RFC: Allow fetching to be forced to true by exchanges

[kitten]

This allows OperationResult to contain fetching: true which would force the useQuery hook to keep fetching: true although data has already been served by an exchange.

This is useful to indicate that more data is being fetched during cache-and-network. Hence the legacy fetching flag could still be reproduced by doing oldFetching = fetching && !data.

[JoviDeCroock]

Personally I would really like stale (cache-and-network refetch) and partial (partial return + fetch) flags.

With fetching going to true the user can get confused because data is returned but fetching is true, this introduces a new case in many applications. This bypasses the point of cache-and-network being a data refresh when we want to show stale data. Many people will have if (fetching) return <Spinner /> in their code at this point.

I've discussed the point of cache-and-network above the following will elaborate on partial.
Partial data introducing fetching is also a bit hit and miss in my opinion since a GQL schema should be closer to what he user really sees so something nullable should also be nullable in the front-end I can't really see why that should introduce fetching.

TLDR: this is breaking

[kitten]

Yes, that's why this is still an RFC. I think we mainly have to validate:

• is this breaking? (If we use fetching: true, then yes)
• should this be a separate state, i.e. stale?
• how granular are we considering cache-and-network and partial background fetches to be?
• can we expect this state to expand even more?
• should there be a proactive flag as discussed previously where we observe using a subscription when the Client is fetching? (potentially expensive and complex)

Edit: on the note of “is something just missing/empty on the client or server”, I also agree that this should be represented in the schema.

In the case of a list for instance, the field shouldn’t be null to indicate an empty list but instead just an empty array or a connection with an empty array of edges.

[lukasluecke]

In my opinion, having separate stale and partial flags would be great (as @JoviDeCroock suggested). If someone wanted to "merge" them back into fetching, they could just wrap that in a custom hook or something.

[JoviDeCroock]

The nice thing about stale is that we could consolidate stale and partial to be the same thing since essentially they are both stale. That would actually be something I can live with since I don't want to overcomplicate our result a lot.

[kitten]

Will be replaced by a new PR with stale based on master