Fix lib/iris/tests/unit/data_manager/test_DataManager.py tests

[lbdreyer]

Fix all tests in lib/iris/tests/unit/data_manager/test_DataManager.py so that they all pass.

I have had to remove/change some tests that used the old API (e.g. cube.fill_value).
I have also had to make some fixes to handling of masked constants which hopefully makes sense

[lbdreyer]

I am putting this back as this ensures you are doing a deepcopy with a data of the correct shape.
Without this you can do:

>>> dm = DataManager(np.array([0,1,2]))
>>> dm._deepcopy(dict(), data=np.array(0))
DataManager(array(0))

Which shouldn't happen

[bjlittle]

@lbdreyer Doing this promotes the dimensionality of the data e.g.

>>> a = ma.masked_array([1, 2, 3], mask=[0, 1, 0])
>>> v = a[1]
>>> v
masked
>>> r = ma.array([v.data], mask=v.mask)
>>> r
masked_array(data = [--],
             mask = [ True],
       fill_value = 1e+20)

>>> r.shape
(1,)
>>> r.ndim
1
>>> r = ma.array(v.data, mask=v.mask)
>>> r
masked_array(data = --,
             mask = True,
       fill_value = 1e+20)

>>> r.shape
()
>>> r.ndim
0

[lbdreyer]

np.asanyarray doesn't work

>>> a = ma.masked_array([8], mask=True)
>>> masked_constant = a[0]
>>> print type(masked_constant)
<class 'numpy.ma.core.MaskedConstant'>
>>> print type(np.asanyarray(masked_constant))
<class 'numpy.ma.core.MaskedConstant'>

[lbdreyer]

I also can't do

>>> a = ma.masked_array(8, mask=True)
>>> print a[0]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/sci/lib/python2.7/site-packages/numpy/ma/core.py", line 2992, in __getitem__
    dout = ndarray.__getitem__(_data, indx)
IndexError: 0-d arrays can't be indexed

[bjlittle]

@lbdreyer Perhaps my point was lost in the example, but shouldn't line 97 be

data = ma.masked_array(data, mask=data.mask)

instead of

data = ma.masked_array([data], mask=data.mask)

[lbdreyer]

Sorry my last comment makes no sense, that's actually relevant to the tests.

Can we change this line to be
data = ma.masked_array(data, mask=data.mask)
?

[bjlittle]

Yup, exactly!

[lbdreyer]

Nope

>>> orig = ma.masked_array([1,2,3], mask=[True, False, False])
>>> m_con = orig[0]
>>> data = ma.masked_array(m_con, mask=m_con.mask)
>>> print type(data)
<class 'numpy.ma.core.MaskedConstant'>

😿

[bjlittle]

Apologies, that should be

data = ma.masked_array(data.data, mask=data.mask)

so for your example above it's m_con.data

[lbdreyer]

Ah! You're right! That does indeed work.

I have now put up a new commit with that change in