Copying a cube with lazy data currently isn't behaving as expected:

from iris.cube import Cube
from iris._lazy_data import as_lazy_data
import numpy as np

c = Cube(as_lazy_data(np.array([0, 1]), chunks=2,))
cc = c.copy()
c is cc
False
c.data is cc.data
True

There's current a failing test on Travis because of this behaviour (it's this test that's failing).

I think we're inheriting this behaviour from dask 0.18.1, so a quick workaround for this is to pin dask to v0.18.0, which doesn't evidence this behaviour.