select_for_update(of) crash when used with values_list since Django 5.2 on Postgres

[Simon Charette]

This was extracted from #36299 as a distinct issue.

---

Queryset evaluation raises an AttributeError if the database backend supports SELECT ... FOR UPDATE. The following code:

with atomic():
    values = (
        User.objects.select_for_update(of=("self",))
        .values_list(
            Concat(F("first_name"), Value(" "), F("last_name")), "email"
        )
        .get(pk=12)
    )

will fail with a stacktrace ending in AttributeError: 'Concat' object has no attribute 'target' using the Postgresql DB backend.

Replicate by Sarah as a regression in 65ad4ade74dc9208b9d686a451cd6045df0c9c3a

tests/select_for_update/tests.py

@@ -13 +13 @@
     router,
     transaction,
 )
+from django.db.models import F, Value
+from django.db.models.functions import Concat
 from django.test import (
     TransactionTestCase,
     override_settings,
@@ -122 +124 @@
             list(Person.objects.select_for_update(no_key=True))
         self.assertIs(self.has_for_update_sql(ctx.captured_queries, no_key=True), True)
 
+    @skipUnlessDBFeature("has_select_for_update_of")
+    def test_for_update_of_values_list(self):
+        with transaction.atomic():
+            values = (
+                Person.objects.select_for_update(
+                    of=("self",),
+                )
+                .values_list(Concat(Value("Dr. "), F("name")), "born")
+            ).get(pk=self.person.pk)
+        self.assertTupleEqual(values, ("Dr. Reinhardt", self.city1.pk))
+
     @skipUnlessDBFeature("has_select_for_update_of")

[Simon Charette]

Self-accepting since #36299 was accepted once the regression was identified.

[Simon Charette]

​PR

[Mariusz Felisiak <felisiak.mariusz@…>]

In 71a19a0:

Fixed #36301 -- Fixed select_for_update(of) crash when using values()/values_list().

Regression in 65ad4ade74dc9208b9d686a451cd6045df0c9c3a which allowed for
annotations to be SELECT'ed before model field references through
values()/values_list() and broke assumptions the select_for_update(of)
table infererence logic had about model fields always being first.

Refs #28900.

Thanks OutOfFocus4 for the report and Sarah for the test.

[Mariusz Felisiak <felisiak.mariusz@…>]

In 5d2a0c5:

[5.2.x] Fixed #36301 -- Fixed select_for_update(of) crash when using values()/values_list().

Regression in 65ad4ade74dc9208b9d686a451cd6045df0c9c3a which allowed for
annotations to be SELECT'ed before model field references through
values()/values_list() and broke assumptions the select_for_update(of)
table infererence logic had about model fields always being first.

Refs #28900.

Thanks OutOfFocus4 for the report and Sarah for the test.

Backport of 71a19a0e475165dbc14c1fe02f552013ee670e4c from main