Fix cube equality for ancillary variables

[stephenworsley]

A subtask for #3422. Fixing this first should make testing for the rest of #3422 easier.

Cube equality (__eq__) is ignoring AVs (and maybe CellMethods) but not AuxCoords.

---

Acceptance Criteria

• Fix cube equality
• Write tests for equality
• Raise issues for any failing tests (tests may be skipped until a further pull request)

[bjlittle]

@stephenworsley Just checking... is this also applicable to iris.coords.CellMeasures?

[pp-mo]

Just about the naming of this issue ... It seems to be mixture of auxiliary and ancillary ?
It's confusing enough already !!

[stephenworsley]

@bjlittle I believe so.

[bjlittle]

@SciTools/iris-devs Are we proposing that iris.cube.Cube equality includes comparing the associated AncillaryVariable and CellMeasure metadata and data?

Just guessing, but that sounds mighty expensive, particularly for large cubes 💣

Kinda glad that #3526 has landed...

[pp-mo]

sounds mighty expensive

But we are (already) doing it for coords, including possibly huge multidimensional AuxCoords,
so where's the difference ??

[bjlittle]

so where's the difference ??

@pp-mo The difference is that we could also potentially be comparing between 1..N AncillaryVariable data payloads of the same dimensionality as the iris.cube.Cube.data, in addition to the iris.cube.Cube.data, plus perhaps a CellMeasure, and then as you mentioned possibly huge multidimensional AuxCoords 🤒

I'm just raising the concern that this just might not scale well, and be not at all practical or performant for the user. Perhaps we should consider options for the user with large data and large metadata, rather than offering Hobson's choice Cube equality; where absolutely everything is always compared.

Just sayin'...

[bjlittle]

@stephenworsley BTW Cube.__eq__ uses the Cube.metadata during equality, which includes Cube.cell_methods, see here and here. HTH, so I guess you should change the description of this issue to align with that for clarity.

The functionality that's missing for Cube equality is only:

• CellMeasure, and
• AncillaryVariable

... as I understand it 😄

[stephenworsley]

Another thing worth fixing, it seems like if a pair of cell methods are added in a different order to two otherwise equal cubes, they are evaluated as not equal. See #3530.

[stephenworsley]

I have detailed some of my thoughts on the direction we could take this in #3536.