(A, on_delete=models.PROTECT)

a = A()
b = B(a=a)  # doesn't set `a_id`, since `a` has no id assigned yet.
A.objects.bulk_create([a])  # sets `a`'s id.
B.objects.bulk_create([b])  # error!

The first bulk_create call sets the id on a, but b.a_id remains None.
When running the second bulk_create, even though b has an a set, it fails to save a onto the b instance, due to a_id being None.

So if you imagine the pregeneration of several related models in a big loop, followed by several consecutive bulk_create calls we run into trouble. It's nicer than having to manually feed in ids during the loop as we needed to in the past, but it still needs some magic.

My current solution is running a little function between each bulk_create call, to fill the foreignkey ids in:

def fill_foreignkey_ids(objects, fields):
    for f in fields:
        for o in objects:
            setattr(o, '%s_id' % f, getattr(o, f).id)

Which makes:

a = A()
b = B(a=a)  # doesn't set `a_id`, since `a` has no id assigned yet.
bulk_a = [a]
bulk_b = [b]
A.objects.bulk_create(bulk_a)  # sets `a`'s id.
fill_foreignkey_ids(bulk_b, ['a'])
B.objects.bulk_create(bulk_b)  # now it works. Lovely.

But this is quite ugly.

This is more of a brainstorm ticket, as I'm not sure how to solve this nicely. Expecting every unsaved model instance to update its foreignkey ids when a bulk_create inserts them sounds seems silly. But maybe we could have the bulk_create method check whether b.a.id is set when b.a_id isn't?

I can submit a PR if this sounds like a reasonable thing to do. It feels a bit magic, but it would greatly improve the usefulness of this new feature.

[Simon Charette]

I haven't tried it myself but I suspect this has little to do with bulk_create itself. Does the following fails in a similar manner?

a = A()
b = B(a=a)
a.save()
b.save()

[Jarek Glowacki]

Yep, that fails too.

The difference is use case. For single saves, it's easy to pass the id through. For bulk_creates it's a bit messier to do that.

I suggested only updating bulk_create (to automatically figure out what to do in that second call) as I wasn't sure it was feasible to make this work in the general case - but maybe by putting some more magic in the foreignkey field's __get__ could work?

[Tim Graham]

I'm not sure if this is a good idea due to magic/complexity, but if we had a patch, we could better evaluate it. I guess your use case doesn't allow constructing B after A.objects.bulk_create()?