[FIX] models: fix behaviour of 'defer_parent_store_computation'

If a model uses nested sets to maintain a hierarchical record
ordering, then bulk inserts (e.g. via a CSV import) are likely to
suffer from an O(n^2) performance penalty, since there is a high
chance that each new record will require many of the existing records'
parent_left and parent_right fields to be updated.

The context parameter 'defer_parent_store_computation' may be passed
to the load() method, in order to defer updating the parent_left and
parent_right fields until after the bulk operation is complete.  This
appears to have last been fully supported in version 7 (via the old
import_data() method).

It is still possible to specify 'defer_parent_store_computation',
e.g. via an XML-RPC call such as

  m.execute_kw(dbname, uid, password, 'stock.location', 'load', [],
               {'fields': fields, 'data': data,
                'context': {'defer_parent_store_computation': True}})

This will currently defer the recalculation of parent_left and
parent_right, but does not correctly call _parent_store_compute()
after the bulk operation is complete.

Add the missing call to _parent_store_compute() to perform a one-time
update of the parent_left and parent_right fields if needed after the
import operation is complete.

Also fix the child_of_domain() and parent_of_domain() expression
methods, so that the 'child_of' and 'parent_of' operators give correct
answers when updates to parent_left and parent_right have been
deferred.

Signed-off-by: Michael Brown <mbrown@fensystems.co.uk>

Closes #13688 (part 2/2)

openerp/models.py

@@ -1040,6 +1040,8 @@ class BaseModel(object):
         :param dict context:
         :returns: {ids: list(int)|False, messages: [Message]}
         """
+        if context is None:
+            context = {}
         cr.execute('SAVEPOINT model_load')
         messages = []
 
@@ -1094,6 +1096,10 @@ class BaseModel(object):
         if any(message['type'] == 'error' for message in messages):
             cr.execute('ROLLBACK TO SAVEPOINT model_load')
             ids = False
+
+        if ids and context.get('defer_parent_store_computation'):
+            self._parent_store_compute(cr)
+
         return {'ids': ids, 'messages': messages}
 
     def _add_fake_fields(self, cr, uid, fields, context=None):

openerp/osv/expression.py

@@ -711,9 +711,11 @@ class expression(object):
             """ Return a domain implementing the child_of operator for [(left,child_of,ids)],
                 either as a range using the parent_left/right tree lookup fields
                 (when available), or as an expanded [(left,in,child_ids)] """
+            if context is None:
+                context = {}
             if not ids:
                 return FALSE_DOMAIN
-            if left_model._parent_store and (not left_model.pool._init):
+            if left_model._parent_store and (not left_model.pool._init) and (not context.get('defer_parent_store_computation')):
                 # TODO: Improve where joins are implemented for many with '.', replace by:
                 # doms += ['&',(prefix+'.parent_left','<',o.parent_right),(prefix+'.parent_left','>=',o.parent_left)]
                 doms = []
@@ -736,7 +738,9 @@ class expression(object):
             """ Return a domain implementing the parent_of operator for [(left,parent_of,ids)],
                 either as a range using the parent_left/right tree lookup fields
                 (when available), or as an expanded [(left,in,parent_ids)] """
-            if left_model._parent_store and (not left_model.pool._init):
+            if context is None:
+                context = {}
+            if left_model._parent_store and (not left_model.pool._init) and (not context.get('defer_parent_store_computation')):
                 doms = []
                 for node in left_model.browse(cr, uid, ids, context=context):
                     if doms: