final design decisions before v1.0

[Ivshti]

in light of using this in stremio-web, stremio-example-seed, stremio-example-sauron, stremio-example-ui and stremio-ng-example,

there are a few more design decisions to make before finalizing:

• instead of CatalogMuxer, implement a #[derive] macro that will allow the user to fill a struct with fields that implement Container; that is much more ergonomic; however, this struct should not have interior mutability, and mutability will need to be handled by the user (implementation-specific) - DOING IT
• should internal APIs use future Streams instead of callbacks - DOING IT
• redux middleware style vs elm-style update function - GOING FOR ELM STYLE
  • elm-style update problem: many things are dependent on the user ctx, so we will have to somehow retrieve it before determining what tasks to issue; if we do a promise of the style with_ctx, then it's kinda ugly (adding async logic to update)
    • or, simple solution: we do update(&mut Model, &Ctx) -> Effects
    • we can still have loopback actions
  • a good reason for loopback actions is that you might want some environment side-effects, e.g. routing; receiving a RecommendedXXX (as a result to Open) to load should lead to a page that will trigger this Load
• consider a AddonEffects trait (with a fn that returns AggrRequest) instead of addon_aggr_req; and in general, a mechanisms for Containers to define thier own Load parameters and effects; could be a trait associated type that we use in Actions - not needed
• ...or, a more generalized Effects trait with fn effects(ctx: Ctx) -> Vec<EnvFuture<Action>> - not needed
• should the update function mutate the struct or return a new struct? - mutate
• consider StandardStremio - a struct with all containers a usual stremio app has - not needed, derive macro handles this great
• AddonM can be enhanced with memoization to avoid re-requesting the same thing (e.g. when going from /detail/movie/tt213 to /detail/movie/tt213/tt213), but for now there's no point of doing this cause (1) the reducer will keep the old response and (2) the browser has a cache - not a problem with elm architecture
• Problem: circular dependency: player needs addon system responses (from AddonM) but needs to emit to the library add-on (back in ContextM) or control it somehow; LibAddon itself needs to be connected after AddonM (to listen to AddonRequest/AddonResponse) - not a problem with elm architecture
  • solution 1: re-think the chain
  • solution 2: drop the chain, there'd be one middleware that manages everything
  • solution 3: use a loopback action
  • solution 4: the player should somehow capture a handle ot the LibAddon and use it directly
  • solution 5: the elm architecture (Player update will handle it's own side effects)
• do we need a common trait between MetaItem, MetaPreview and LibItem? - a common trait between LibItem/MetaPreview/MetaItem #53
• do we need LibItemPreview? - no
• when saving the last stream, save the whole stream object but compressed OR save the stream hash; hash can be saved within the libitem - whole stream makes more sense, could be part of PlayerPreferences?; this decision may change
• design decision: models (containers; or should we perhaps rename them to models) should contain raw, but complete data; their update function should focus on how they mutate, and do as little processing as possible; they will do sorting and filtering (e.g. Search) though
• figure out environment-specific settings; perhaps use the Environment trait?

see also: #28

[Ivshti]

Due to many complicated bits, like the player and detail page, we're most likely elm-ing it.

• we can do update(&mut self, &Ctx) for regular containers
• the Ctx itself will be a container that exposes an update(&mut self, &action)
• the Ctx container may actually become cleaner than the contextM since we'll just make a message Internal::UpdateCtx and eliminate the need for interior mutability
• a #[derive] macro will allow us to compose many types that implement update
• this will compose in nicely with elm-like frameworks (seed-rs) but it will also compose great with React/Conrod/whatever with a bit of glue: create a struct that manages mutability and Ctx; we can also provide glue:
  • something that will return a channel Sender (for actions) and a Receiver (for new states and other outputs, e.g. errors), and will manage context and will consume a struct that implements update
• LibAddon's interior mutability might need to be re-thought too
• we should probably sync the library on log-in; It makes a lot of sense for UX

Example API:

// TODO all those will have to take a type arg for Env....
#[derive(StremioModel)]
pub struct StremioModel {
   pub discover: CatalogFiltered,
   pub board: CatalogGrouped,
   pub player: Player<PlayerImpl>
}

let mut stremio = StremioModel::new();
let ctx = Ctx::new();
stremio.update(&ctx); // will return a Vec of futures that have to be handled

// or more sophisticated
// TODO: figure out a better name for this
let (tx, rx) = App::with_managed_context::<Env>();
// send actions to tx
//rx.for_each() // do something with the new &model

NOTE: the models might have to take AddonTransport besides Env