Check collection less than two items remaining before firing scalar backref remove

Fixed long-standing issue where duplicate collection members would cause a
backref to delete the association between the member and its parent object
when one of the duplicates were removed, as occurs as a side effect of
swapping two objects in one statement.

Fixes: #1103
Change-Id: Ic12877f7bd5a4eb688091725a78410748e7fdf16

2 files changed, 32 insertions, 1 deletions

diff --git a/lib/sqlalchemy/util/__init__.py b/lib/sqlalchemy/util/__init__.py
index 9229d0797..d8c28d6af 100644
--- a/lib/sqlalchemy/util/__init__.py
+++ b/lib/sqlalchemy/util/__init__.py
@@ -21,7 +21,8 @@ from ._collections import KeyedTuple, ImmutableContainer, immutabledict, \
     UniqueAppender, PopulateDict, EMPTY_SET, to_list, to_set, \
     to_column_set, update_copy, flatten_iterator, has_intersection, \
     LRUCache, ScopedRegistry, ThreadLocalRegistry, WeakSequence, \
-    coerce_generator_arg, lightweight_named_tuple, collections_abc
+    coerce_generator_arg, lightweight_named_tuple, collections_abc, \
+    has_dupes
 
 from .langhelpers import iterate_attributes, class_hierarchy, \
     portable_instancemethod, unbound_method_to_callable, \
diff --git a/lib/sqlalchemy/util/_collections.py b/lib/sqlalchemy/util/_collections.py
index 3152458bf..43440134a 100644
--- a/lib/sqlalchemy/util/_collections.py
+++ b/lib/sqlalchemy/util/_collections.py
@@ -1057,3 +1057,33 @@ def _iter_id(iterable):
 
     for item in iterable:
         yield id(item), item
+
+
+def has_dupes(sequence, target):
+    """Given a sequence and search object, return True if there's more
+    than one, False if zero or one of them.
+
+
+    """
+    # compare to .index version below, this version introduces less function
+    # overhead and is usually the same speed.  At 15000 items (way bigger than
+    # a relationship-bound collection in memory usually is) it begins to
+    # fall behind the other version only by microseconds.
+    c = 0
+    for item in sequence:
+        if item is target:
+            c += 1
+            if c > 1:
+                return True
+    return False
+
+# .index version.  the two __contains__ calls as well
+# as .index() and isinstance() slow this down.
+# def has_dupes(sequence, target):
+#    if target not in sequence:
+#        return False
+#    elif not isinstance(sequence, collections_abc.Sequence):
+#        return False
+#
+#    idx = sequence.index(target)
+#    return target in sequence[idx + 1:]