cube printout incorrectly displays a masked scalar coord

[lbdreyer]

🐛 Bug Report

My cube has a scalar coordinate with a masked point. When I print out the coordinate, it displays the point as masked (i.e. --) as expected, but if I print the whole cube, it displays the underlaying data under the mask.

I have only tested this with scalar coordinates but it may also be applicable to cell measure/ancillary variables.

How To Reproduce

>>> from iris.cube import Cube
>>> from iris.coords import AuxCoord
>>> import numpy.ma as ma
>>> cube = Cube([1])
>>> coord = AuxCoord(ma.masked, 'height')
>>> cube.add_aux_coord(coord)
>>> print(cube.coord('height'))
AuxCoord :  height / (unknown)
    points: [--]
    shape: (1,)
    dtype: float64
    standard_name: 'height'
>>> print(cube)
unknown / (unknown)                 (-- : 1)
    Scalar coordinates:
        height                      0.0

In the above you can see that when I just print the height coordinate it is displayed as -- but printing the cube displays it as 0

Expected behaviour

I would expect that printing the above would return:

>>> print(cube)
unknown / (unknown)                 (-- : 1)
    Scalar coordinates:
        height                      --

And the same goes for scalar cell measures and ancillary variables

[pp-mo]

I thought this must lie in the cube_summary or cube_printout code.

But actually it seems the culprit is the use of np.array here in the Coord.cell() code

Ultimately, that's probably because

>>> print(np.asanyarray(np.ma.masked_array([2], mask=[1])))
[--]
>>> 
>>> print(np.array(np.ma.masked_array([2], mask=[1])))
[2]

This should probably be replaced with an asanyarray, but it's a bit awkward as that does not provide the ndmin keyword used in the above code link.

A search for other related usages in this code is certainly called for, e.g. in Coord.cells() (plural!).
But in fact it seems that .cells() is not equivalent to .cell() in these cases ...

>>> mco = AuxCoord(np.ma.masked_array([2], mask=[1]))
>>> print(mco.points)
[--]
>>> np.array(mco.points)
array([2])
>>>
>>> next(mco.cells())
Cell(point=masked, bound=None)
>>> mco.cell(0)
Cell(point=0.0, bound=None)
>>> 

Note also : in the last example, the spurious un-masked value is 0.0, not 2.
So, it's not just the use of np.array, and something else weird is happening here !!