Model consistency, serialisation and methods

[felix-hilden]

Currently a dataclass behaves like this:

• asdict: strip dataclass but keep other classes, like Timestamp and enums
• repr: Dataclass representation (class name, fields and values recursively)
• str: JSON encode all types

We also have a ModelList to allow string dataclasses. But this is not really comprehensive. It would be nice to include the behavior to all responses, like the tuple of results returned from search. This could be achieved by making all return types share the repr/str structure and using those types as well when parsing inner lists. We could also provide an asdict equivalent for lists like aslist (maybe something else), or call both asbuiltin or something like that.

[felix-hilden]

When printing models, str is naturally called. Should it produce the JSON representation? The answer could be no, and we could provide a json method. str(tracks) is far more ambiguous than tracks.json(), and then print(tracks) could produce the readable representation.

[felix-hilden]

We now have a readable repr for models (not lists). I experimented a bit with providing a similar repr for ModelList, but it seems that pretty printing uses the repr of lists, which breaks the whole thing. So the lists would need to be converted back to builtin ones in asdict. As far as I know, there is no way to do that in dataclasses, but attrs might.

I think we should set the thing up like this:

• Endpoints that would return tuples or lists return a ModelList and inner lists are parsed to ModelLists
• Both models and lists have their repr and str be a readable summary of the fields much like now
• Both should define asbuiltin (or similar) that parses dataclasses to dictionaries and ModelLists to builtin lists, enums and timestamps are preserved
• Both should define json (or similar) to serialise like str does now
• Both should define pprint to use the builtin representation to view the entire contents of the object

Enums and timestamps could even be converted back to strings. And now that pprint is essentially replaced by the repr it could be discarded.

[felix-hilden]

Also, serialise should be moved under model. There is no reason to have it as a separate module, because it shouldn't really be imported or used.

[felix-hilden]

I think we should not preserve timestamps and enums when converting the models. They are technically available outside Tekore, but strings are so much more common and straightforward that we'll use them.