material-motion / indefinite-observable-js Goto Github PK

The coverage badge points at develop: it should point at stable for releases.

Closed by 4443d15

Can't add the [$$observable] line of

export interface Observable<T> {
  subscribe(listener: Observer | Next): Subscription;
  [$$observable](): Observable<T>;
}

because $$observable throws this error:

TS1169: A computed property name in an interface must directly refer to a built-in symbol.

Symbol didn't exist until Safari 8, so using Symbol.observable rather than $$observable would affect our compatibility with Safari 8 for no user-justifiable reason.

What do?

IndefiniteSubject isn't documented in the README and hasn't been implemented in Swift/Java. IndefiniteObservable has. Therefore, v1.0.0 is blocked by documentation here and implementation across the platforms.

Potential solutions:

• Fork the present repo into indefinite-subject, delete the indefinite-observable code, and depend on indefinite-observable for types and wrapWithObserver.
• Move indefinite-subject into materal-motion-js/packages.
• Move IndefiniteSubject.ts (and the associated tests) into material-motion-js/packages/streams

First draft here: https://github.com/material-motion/indefinite-observable-js/tree/ci

Building on work of @tofuness in the other repo

From @jverkoey's over-the-shoulder code review.

https://github.com/tc39/proposal-observable

What was creator was changed to connect on the docs branch, but that never made it back to develop.

I'd just do it, but since the native platforms are using subscriber, maybe we should be consistent.

Port .travis.yml to circle.yml and delete .travis.yml.

Travis works, but we're using Circle to be compatible with Phabricator in the main repo, so we should be consistent here.

Should just be a simple index.html with a direct import of indefinite-observable.js and inline JavaScript.

You probably shouldn't be able to call unsubscribe on a subscription that isn't subscribed.

The TC39 spec tracks this in a boolean called closed. Can we implement this while maintaining our goal of a fits-on-a-page minimal correct implementation?