Update Signature of Actions and Actions#ofType to infer action interface in Effects

[MikeRyanDev]

Using the new mapped types feature in TypeScript 2.8 we can now write a type safe version of Actions#ofType that can correctly infer the interface of the action based on the provided action type:

type Action<Type extends string> = { type: Type };

class Actions<T extends Action> extends Observable<T> { }

type ActionsOfType<A, T extends string> = A extends Action<T> ? A : never;

function ofType<V, T1 extends string>(t1: T1): OperatorFunction<V, ActionsOfType<V, T1>>;
function ofType(...types: string[]) {
  return function (source: Observable<any>) {
    return source.pipe(filter(action => types.indexOf(action.type) !== -1)) as any;
  }
}

This will be a major breaking change for Effects. To minimize the impact we should add an @ngrx/effects/compat submodule that exports a backwards compatible version of Actions and ofType so that developers can incrementally upgrade. We will remove @ngrx/effects/compat in 7.0.

[timdeschryver]

I don't know how much effort it will cost or if it's even possible.
But I was thinking if this could be resolved with a typescript linter (think of rxjs-tslint)?
Because then we could add a step during ng update to fix it automatically.

[timdeschryver]

I have something ready at tdeschryver/platform/oftype, without the compat module for now.
When the 7x development has started I can create a PR.

[rkirov]

This features is on the critical path for turning on --strictFunctionTypes in google's TS codebase, so I am very invested in getting a solution checked in ASAP.

Here is the problem. Currently the following code compiles without an error:

      this.actions.pipe(
          ofType(ActionTypes.CREATE),
          mergeMap(
              (action: CreatetAction) => {...}

How it actually works is that after ofType TS is inferring an Observalbe<Action>, but without strictFnTypes one is allowed to pipe a callback of type (action: CreateAction) => ... to such an observable. Of course, such a pipe is wrong in general and with strictFnType this starts throwing a compiler error.

It appears there are 3 different solutions for better typing 'oftype' floating around - @MikeRyanDev, @tdeschryver and by @mtaran-google. They are slightly different in terms of naming and maybe support for multiple arguments, but for my purposes I think any of them will fix the majority of the failures we are seeing. I want to get ball rolling, so who will create an actual PR for this, so that we can start reviewing it?

[mtaran-google]

The version I came up with is at https://gist.github.com/mtaran-google/8f30d4853e225e830d9a3e712eb96618

Blame clang-format for the indentation.

[alex-okrushko]

@rkirov Hey Rado, Does this patch passes Google tests? (I haven't tried - not sure if you ran those).

[rkirov]

I did an internal test and there is just a single failure (though it only tests pipable ofType).