🔧 Add types for DefaultFieldsAttributeDict subclasses

[chrisjsewell]

Improves type checking and LSP completion

(in some sense this a non-breaking alternative to changing them to dataclasses)

[sphuber]

Thanks. How breaking would it be if we change all these extendeddicts variants to dataclasses? I don't think that plugins really rely on the "additional" functionality of some of them, such as the defaultkeys and extrakeys methods of the DefaultFieldsAttributeDict class. I think we should start the process of deprecating these with the intention to replace them with dataclasses. And then we decide when we actually remove them. I can do some tests with aiida-quantumespresso to see if anything would be affected, but I think very little, if anything should break. What do you think?

[chrisjsewell]

What do you think?

Yeh I think plugins could break; mainly scheduler plugins though, which obviously won't be checked by aiida-quantumespresso. Also serialisation to checkpoints is the other consideration.

I think let's merge this as a quick incremental step, then it shouldn't be to hard to try converting to dataclasses

[sphuber]

mainly scheduler plugins though, which obviously won't be checked by aiida-quantumespresso.

Yeah fair enough, but there we will have the scheduler plugins in aiida-core to get a sense of impact.

Also serialisation to checkpoints is the other consideration.

Our checkpoint serialization already has a representer for dataclasses, so that should be fine. Of course if people upgrade when they have unfinished processes, then when they restart and they get deserialized, that might cause problems. But that has never been supported. We explicitly tell people to finish any running processes before upgrading.

That reminds me that we cannot deprecate/remove AttributeDict probably as that is actively used a lot, for example by the Process.inputs and WorkChain.ctx. So we might have to keep that one around. But that is ok. It is mostly the other derivative FixedFieldsAttributeDict and DefaultFieldsAttributeDict that we want to get rid off.