Fixed #14056 -- Made sure LEFT JOIN aren't trimmed in ORDER BY

If LEFT JOINs are required for correct results, then trimming the join
can lead to incorrect results. Consider case:

TBL A: ID | TBL B: ID  A_ID
       1           1   1
       2
Now A.order_by('b__a') did use a join to B, and B's a_id column. This
was seen to contain the same value as A's id, and so the join was
trimmed. But this wasn't correct as the join is LEFT JOIN, and for row
A.id = 2 the B.a_id column is NULL.

django/db/models/sql/compiler.py

@@ -319,10 +319,10 @@ class SQLCompiler(object):
 
         for name in self.query.distinct_fields:
             parts = name.split(LOOKUP_SEP)
-            field, cols, alias, _, _ = self._setup_joins(parts, opts, None)
-            cols, alias = self._final_join_removal(cols, alias)
-            for col in cols:
-                result.append("%s.%s" % (qn(alias), qn2(col)))
+            _, targets, alias, joins, path, _ = self._setup_joins(parts, opts, None)
+            targets, alias, _ = self.query.trim_joins(targets, joins, path)
+            for target in targets:
+                result.append("%s.%s" % (qn(alias), qn2(target.column)))
         return result
 
     def get_ordering(self):
@@ -421,7 +421,7 @@ class SQLCompiler(object):
         return result, params, group_by
 
     def find_ordering_name(self, name, opts, alias=None, default_order='ASC',
-            already_seen=None):
+                           already_seen=None):
         """
         Returns the table alias (the name might be ambiguous, the alias will
         not be) and column name for ordering by the given 'name' parameter.
@@ -429,11 +429,11 @@ class SQLCompiler(object):
         """
         name, order = get_order_dir(name, default_order)
         pieces = name.split(LOOKUP_SEP)
-        field, cols, alias, joins, opts = self._setup_joins(pieces, opts, alias)
+        field, targets, alias, joins, path, opts = self._setup_joins(pieces, opts, alias)
 
         # If we get to this point and the field is a relation to another model,
         # append the default ordering for that model.
-        if field.rel and len(joins) > 1 and opts.ordering:
+        if field.rel and path and opts.ordering:
             # Firstly, avoid infinite loops.
             if not already_seen:
                 already_seen = set()
@@ -445,10 +445,10 @@ class SQLCompiler(object):
             results = []
             for item in opts.ordering:
                 results.extend(self.find_ordering_name(item, opts, alias,
-                        order, already_seen))
+                                                       order, already_seen))
             return results
-        cols, alias = self._final_join_removal(cols, alias)
-        return [(alias, cols, order)]
+        targets, alias, _ = self.query.trim_joins(targets, joins, path)
+        return [(alias, [t.column for t in targets], order)]
 
     def _setup_joins(self, pieces, opts, alias):
         """
@@ -461,13 +461,12 @@ class SQLCompiler(object):
         """
         if not alias:
             alias = self.query.get_initial_alias()
-        field, targets, opts, joins, _ = self.query.setup_joins(
+        field, targets, opts, joins, path = self.query.setup_joins(
             pieces, opts, alias)
         # We will later on need to promote those joins that were added to the
         # query afresh above.
         joins_to_promote = [j for j in joins if self.query.alias_refcount[j] < 2]
         alias = joins[-1]
-        cols = [target.column for target in targets]
         if not field.rel:
             # To avoid inadvertent trimming of a necessary alias, use the
             # refcount to show that we are referencing a non-relation field on
@@ -478,28 +477,7 @@ class SQLCompiler(object):
         # Ordering or distinct must not affect the returned set, and INNER
         # JOINS for nullable fields could do this.
         self.query.promote_joins(joins_to_promote)
-        return field, cols, alias, joins, opts
-
-    def _final_join_removal(self, cols, alias):
-        """
-        A helper method for get_distinct and get_ordering. This method will
-        trim extra not-needed joins from the tail of the join chain.
-
-        This is very similar to what is done in trim_joins, but we will
-        trim LEFT JOINS here. It would be a good idea to consolidate this
-        method and query.trim_joins().
-        """
-        if alias:
-            while 1:
-                join = self.query.alias_map[alias]
-                lhs_cols, rhs_cols = zip(*[(lhs_col, rhs_col) for lhs_col, rhs_col in join.join_cols])
-                if set(cols) != set(rhs_cols):
-                    break
-
-                cols = [lhs_cols[rhs_cols.index(col)] for col in cols]
-                self.query.unref_alias(alias)
-                alias = join.lhs_alias
-        return cols, alias
+        return field, targets, alias, joins, path, opts
 
     def get_from_clause(self):
         """

tests/queries/tests.py

@@ -25,7 +25,7 @@ from .models import (
     OneToOneCategory, NullableName, ProxyCategory, SingleObject, RelatedObject,
     ModelA, ModelB, ModelC, ModelD, Responsibility, Job, JobResponsibilities,
     BaseA, FK1, Identifier, Program, Channel, Page, Paragraph, Chapter, Book,
-    MyObject, Order, OrderItem)
+    MyObject, Order, OrderItem, SharedConnection)
 
 class BaseQuerysetTest(TestCase):
     def assertValueQuerysetEqual(self, qs, values):
@@ -2977,3 +2977,14 @@ class RelatedLookupTypeTests(TestCase):
         self.assertQuerysetEqual(
             ObjectB.objects.filter(objecta__in=[wrong_type]),
             [ob], lambda x: x)
+
+class Ticket14056Tests(TestCase):
+    def test_ticket_14056(self):
+        s1 = SharedConnection.objects.create(data='s1')
+        s2 = SharedConnection.objects.create(data='s2')
+        s3 = SharedConnection.objects.create(data='s3')
+        PointerA.objects.create(connection=s2)
+        self.assertQuerysetEqual(
+            SharedConnection.objects.order_by('pointera__connection', 'pk'),
+            [s1, s3, s2], lambda x: x
+        )