Opened 3 days ago
Last modified 3 days ago
#36198 assigned Bug
Unique Constraints defined with __date lookups on F expressions fail validation but create database constraints sucessfully
Reported by: | Tom Hall | Owned by: | Simon Charette |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | 5.1 |
Severity: | Normal | Keywords: | constraint lookup unique validation |
Cc: | Triage Stage: | Accepted | |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
Unique constraints defined with date lookups on F expressions create the correct index and work as expected at the database/ORM level but raise false positives during validation.
Initially discovered this for date lookups but it would seem to apply to the other date/time lookups, at this point I have explicitly confirmed the same behaviour for year, iso_year and minute.
Minimal reproducable example:
# models.py from django.db import models class DateLookup(models.Model): timestamp = models.DateTimeField() class Meta: constraints = [ # triggering validation errors for objects that can actually be saved models.UniqueConstraint( models.F("timestamp__date"), name="%(app_label)s_%(class)s_unique_date_F_expr", ), ] # tests.py import datetime as dt from django.core.exceptions import ValidationError from django.db import IntegrityError from django.utils import timezone from django.test import TestCase from .models import DateLookup TIMESTAMP = timezone.make_aware(dt.datetime(2025, 2, 18, 8, 18, 6)) class TestDateLookupConstraint(TestCase): def test_demonstrate_constraint_fails(self): """Demonstrate constraint validation failure at python level when db constraint is not triggered""" DateLookup.objects.create(timestamp=TIMESTAMP) new = DateLookup(timestamp=TIMESTAMP - dt.timedelta(days=1)) new.save() #success self.assertTrue(new.pk) try: new.full_clean() #ValidationError except ValidationError: self.fail("False Positive") def test_constraint_db(self): """Demonstrate DB constraint is triggered sucessfully.""" DateLookup.objects.create(timestamp=TIMESTAMP) new = DateLookup(timestamp=TIMESTAMP) with self.assertRaises(ValidationError): new.full_clean() with self.assertRaises(IntegrityError): new.save() self.assertFalse(new.pk)
Output of sqlmigrate command for SQLite backend:
BEGIN; -- -- Create model DateLookup -- CREATE TABLE "minimal_example_datelookup" ("id" integer NOT NULL PRIMARY KEY AUTOINCREMENT, "timestamp" datetime NOT NULL); CREATE UNIQUE INDEX "minimal_example_datelookup_unique_date_F_expr" ON "minimal_example_datelookup" ((django_datetime_cast_date("timestamp", 'UTC', 'UTC'))); COMMIT;
The workaround is to explicitly use the corresponding database functions in the constraint definition; this version works correctly.
class TruncDateExpression(models.Model): timestamp = models.DateTimeField() class Meta: constraints = [ models.UniqueConstraint( models.functions.TruncDate("timestamp"), name="%(app_label)s_%(class)s_unique_date", ), ]
The docs for constraint expressions don't show any examples of lookups being used in this way but also don't explicitly disallow them either, and given that the correct constraint is in fact being generated on the database, it implies this could be fixed to work as expected. Otherwise, at a minimum the documentation should be updated to forbid the use of F expression lookups, and the constraint should fail to apply during migration.
Initially discovered this and investigated on the django forum, see there for additional context and my efforts at debugging the validation step.
Change History (8)
comment:1 by , 3 days ago
Description: | modified (diff) |
---|
comment:2 by , 3 days ago
Triage Stage: | Unreviewed → Accepted |
---|
follow-up: 7 comment:3 by , 3 days ago
To me this is a duplicate or at least closely related to #34871 and is due to the fact that models.F("timestamp__date")
is not turned into something alike UnresolvedTransform(F("timestamp"), "date")
which would allow proper replacement of F("timestamp")
by Value(TIMESTAMP - dt.timedelta(days=1), DateLookup._meta.get_field("timestamp"))
.
The could be ways to resolve this particular issue by adapting F.replace_expressions
when LOOKUP_SEP in self.name
and the replacement (a Value
instance in the case of constraint validation) has an output_field
though which might be a less invasive solution.
comment:4 by , 3 days ago
comment:5 by , 3 days ago
Has patch: | set |
---|
comment:6 by , 3 days ago
Owner: | set to |
---|---|
Status: | new → assigned |
comment:7 by , 3 days ago
Replying to Simon Charette:
To me this is a duplicate or at least closely related to #34871 and is due to the fact that
models.F("timestamp__date")
is not turned into something alikeUnresolvedTransform(F("timestamp"), "date")
which would allow proper replacement ofF("timestamp")
byValue(TIMESTAMP - dt.timedelta(days=1), DateLookup._meta.get_field("timestamp"))
.
I did spot that one and meant to link to it in the description as maybe related but wasn't familiar enough at all with the underlying mechanics to be confident. And then I forgot. First time here and getting used to the system. Nice work on fixing it up, I suspected something in F.replace_expressions
was needed.
Can try writing something up for the docs to demonstrate this being used if that would be helpful?
comment:8 by , 3 days ago
Can try writing something up for the docs to demonstrate this being used if that would be helpful?
Thanks for the offer but we don't normally document bug fixing so in this case I don't think any documentation update is necessary.
I'll adjust the PR to include Sarah's test as well.
Thank you for the ticket!
I can replicate and I agree that if this is not allowed it is not clear in the docs
Here is a possible regression test
tests/constraints/models.py
tests/constraints/tests.py