Improve ListSerializer

[felipediel]

Description

#8979 prevents list serializers from having data and instances with different lengths:

https://github.com/encode/django-rest-framework/pull/8979/files#diff-c33f1f011c9f5cf3ed1357b809ebf2b4ed0b808b227c6c007291bf9b259422ecR612-R617

This design choice breaks some use cases. Let's say we want the list serializer to delete from queryset all elements that are not provided in the data during an update. The queryset will be bigger than the data in this case:

# Example: serializer that will break after #8979

class BankBulkUpdateSerializer(serializers.ListSerializer):
    """Bank bulk update serializer."""

    def update(
        self,
        instance: models.QuerySet[Bank],
        validated_data: list[OrderedDict],
    ) -> list[Bank]:
        """Bulk update banks."""
        bank_dict = {bank.pk: bank for bank in instance}
        banks_to_create: list[Bank] = []
        banks_to_update: list[Bank] = []
        banks_created: list[Bank] = []
        banks_updated: list[Bank] = []
        bank_pks_to_keep: list[int] = []

        for bank_data in validated_data:
            bank_id = bank_data.get("id")
            bank = bank_dict.get(bank_id) if bank_id is not None else None

            if bank is not None:
                for attr, value in bank_data.items():
                    setattr(bank, attr, value)
                banks_to_update.append(bank)
                bank_pks_to_keep.append(bank.pk)

            else:
                bank = Bank(**bank_data)
                banks_to_create.append(bank)

        with db_transaction.atomic():
            instance.exclude(pk__in=bank_pks_to_keep).delete()

            if banks_to_update:
                update_fields = ["name", "compe", "ispb", "website"]
                Bank.objects.bulk_update(banks_to_update, update_fields)
                banks_updated = banks_to_update

            if banks_to_create:
                banks_created: list[Bank] = Bank.objects.bulk_create(
                    banks_to_create
                )

        return sorted(banks_created + banks_updated, key=lambda a: a.name)

The intention of the original PR is good, it solves the problem of not passing initial data and instances to child serializers correctly. But instead of indexing, it's better to relate the elements by id, so we can have different lengths. This PR addresses the issue.

refs #8926 #8979

[felipediel]

I'm not sure this is the best place to do this. I'd like to hear the opinion of someone more familiar with this codebase.

[auvipy]

can you give a more verbose name to rel_mgr please? also share a bit more of what this would be doing?

[felipediel]

I just realized there's a side case here. The child may not be a ModelSerializer. I'll leave it as a draft while I think about it, suggestions are welcome.

We could create a new ModelListSerializer class with these changes and have it as the default list_serializer_class of ModelSerializer. @auvipy let me know what you think.

[auvipy]

I think we should have appropriate test cases for this before even trying to fix them

[felipediel]

Agree, I'll do it

[auvipy]

as you can see the CI builds are failing, can you please rebase and make it green at least? possibly with very basic test? you can try incremental approach to fix the problem

[felipediel]

Yes, I will do it. Sorry for the delay, I'm struggling to find time to continue this PR, any help is welcome