Cube merge scalar coordinate signature comparison

[bjlittle]

Require to review the current light weight scalar coordinate signature comparison between proto-cubes within cube merge e.g.

>>> import numpy as np
>>> import iris
>>> from iris.coords import DimCoord, AuxCoord
>>> cube = iris.cube.Cube(np.arange(12).reshape(3, 4), standard_name='air_temperature', units='K')
>>> cube.add_dim_coord(DimCoord(np.arange(3), standard_name='latitude', units='degrees'), 0)
>>> cube.add_dim_coord(DimCoord(np.arange(4), standard_name='longitude', units='degrees'), 1)
>>> other = cube.copy()
>>> cube.add_aux_coord(AuxCoord(np.array([10], dtype=np.int), standard_name='height', units='m'))
>>> other.add_aux_coord(AuxCoord([20.], bounds=[[15., 25.]], standard_name='height', units='m'))
>>> merged = iris.cube.CubeList([cube, other]).merge()[0]
>>> print merged
air_temperature / (K)               (height: 2; latitude: 3; longitude: 4)
     Dimension coordinates:
          height                           x            -             -
          latitude                         -            x             -
          longitude                        -            -             x
>>> merged.coord('height')
DimCoord(array([ 10.,  20.]), standard_name='height', units=Unit('m'))

Ref: #475

[rhattersley]

Thanks for creating this @bjlittle and adding a nice concrete example.

For future reference, the problem is merge is not paying attention to:

1. the dtype of scalar,
2. whether the scalar has bounds.

Are there any other facets?

[bjlittle]

Apart from the cell point and bounds dtype, the meta-data associated with a scalar coordinate also contains the circular flag for DimCoord scalar coordinates. When the scalar coordinate is an AuxCoord clearly this doesn't apply.

With regards to string scalar coordinates we'd want to decide whether we force all similar string scalar coords to have the same length or not - at the moment merge will calculate the maximum similar scalar coord string length, which seems pretty flexible (and appropriate). Whereas we request that all extensive (vector) string auxiliary coordinates have the exact same length - which is okay by me.

Might want to consider the case when similar scalar coordinates have mixed types i.e. one cube has a height DimCoord scalar coordinate, whereas another has a height AuxCoord scalar coordinate. Should these be considered compatible? Dunno, but something else to ponder when head scratching over these collective issues.

Ummm think that's it ... for now.

[pelson]

Closing due to inactivity.