Opened 3 years ago
Closed 3 years ago
#33808 closed Bug (duplicate)
Specific subquery produces wrong SQL (error 500)
Reported by: | Fabio Zoratti | Owned by: | nobody |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | 3.2 |
Severity: | Normal | Keywords: | database, count, query, orm |
Cc: | Triage Stage: | Unreviewed | |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
You can find in attachment a minimal working example. I created the zip with
django-admin startproject mwe ./manage.py startapp testapp
Then I only edited the file testapp/models.py
testapp/tests.py
and mwe/settings.py
(to add the app to the installed apps) and ran ./manage.py makemigrations testapp
.
Running the tests with the usual ./manage.py test
triggers the problem. I printed the query produced by django, simply adding a print
in django/db/backends/sqlite/base.py
. The query is the following:
SELECT DISTINCT "testapp_firstmodel"."id", COUNT("testapp_secondmodel_related_field2"."secondmodel_id" IN "testapp_secondmodel_related_field1"."secondmodel_id") AS "howmany" FROM "testapp_firstmodel" LEFT OUTER JOIN "testapp_secondmodel_related_field1" ON ("testapp_firstmodel"."id" = "testapp_secondmodel_related_field1"."firstmodel_id") LEFT OUTER JOIN "testapp_secondmodel_related_field2" ON ("testapp_firstmodel"."id" = "testapp_secondmodel_related_field2"."firstmodel_id") GROUP BY "testapp_firstmodel"."id" LIMIT 21
Adding a layer of parenthesis in the query inside the Count
fixes the issue.
SELECT DISTINCT "testapp_firstmodel"."id", COUNT("testapp_secondmodel_related_field2"."secondmodel_id" IN ("testapp_secondmodel_related_field1"."secondmodel_id")) AS "howmany" FROM "testapp_firstmodel" LEFT OUTER JOIN "testapp_secondmodel_related_field1" ON ("testapp_firstmodel"."id" = "testapp_secondmodel_related_field1"."firstmodel_id") LEFT OUTER JOIN "testapp_secondmodel_related_field2" ON ("testapp_firstmodel"."id" = "testapp_secondmodel_related_field2"."firstmodel_id") GROUP BY "testapp_firstmodel"."id" LIMIT 21
This should fix the issue: https://github.com/django/django/pull/15798
Attachments (1)
Change History (6)
by , 3 years ago
comment:1 by , 3 years ago
I used git bisect
to find the commit that broke the code. The commit is
commit 3a505c70e7b228bf1212c067a8f38271ca86ce09 (HEAD, refs/bisect/bad) Author: Simon Charette <charette.s@gmail.com> Date: Wed Mar 6 01:24:41 2019 -0500 Refs #27149, #29542 -- Simplified subquery parentheses wrapping logic.
The diff clearly shows that some parenthesis are being removed from the query
git show 3a505c70e7b228bf1212c067a8f38271ca86ce09 diff --git a/django/db/models/lookups.py b/django/db/models/lookups.py index e78ffdf390..910291094e 100644 --- a/django/db/models/lookups.py +++ b/django/db/models/lookups.py @@ -89,8 +89,7 @@ class Lookup: value = self.apply_bilateral_transforms(value) value = value.resolve_expression(compiler.query) if hasattr(value, 'as_sql'): - sql, params = compiler.compile(value) - return '(' + sql + ')', params + return compiler.compile(value) else: return self.get_db_prep_lookup(value, connection)
comment:2 by , 3 years ago
Summary: | Count manytomany__in produces wrong SQL → Specific subquery produces wrong SQL |
---|
comment:3 by , 3 years ago
Summary: | Specific subquery produces wrong SQL → Specific subquery produces wrong SQL (error 500) |
---|
comment:4 by , 3 years ago
Description: | modified (diff) |
---|---|
Has patch: | set |
comment:5 by , 3 years ago
Resolution: | → duplicate |
---|---|
Status: | new → closed |
Thanks for the report, however #32673 is a regression in Django 3.0 reported when it was already EOL. Per our backporting policy this means it doesn't qualify for a backport to 3.2.x anymore. See Django’s release process for more details. Also, Django 3.2 is in extended support so it doesn't receive bugfixes anymore (except security patches).
Moreover, using the __in
lookup with ManyToManyField
has never been officially supported, you should use Q(sm_f1=F("sm_f2"))
instead of Q(sm_f1__in=F("sm_f2"))
, it's a duplicate of #31135.
Minimal django project reproducing the issue