actionsDict is not bound to this

[esamattis]

Ie. cannot call another action via this.

Example:

	const initialState = { s: "" }
    const red = createRedutser(initialState, {
      a(state, act: {}) {
        return this.b(state, act)
      },

      b(state, act: {}) {
        return {
          s: state.s + "b",
        }
      },
    })

This gives weird type errors

and also errors in runtime because this is undefined.

I managed to workaround this with following helper:

import {mapValues} from "lodash";

type IState = typeof initialState

type ActionCreators = {
    [key: string]: (state: IState, action: any) => IState;
};

export function bindThis<T extends ActionCreators>(obj: T): T {
    const mapped = mapValues(obj, value => {
        if (typeof value === "function") {
            return value.bind(obj);
        }

        return value;
    });

    return (mapped as any) as T;
}

// Usage

const redutser = createRedutser(
    initialState,
    bindThis({ /* actions */ })
);

[wkrueger]

Having a look at it. (Somehow this project got unwatched on my github so i never saw your issues).

[wkrueger]

While the binding issue has been fixed, I couldnt find a decent fix for the inference issues without changing createRedutser signature. For now, you can use createRedutser2 when using this in the Dict. Ugly and sad... :(

[esamattis]

I Might have found a solution for the inference issue. Here's a patch for it

esamattis@a34d87e

This builds on the PR #4.

The idea is to explicitly add typing for this

I did not sent PR for this because this breaks the subdomain specs and currently I don't have time to look at it.

[wkrueger]

I probably have tried something similar and ended up in a loophole where one of the inferred types got to {} (which actually means: I couldnt infer but hey I wont throw you an error).

I'll have a look on your gist, thanks!

In the worst case, ts3.0 is promising partial inference which may fix things that require this curry workaround.

[esamattis]

I probably have tried something similar and ended up in a loophole where one of the inferred types got to {} (which actually means: I couldnt infer but hey I wont throw you an error).

Do you see this with TS 2.9? Seems to be working for me just fine.

[wkrueger]

Seems it only happens when you return something from this.
"bindToSelf" test (modified) continues to fail on ts 2.9.

About your suggestion, that erases the this parameter to a generic object instead of receiving the actual object through generics.

I have also tried adding a generic Self parameter to this too (or even adding a 3rd parameter), but that ends up propagating to the whole codebase and still bringing other new issues (ex: circular references).

So, ugly as it seems, createRedutser2 still is currently more elegant than the other solutions. I could bring a fancier name for it like createRedutserCurried, but well, that is also not much better...

[esamattis]

Yeah, I misunderstood the typing situation. this is actually an any type but vscode was still clever enought to autocomplete it seemingly right so I misunderstood it.

But I played with style of API from scratch and found out that it can be implemented using classes. So I ended up creating Immer Reducer based on it.

[wkrueger]

Offtopic: I have put redux on soak for the moment. One big issue was the struggle to bring type inference to stateful (classy) connected components.