Isolated BandsData will raise exception when running verdi data bands list

[unkcpz]

bands = load_node(<bands_node>)
bands_clone = bands.clone()
bands_clone.store()

will cause verdi data bands list break.

File "/home/unkcpz/pyProject/aiida_core/aiida/backends/djsite/queries.py", line 178, in get_bands_and_parents_structure
    struc_pks = [structure_dict[pk] for pk in pks]
File "/home/unkcpz/pyProject/aiida_core/aiida/backends/djsite/queries.py", line 178, in <listcomp>
    struc_pks = [structure_dict[pk] for pk in pks]
KeyError: 230

Although I have no idea whether it is reasonable to have BandsData node without structure as its ancestor, this is still a case to consider. but I don't know which of the following measures is more reasonable

1. forbid users from creating BandsData instance without set a StructureData as its ancestor node.
2. when running verdi data bands list show None as its formula value.

I am in favor of the second personally.

[ltalirz]

Thanks @unkcpz for pointing this out. I agree with your assessment and would vote for 2.

Mentioning @giovannipizzi and @sphuber for comment.

[greschd]

Agree with option 2, I'm fairly certain I've created BandsData without associated StructureData in some places.

[sphuber]

This really depends on the original design of BandsData. If it really needs an associated StructureData then option 2 would be a bit of a workaround. But given that one can already created a separate BandsData object, interfaces should not except, so I would also vote for option 2

[ltalirz]

Thanks for the input - @unkcpz this is the "go" signal ;-)

[unkcpz]

I make a PR. but one thing need to be discussed, when a bands object is cloned, its StructureData should also be cloned? Maybe this behavior is what users expected when calling the .clone() function.

[greschd]

I don't think it should clone the StructureData. Looking at the code, the "linked structure" is anyway just determined through a query for the closest ancestor, and .clone() definitely shouldn't do any node linking - the graph relationships of the clone should be up to the calling code (e.g. in the caching code).

That makes me wonder, though, if that linked structure is always the right one. It looks for the closest parent, but for example if a calculation creates both structure and bands in one step, the correct structure might be a sibling node.