Refactored qs.add_q() and utils/tree.py

The sql/query.py add_q method did a lot of where/having tree hacking to
get complex queries to work correctly. The logic was refactored so that
it should be simpler to understand. The new logic should also produce
leaner WHERE conditions.

The changes cascade somewhat, as some other parts of Django (like
add_filter() and WhereNode) expect boolean trees in certain format or
they fail to work. So to fix the add_q() one must fix utils/tree.py,
some things in add_filter(), WhereNode and so on.

This commit also fixed add_filter to see negate clauses up the path.
A query like .exclude(Q(reversefk__in=a_list)) didn't work similarly to
.filter(~Q(reversefk__in=a_list)). The reason for this is that only
the immediate parent negate clauses were seen by add_filter, and thus a
tree like AND: (NOT AND: (AND: condition)) will not be handled
correctly, as there is one intermediary AND node in the tree. The
example tree is generated by .exclude(~Q(reversefk__in=a_list)).

Still, aggregation lost connectors in OR cases, and F() objects and
aggregates in same filter clause caused GROUP BY problems on some
databases.

Fixed #17600, fixed #13198, fixed #17025, fixed #17000, fixed #11293.

1 files changed, 46 insertions, 66 deletions

diff --git a/django/utils/tree.py b/django/utils/tree.py
index ce490224e0..4152b4600b 100644
--- a/django/utils/tree.py
+++ b/django/utils/tree.py
@@ -19,14 +19,9 @@ class Node(object):
         """
         Constructs a new Node. If no connector is given, the default will be
         used.
-
-        Warning: You probably don't want to pass in the 'negated' parameter. It
-        is NOT the same as constructing a node and calling negate() on the
-        result.
         """
         self.children = children and children[:] or []
         self.connector = connector or self.default
-        self.subtree_parents = []
         self.negated = negated
 
     # We need this because of django.db.models.query_utils.Q. Q. __init__() is
@@ -59,7 +54,6 @@ class Node(object):
         obj = Node(connector=self.connector, negated=self.negated)
         obj.__class__ = self.__class__
         obj.children = copy.deepcopy(self.children, memodict)
-        obj.subtree_parents = copy.deepcopy(self.subtree_parents, memodict)
         return obj
 
     def __len__(self):
@@ -83,74 +77,60 @@ class Node(object):
         """
         return other in self.children
 
-    def add(self, node, conn_type):
+    def _prepare_data(self, data):
         """
-        Adds a new node to the tree. If the conn_type is the same as the root's
-        current connector type, the node is added to the first level.
-        Otherwise, the whole tree is pushed down one level and a new root
-        connector is created, connecting the existing tree and the new node.
+        A subclass hook for doing subclass specific transformations of the
+        given data on combine() or add().
         """
-        if node in self.children and conn_type == self.connector:
-            return
-        if len(self.children) < 2:
-            self.connector = conn_type
-        if self.connector == conn_type:
-            if isinstance(node, Node) and (node.connector == conn_type or
-                    len(node) == 1):
-                self.children.extend(node.children)
-            else:
-                self.children.append(node)
-        else:
-            obj = self._new_instance(self.children, self.connector,
-                    self.negated)
-            self.connector = conn_type
-            self.children = [obj, node]
+        return data
 
-    def negate(self):
+    def add(self, data, conn_type, squash=True):
         """
-        Negate the sense of the root connector. This reorganises the children
-        so that the current node has a single child: a negated node containing
-        all the previous children. This slightly odd construction makes adding
-        new children behave more intuitively.
+        Combines this tree and the data represented by data using the
+        connector conn_type. The combine is done by squashing the node other
+        away if possible.
 
-        Interpreting the meaning of this negate is up to client code. This
-        method is useful for implementing "not" arrangements.
-        """
-        self.children = [self._new_instance(self.children, self.connector,
-                not self.negated)]
-        self.connector = self.default
+        This tree (self) will never be pushed to a child node of the
+        combined tree, nor will the connector or negated properties change.
 
-    def start_subtree(self, conn_type):
-        """
-        Sets up internal state so that new nodes are added to a subtree of the
-        current node. The conn_type specifies how the sub-tree is joined to the
-        existing children.
+        The function returns a node which can be used in place of data
+        regardless if the node other got squashed or not.
+
+        If `squash` is False the data is prepared and added as a child to
+        this tree without further logic.
         """
-        if len(self.children) == 1:
-            self.connector = conn_type
-        elif self.connector != conn_type:
-            self.children = [self._new_instance(self.children, self.connector,
-                    self.negated)]
+        if data in self.children:
+            return data
+        data = self._prepare_data(data)
+        if not squash:
+            self.children.append(data)
+            return data
+        if self.connector == conn_type:
+            # We can reuse self.children to append or squash the node other.
+            if (isinstance(data, Node) and not data.negated
+                    and (data.connector == conn_type or len(data) == 1)):
+                # We can squash the other node's children directly into this
+                # node. We are just doing (AB)(CD) == (ABCD) here, with the
+                # addition that if the length of the other node is 1 the
+                # connector doesn't matter. However, for the len(self) == 1
+                # case we don't want to do the squashing, as it would alter
+                # self.connector.
+                self.children.extend(data.children)
+                return self
+            else:
+                # We could use perhaps additional logic here to see if some
+                # children could be used for pushdown here.
+                self.children.append(data)
+                return data
+        else:
+            obj = self._new_instance(self.children, self.connector,
+                                     self.negated)
             self.connector = conn_type
-            self.negated = False
-
-        self.subtree_parents.append(self.__class__(self.children,
-                self.connector, self.negated))
-        self.connector = self.default
-        self.negated = False
-        self.children = []
+            self.children = [obj, data]
+            return data
 
-    def end_subtree(self):
+    def negate(self):
         """
-        Closes off the most recently unmatched start_subtree() call.
-
-        This puts the current state into a node of the parent tree and returns
-        the current instances state to be the parent.
+        Negate the sense of the root connector.
         """
-        obj = self.subtree_parents.pop()
-        node = self.__class__(self.children, self.connector)
-        self.connector = obj.connector
-        self.negated = obj.negated
-        self.children = obj.children
-        self.children.append(node)
-
+        self.negated = not self.negated