Reference can return a cached invalidated object if the local key is the primary key

I do a remove() through a Reference. After doing the remove, the deleted row reappears in the Reference, even when I refetch the object containing the Reference. A test program is below. I would expect it to print None twice, showing that both References are empty.

I tried various workaround for this, including calling .autoreload() and .invalidate() on both the object holding the Reference and the Reference itself. (Workarounds are inserted at "#***" in the test program.) The problem appears to be with store._alive. The only workarounds that work are:

    store._alive.clear()
and
    store.reset()

which are fairly draconian.

I've duplicated this on 0.14 and 0.15, with both sqlite and postgresql.

#-----------------------------------------------------------------
from storm.locals import *

class T1(Storm):
   __storm_table__ = 't1'
   id = Int(primary=True)
   t2 = Reference(id, 'T2.t1_id')

   def __init__(self, id):
       self.id = id

class T2(Storm):
   __storm_table__ = 't2'
   t1_id = Int(primary=True) # foreign key referencing T1

   def __init__(self, t1_id):
       self.t1_id = t1_id

db = create_database('sqlite:')
store = Store(db)

store.execute('create temporary table t1 (id INTEGER PRIMARY KEY)')
store.execute('create temporary table t2 (t1_id INTEGER PRIMARY KEY)')

# Store two rows in T1 and two corresponding rows in T2
t1a = T1(33); store.add(t1a)
t1b = T1(66); store.add(t1b)
t2a = T2(33); store.add(t2a)
t2b = T2(66); store.add(t2b)
store.commit()

# Now remove both T2 rows, but in two different ways.
store.remove(t2a) # removing t2a directly
store.remove(t1b.t2) # remove t2b indirectly, via the Reference
store.commit()

#*** Tried various workarounds at this point (see above).

# Confirm there are no T2 rows now.
print "# of T2 rows:", store.find(T2).count()

# Get the T1 objects again.
t1a_new = store.find(T1, T1.id == 33).one()
t1b_new = store.find(T1, T1.id == 66).one()

# Shouldn't these both print None, because both T2 rows have been deleted?
print "t1a_new.t2:", t1a_new.t2
print "t1b_new.t2:", t1b_new.t2