It was likely overlooked by the patch. Looks like Model.__copy__ should make sure to make a deep-copy of self._state now that fields are cached in self._state.fields_cache.
Using copy.deepcopy will circumvent the issue but I feel like copy.copy is common enough pattern that we should maintain compatibility for it. After all, the fact the _state_ of the model is now stored in _state.fields_cache is no more than an implementation detail and it would be normal to expect shallow copying to work in this case. On the other hand the fact this has been happening since Django 2.0 and has just been reported and that shallow copies won't work correctly for prefetched relationships are both non-negligible argument for a wontfix since this pattern was never explicitly documented. There is a precedent for a similar undocumented pattern though [0]. Thoughts? Simon [0] https://github.com/django/django/commit/dee5896b55385d2c66d8c1b8386604868f7fc6b4 Le mardi 4 août 2020 à 10:48:43 UTC-4, gertb...@gmail.com a écrit : > Hi Simon, > > I think the commit is bfb746f983aa741afa3709794e70f1e0ab6040b5 "Refs > #16043 -- Refactored internal fields value cache". > > Cheers > > On Tue, 4 Aug 2020 at 15:52, charettes <chare...@gmail.com> wrote: > >> Hello Gert, that seems a bit surprising to me and was likely not a >> desired change. >> >> Could you bisect the exact commit that caused the regression[0]? That >> would certainly help determining the action to take here. >> >> Thanks, >> Simon >> >> [0] >> https://docs.djangoproject.com/en/3.1/internals/contributing/triaging-tickets/#bisecting-a-regression >> >> Le mardi 4 août 2020 à 08:10:00 UTC-4, gertb...@gmail.com a écrit : >> >>> (initially posted to >>> https://forum.djangoproject.com/t/fk-field-caching-behavior-change-between-1-11-x-and-2-x/3151 >>> >>> on 26 June but this mailing list might be more appropriate) >>> >>> Hi >>> >>> Whilst upgrading a codebase from 1.11.x to 2.0/2.2 I noticed a weird >>> change in behavior of FK fields when copying model instances. >>> >>> At the bottom of the post there is a testcase that succeeds on 1.11.x >>> and fails on 2.x >>> I think the commit that changed the behavior is >>> bfb746f983aa741afa3709794e70f1e0ab6040b5 >>> >>> So my question is two fold: >>> >>> 1. Is the behavior in >=2.0 correct? It seems quite unexpected. >>> 2. What is the recommended way to clone a model instance? To date we >>> have been using copy() in a similar fashion to the test without >>> issue. deepcopy seems to work fine in >=2.0 but we haven’t done too >>> much testing yet. >>> >>> Test (placed in tests/model_fields/test_field_caching_change.py): >>> >>> ---------------------code-------------------------- >>> >>> import copy >>> from django.test import TestCase >>> >>> from .models import Bar, Foo >>> >>> >>> class ForeignKeyCachingBehaviorTest(TestCase): >>> def test_copy(self): >>> foo1 = Foo.objects.create(a='foo1', d=1) >>> foo2 = Foo.objects.create(a='foo2', d=2) >>> bar1 = Bar.objects.create(a=foo1, b='bar1') >>> >>> bar2 = copy.copy(bar1) >>> >>> bar2.pk = None >>> bar2.a = foo2 >>> >>> # bar2 points to foo2 >>> self.assertEqual(bar2.a, foo2) >>> self.assertEqual(bar2.a.id, bar2.a_id) >>> >>> # bar1 is unchanged and must still point to foo1 >>> # These fail on Django >= 2.0 >>> self.assertEqual(bar1.a, foo1) >>> self.assertEqual(bar1.a.id, bar1.a_id) >>> >>> ---------------------code-------------------------- >>> >>> and executed that via: >>> >>> python3.6 tests/runtests.py --parallel 1 model_fields >>> >>> Regards >>> Gert Burger >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "Django developers (Contributions to Django itself)" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to django-develop...@googlegroups.com. >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/django-developers/c284e35e-ae02-44e5-b44d-4994eff9b248n%40googlegroups.com >> >> <https://groups.google.com/d/msgid/django-developers/c284e35e-ae02-44e5-b44d-4994eff9b248n%40googlegroups.com?utm_medium=email&utm_source=footer> >> . >> > -- You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group. To unsubscribe from this group and stop receiving emails from it, send an email to django-developers+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/0b5d48d0-5594-40d6-9dce-80995f8635ecn%40googlegroups.com.
