merge is calling expire on pending/transient instances

[Mike Bayer]

1. InstanceState.expire_attributes should raise if self.key is None

2. InstanceState should gain has_identity and we should lose orm.util._state_has_identity so this check always available

3. ColumnProperty.merge should be checking has_identity before expiring

   #!python from sqlalchemy import * from sqlalchemy.orm import * from sqlalchemy.orm import attributes from sqlalchemy import exc

   metadata = MetaData() foo = Table('foo', metadata, Column('id', Integer, primary_key=True), Column('data', String) ) engine = create_engine('sqlite://', echo=True) metadata.create_all(engine)

   class Foo(object): pass

   mapper(Foo, foo) session = sessionmaker()()

   f1 = Foo() session.add(f1)

   state = attributes.instance_state(f1) attributes.instance_state(f1).expire_attributes(state.dict, 'data')

   try: print f1.data except exc.UnboundExecutionError: # should not be hitting the DB print "raised !"

   session.bind = engine f2 = Foo() f2.id = 2 f2 = session.merge(f2)

   session.bind = None try: f2.data except exc.UnboundExecutionError: # should not be hitting the DB print "raised !"

also see related test in #1681

[Mike Bayer]

6ea6673376609ce6a5e26f9f20425cffee96bcd8. Interestingly, disallowing state.expire_attributes() on a "pending" state is not at all an option here, since expiration occurs within the flush() process for states that have just become persistent, but don't yet have their key assigned. In some cases the loader is even fired off before the flush finishes and hence before state.key is assigned. The key assignment currently happens at the end in most cases in _register_newly_persistent() since we bookkeep for some "key switch" scenarios there after all SQL has been emitted.

[Mike Bayer]

Removing milestone: 0.6.1 (automated comment)