The following behaviour when copying a dask array created using from_array and then computing the result isn't what I'd expect:

import dask.array as da
import numpy as np

d = da.from_array(np.array([0, 1]), chunks=2,)
dc = d.copy()

d is dc
False
d.compute() is dc.compute()
True

A dask array not created using from_array does not show this behaviour:

d1 = da.ones(10, chunks=2,)

d1c = d1.copy()
d1 is d1c
False
d1.compute() is d1c.compute()
False

For reference, this behaviour seems to be tripping up an Iris test, resulting in this test failure.

I'm using Python 3.6.5 and dask 0.18.1 (this behaviour doesn't appear in v0.18.0).