[Experiment] ViewState opportunities suggestions

[tgrapperon]

Hello!
I was wondering yesterday if we couldn't improve the ViewState story in TCA, that is small observable sub-states for ViewStores to work with instead of whole feature's states.
I've made this quick proof of concept that observes each KeyPath access to the ViewStore's state and compare them with what's available in the State. If State offers properties that are not used, it presents a runtime warning where it proposes a ViewState declaration.

For example, activating this by using .lookForViewStateOpportunities() in the Animation case study gives:

Likely overkill in this case, but the `ViewState` declaration is spot-on.

This is of course not a perfect solution:

• There are potentially many false positives caused by direct accesses to viewStore.state and its descendants that are missed and considered unaccessed.
• There are probably cases where the warning could be emitted before an escaped ViewStore accesses the properties. For example in modal/navigation contexts. I've added a small delay to the comparison to give it more time to unfold/evaluate all the body's hierarchy, like in GeometryReaders closures for example.
• This is not working yet with composed KeyPath, for example viewStore.array.count, but I feel it should be possible to match them with more work.
• It's not working well yet with property wrappers. I've envisaged several approaches for the whole feature, but finally opted to exploit custom-dump facilities around KeyPaths, because establishing the link between a KeyPath and its property name is far from trivial. This doesn't work well with property wrappers right now, but it could hopefully be improved for a general purpose. In this case, this experiment could be updated to support @BindableState properties.

This is also not working well yet when already using a ViewState (for example, the one it suggested 😬)!

The code was written in one go, in some hackathon/proof of concept style where I only wanted to see if it could work. There are of course many ways to refactor/reorganize it, or maybe rethink this at a deeper level. It is maybe possible to build the ViewStore's removeDuplicates automatically, but I don't think it would be worth it performance-wise if it ever works.

I'm not pushing it farther yet because of these issues. The ViewState declaration suggestion seems very enticing though.
What do you think about it?

[stephencelis]

@tgrapperon Thanks for sharing! This looks like a lot of fun! Will try to take it for a spin soon 😄