unique_together does not work when any of the listed fields contains a FK

I mean, when saving an "application" object, of course

In fact the problem occures if there is at least one ForeignKey? field specified in unique_together list (no matter if it is an edit_inline or not)

Yeah, what's the deal with this?

I am using unique_together with ForeignKey fields in a model that looks like:

class Course(models.Model):
    """A course number and title combination."""
    
    department = models.ForeignKey(Department)
    number = models.CharField('course number', maxlength=6)
    title = models.CharField('course title', maxlength=100)
    
    class Meta:
        # Don't allow duplicate courses.
        unique_together = (("department", "number", "title"),)

I have no problems saving Department or Course objects. I also have no problems when I add edit_inline=True in the department field as long as at least one other field has core=True. It's only if I fail to add a core=True to any field that I get the following IntegrityError:

Traceback (most recent call last):
File "/usr/lib/python2.4/site-packages/django/core/handlers/base.py" in get_response
  74. response = callback(request, *callback_args, **callback_kwargs)
File "/usr/lib/python2.4/site-packages/django/contrib/admin/views/decorators.py" in _checklogin
  55. return view_func(request, *args, **kwargs)
File "/usr/lib/python2.4/site-packages/django/views/decorators/cache.py" in _wrapped_view_func
  39. response = view_func(request, *args, **kwargs)
File "/usr/lib/python2.4/site-packages/django/contrib/admin/views/main.py" in add_stage
  254. new_object = manipulator.save(new_data)
File "/usr/lib/python2.4/site-packages/django/db/models/manipulators.py" in save
  198. new_rel_obj.save()
File "/usr/lib/python2.4/site-packages/django/db/models/base.py" in save
  203. ','.join(placeholders)), db_values)
File "/usr/lib/python2.4/site-packages/django/db/backends/util.py" in execute
  12. return self.cursor.execute(sql, params)

  IntegrityError at /admin/courses/department/add/
  ERROR: duplicate key violates unique constraint "courses_course_department_id_key" INSERT INTO "courses_course" ("department_id","number","title") VALUES (6,'','')

I also have a model that uses unique_together using two fields that are both ForeignKeys and I don't experience any problem there with or without edit_inline=True.

Notice that "This validation needs to occur after html2python to be effective." but validator always run before html2python. so I change ForeignKey? to use field.name instead of field.attname.

Index: db/models/manipulators.py
===================================================================
--- db/models/manipulators.py   (revision 4643)
+++ db/models/manipulators.py   (working copy)
@@ -283,7 +283,7 @@
         # This is really not going to work for fields that have different
         # form fields, e.g. DateTime.
         # This validation needs to occur after html2python to be effective.
-        field_val = all_data.get(f.attname, None)
+        field_val = all_data.get(f.name, None)
         if field_val is None:
             # This will be caught by another validator, assuming the field
             # doesn't have blank=True.

New version: unique together fields could be null=True now.

Index: db/models/manipulators.py
===================================================================
--- db/models/manipulators.py   (revision 4643)
+++ db/models/manipulators.py   (working copy)
@@ -283,13 +283,13 @@
         # This is really not going to work for fields that have different
         # form fields, e.g. DateTime.
         # This validation needs to occur after html2python to be effective.
-        field_val = all_data.get(f.attname, None)
-        if field_val is None:
+        field_val = f.get_manipulator_new_data(all_data)
+        if not f.null and field_val is None:
             # This will be caught by another validator, assuming the field
             # doesn't have blank=True.
             return
         if isinstance(f.rel, ManyToOneRel):
-            kwargs['%s__pk' % f.name] = field_val
+            kwargs[f.get_validator_unique_lookup_type()] = field_val
         else:
             kwargs['%s__iexact' % f.name] = field_val
     try:

I can confirm that the first patch submitted by favo works.

Maybe the problem is backend specific? For anyone seeing this problem, which database backend are you using?

My comments from above were on postgresql.

Same here, PostgreSQL.

I was on MySQL when I hit this problem.

I had this same problem until I switched the order of my fields in the unique_together clause. In my example, using 'slug' first threw an integrity error, but having 'site' first validates correctly. (I'm working in 91-bugfixes with PostgreSQL)

 slug = meta.SlugField(prepopulate_from=('name',))
 site = meta.ForeignKey(Site)

 class META:
     unique_together=(('site', 'slug'),)

#2019 marked as a duplicate of this.

Replying to favo@exoweb.net:

I change ForeignKey? to use field.name instead of field.attname.

This change was made in [4028].

i'm not sure what's the status of this ticket, but maybe a clean reproducible test-case will help.

this is with from-svn-django (r6022), postgresql8, "postgresql" driver.

from django.db.models import *

class A(Model):
    name = CharField(maxlength=200)

    class Admin: pass

class B(Model):
    name = CharField(maxlength=200)

    class Admin: pass

class C(Model):
    name = CharField(maxlength=200)
    a = ForeignKey(A)
    b = ForeignKey(B)

    class Meta:
        unique_together = (("a","b"),)

    class Admin: pass

in the admin:

1. create an instance of A
2. go to create an instance of C, and fill out the "name" field, and select an "a" object, but do not select a "b" object
3. save the instance

you will get an exception saying:

ERROR: invalid input syntax for integer: 
"" SELECT "main_c"."id","main_c"."name","main_c"."a_id","main_c"."b_id" 
FROM "main_c" INNER JOIN "main_a" AS "main_c__a" ON "main_c"."a_id" = "main_c__a"."id" 
WHERE ("main_c"."b_id" = '' AND "main_c__a"."id" ILIKE '1')

(isn't that ILIKE '1' quite nice for a numeric-key? :))

btw. this works fine in sqlite3, probably because it simply accepts this syntax...

the stacktrace:

Traceback (most recent call last):
File "/home/gabor/src/django/django/core/handlers/base.py" in get_response
  77. response = callback(request, *callback_args, **callback_kwargs)
File "/home/gabor/src/django/django/contrib/admin/views/decorators.py" in _checklogin
  55. return view_func(request, *args, **kwargs)
File "/home/gabor/src/django/django/views/decorators/cache.py" in _wrapped_view_func
  39. response = view_func(request, *args, **kwargs)
File "/home/gabor/src/django/django/contrib/admin/views/main.py" in add_stage
  257. errors = manipulator.get_validation_errors(new_data)
File "/home/gabor/src/django/django/oldforms/__init__.py" in get_validation_errors
  61. errors.update(field.get_validation_errors(new_data))
File "/home/gabor/src/django/django/oldforms/__init__.py" in get_validation_errors
  378. self.run_validator(new_data, validator)
File "/home/gabor/src/django/django/oldforms/__init__.py" in run_validator
  368. validator(new_data.get(self.field_name, ''), new_data)
File "/home/gabor/src/django/django/utils/functional.py" in _curried
  3. return _curried_func(*(args+moreargs), **dict(kwargs, **morekwargs))
File "/home/gabor/src/django/django/db/models/manipulators.py" in manipulator_validator_unique_together
  303. old_obj = self.manager.get(**kwargs)
File "/home/gabor/src/django/django/db/models/manager.py" in get
  69. return self.get_query_set().get(*args, **kwargs)
File "/home/gabor/src/django/django/db/models/query.py" in get
  261. obj_list = list(clone)
File "/home/gabor/src/django/django/db/models/query.py" in __iter__
  114. return iter(self._get_data())
File "/home/gabor/src/django/django/db/models/query.py" in _get_data
  482. self._result_cache = list(self.iterator())
File "/home/gabor/src/django/django/db/models/query.py" in iterator
  189. cursor.execute("SELECT " + (self._distinct and "DISTINCT " or "") + ",".join(select) + sql, params)
File "/home/gabor/src/django/django/db/backends/util.py" in execute
  19. return self.cursor.execute(sql, params)
File "/home/gabor/src/django/django/db/backends/postgresql/base.py" in execute
  47. return self.cursor.execute(smart_str(sql, self.charset), self.format_params(params))

  ProgrammingError at /admin/main/c/add/
  ERROR: invalid input syntax for integer: "" SELECT "main_c"."id","main_c"."name","main_c"."a_id","main_c"."b_id" FROM "main_c" INNER JOIN "main_a" AS "main_c__a" ON "main_c"."a_id" = "main_c__a"."id" WHERE ("main_c"."b_id" = '' AND "main_c__a"."id" ILIKE '2')

I found that the above patches weren't effective for 0.96, and also did not handle BooleanField correctly with sqlite3, as the value is stored as True/False as opposed to 'on' -- so the generated <name>__iexact = 'on' would not match results correctly. I wrote the following patch for 0.96 that seems to resolve this issue:

Index: manipulators.py
===================================================================
--- manipulators.py
+++ manipulators.py
@@ -278,6 +278,7 @@

 def manipulator_validator_unique_together(field_name_list, opts, self, field_data, all_data):
     from django.db.models.fields.related import ManyToOneRel
+    from django.db.models.fields import BooleanField
     from django.utils.text import get_text_list
     field_list = [opts.get_field(field_name) for field_name in field_name_list]
     if isinstance(field_list[0].rel, ManyToOneRel):
@@ -294,7 +295,15 @@
             # doesn't have blank=True.
             return
         if isinstance(f.rel, ManyToOneRel):
-            kwargs['%s__pk' % f.name] = field_val
+            if field_val is '':
+                kwargs['%s__isnull' % f.name] = True
+            else:
+                kwargs['%s__pk' % f.name] = field_val
+        elif isinstance(f, BooleanField):
+            if field_val in ('on', 'True', 'true', True):
+               kwargs['%s' % f.name] = True
+            else:
+               kwargs['%s' % f.name] = False
         else:
             kwargs['%s__iexact' % f.name] = field_val
     try:

I've tried the above scenarios and the only way I am currently able to produce an IntegrityError with unique_together FKs is by adding an actual duplicate -- but this is the issue in #8209.

I'm going to look into #8209 and then see if this is resolved..

I cannot reproduce this one so I'll release it. Perhaps someone else can resolve?

I suspect that this is no longer a problem, especially since manipulators are no longer used and about to be removed. Can anyone verify that this still is a problem?

Agreed. The impending removal of manipulators (part of #7742) would seem to make this fixed through attrition.