[POC] "Back to databag" (i.e. `dict` representation of Abstract Relation Data

[juditnovak]

See how it works on ZooKeeper: https://github.com/canonical/zookeeper-operator/pull/122

If the concept is welcome, I'm adding the tests

[juditnovak]

Summary of the offline conversation with @welpaolo :

• We evaluated what if we use direct inheritace from dict for *Data classes. i.e. no *Data.as_dict() method (as the POC has it at the time this comment is written). We agreed that it seems a good idea, as this allows for instantiating the object directly, instead of requiring a second object to benefit from dict access (i.e. X instance, Y = X.as_dict()).
• However, the above means a bigger, conceptual decision. I.e. that Python Standard Library dict would become our confirmed interface for Abstract Relation Data.
  • @welpaolo had once concern: In case Python Standard Lib introduces a (breaking) change, we must follow -- and/or we might need to pin Python versions over the lib
    • Though the above is a valid concern, in my opinion if dict has a (breaking) change, it's likely to have a strong impact on the rest of our code anyway. So adjusting Relation Data may just be part of handling the problem overall.
    • Since no matter what dict IS a key-value mapping, it should ever correspond to the representation of Relation Data from our perspective. Thus overall it should be convenient to be used as an interface.

[juditnovak]

@welpaolo I was an idiot. (Let's hope it's past tense :rofl: )

We need two separate objects.

Data (here still called DataRelation) is bound to a relation-name.

While the dict is representing a specific relation within that "name".

So we can't do better than as_dict(relation_id) :-/

Sorry I was thinking on a wrong track :-/