Fixed #8106 -- Untangled some problems with complex select_related() queries

and models that have multiple paths to them from other models.


git-svn-id: http://code.djangoproject.com/svn/django/trunk@8559 bcc190cf-cafb-0310-a4f2-bffc1f526a37

django/db/models/sql/query.py

@@ -913,7 +913,7 @@ class Query(object):
 
     def fill_related_selections(self, opts=None, root_alias=None, cur_depth=1,
             used=None, requested=None, restricted=None, nullable=None,
-            dupe_set=None):
+            dupe_set=None, avoid_set=None):
         """
         Fill in the information needed for a select_related query. The current
         depth is measured as the number of connections away from the root model
@@ -933,8 +933,9 @@ class Query(object):
             used = set()
         if dupe_set is None:
             dupe_set = set()
+        if avoid_set is None:
+            avoid_set = set()
         orig_dupe_set = dupe_set
-        orig_used = used
 
         # Setup for the case when only particular related fields should be
         # included in the related selection.
@@ -948,8 +949,12 @@ class Query(object):
         for f, model in opts.get_fields_with_model():
             if not select_related_descend(f, restricted, requested):
                 continue
+            # The "avoid" set is aliases we want to avoid just for this
+            # particular branch of the recursion. They aren't permanently
+            # forbidden from reuse in the related selection tables (which is
+            # what "used" specifies).
+            avoid = avoid_set.copy()
             dupe_set = orig_dupe_set.copy()
-            used = orig_used.copy()
             table = f.rel.to._meta.db_table
             if nullable or f.null:
                 promote = True
@@ -962,7 +967,7 @@ class Query(object):
                     lhs_col = int_opts.parents[int_model].column
                     dedupe = lhs_col in opts.duplicate_targets
                     if dedupe:
-                        used.update(self.dupe_avoidance.get(id(opts), lhs_col),
+                        avoid.update(self.dupe_avoidance.get(id(opts), lhs_col),
                                 ())
                         dupe_set.add((opts, lhs_col))
                     int_opts = int_model._meta
@@ -976,13 +981,13 @@ class Query(object):
 
             dedupe = f.column in opts.duplicate_targets
             if dupe_set or dedupe:
-                used.update(self.dupe_avoidance.get((id(opts), f.column), ()))
+                avoid.update(self.dupe_avoidance.get((id(opts), f.column), ()))
                 if dedupe:
                     dupe_set.add((opts, f.column))
 
             alias = self.join((alias, table, f.column,
-                    f.rel.get_related_field().column), exclusions=used,
+                    f.rel.get_related_field().column),
-                    promote=promote)
+                    exclusions=used.union(avoid), promote=promote)
             used.add(alias)
             self.related_select_cols.extend(self.get_default_columns(
                 start_alias=alias, opts=f.rel.to._meta, as_pairs=True)[0])
@@ -998,7 +1003,7 @@ class Query(object):
             for dupe_opts, dupe_col in dupe_set:
                 self.update_dupe_avoidance(dupe_opts, dupe_col, alias)
             self.fill_related_selections(f.rel.to._meta, alias, cur_depth + 1,
-                    used, next, restricted, new_nullable, dupe_set)
+                    used, next, restricted, new_nullable, dupe_set, avoid)
 
     def add_filter(self, filter_expr, connector=AND, negate=False, trim=False,
             can_reuse=None):

tests/regressiontests/select_related_regress/models.py

@@ -28,11 +28,35 @@ class Connection(models.Model):
     def __unicode__(self):
         return u"%s to %s" % (self.start, self.end)
 
+# Another non-tree hierarchy that exercises code paths similar to the above
+# example, but in a slightly different configuration.
+class TUser(models.Model):
+    name = models.CharField(max_length=200)
+
+class Person(models.Model):
+    user = models.ForeignKey(TUser, unique=True)
+
+class Organizer(models.Model):
+    person = models.ForeignKey(Person)
+
+class Student(models.Model):
+    person = models.ForeignKey(Person)
+
+class Class(models.Model):
+    org = models.ForeignKey(Organizer)
+
+class Enrollment(models.Model):
+    std = models.ForeignKey(Student)
+    cls = models.ForeignKey(Class)
+
 __test__ = {'API_TESTS': """
 Regression test for bug #7110. When using select_related(), we must query the
 Device and Building tables using two different aliases (each) in order to
 differentiate the start and end Connection fields. The net result is that both
-the "connections = ..." queries here should give the same results.
+the "connections = ..." queries here should give the same results without
+pulling in more than the absolute minimum number of tables (history has
+shown that it's easy to make a mistake in the implementation and include some
+unnecessary bonus joins).
 
 >>> b=Building.objects.create(name='101')
 >>> dev1=Device.objects.create(name="router", building=b)
@@ -57,4 +81,23 @@ the "connections = ..." queries here should give the same results.
 >>> connections.query.count_active_tables()
 7
 
+Regression test for bug #8106. Same sort of problem as the previous test, but
+this time there are more extra tables to pull in as part of the
+select_related() and some of them could potentially clash (so need to be kept
+separate).
+
+>>> us = TUser.objects.create(name="std")
+>>> usp = Person.objects.create(user=us)
+>>> uo = TUser.objects.create(name="org")
+>>> uop = Person.objects.create(user=uo)
+>>> s = Student.objects.create(person = usp)
+>>> o = Organizer.objects.create(person = uop)
+>>> c = Class.objects.create(org=o)
+>>> e = Enrollment.objects.create(std=s, cls=c)
+
+>>> e_related = Enrollment.objects.all().select_related()[0]
+>>> e_related.std.person.user.name
+u"std"
+>>> e_related.cls.org.person.user.name
+u"org"
 """}