Allow creating models with fields values as a Subquery() with F() expressions

[Charlie McBride]

I understand why F() expressions are generally disallowed inside insert statements, the columns you are referencing don't yet exist, so it wouldn't make any sense to do so. However, if you are performing an insert with a Subquery (as in the example below), it's possible to have otherwise valid statements rejected because of the blanket blacklist of F() expressions during inserts.

For example:

given the following models:

class Item(models.Model):
    pass

class ItemVersion(models.Model):
    item = models.ForeignKey(Item, related_name='versions')
    version_number = models.IntegerField(default=0)

    class Meta:
        unique_together = ('item', 'version_number',)

I would like to be able to do the following operation to mitigate race conditions as described here (​https://docs.djangoproject.com/en/2.1/ref/models/expressions/#avoiding-race-conditions-using-f)

    item = Item.objects.create()
    # arbitrary number of other items created/destroyed etc
    item_version_2 = ItemVersion.objects.create(
        item=item,
        version_number=Subquery(
            item.versions.order_by('-version_number').annotate(
                max_version_number=Coalesce(Max('version_number'), 0)
            ).annotate(
                new_version_number=F('max_version_number) + 1
            ).values('new_version_number')
         )
     )

As written, I would expect the F() in the inner expression to always be resolvable, because it is in a Subquery (and not the result of an insert). However this query is blocked by the compiler because "F() expressions can only be used to update, not to insert." Would it be possible to allow F expressions in a Subquery even if it is being used in an insert? Is there an edge case that I'm missing that caused the team to not consider this?

[Tim Graham]

It seems okay, if it's feasible to implement.

[Sarah Boyce]

I think this is resolved (I'm not sure when), raised a PR just to show that I think it's been fixed ​https://github.com/django/django/pull/16451 (but I will close the PR later).
I feel like I'm not supposed to close tickets so I'll let a DSF team member check I haven't missed anything and close the ticket.

[Tim Graham]

I did a bisect and confirmed this was fixed in Django 3.0 by 35431298226165986ad07e91f9d3aca721ff38ec.

[Sarah Boyce]

Thank you Tim! Updated the commit message to credit 👍

[Mariusz Felisiak <felisiak.mariusz@…>]

In 05bcd5ba:

Refs #30129 -- Added test for create() with F() expression in Subquery.

Fixed in 35431298226165986ad07e91f9d3aca721ff38ec.