krawaller


Exploring a pattern that lets us group related logic and reduce boilerplate in Redux apps

How to reduce boilerplate in Redux

Tags: reduxtypescriptcyclejs

Premise

Redux is frequently bemoaned for forcing you to write a fair bit of boilerplate. Sometimes that critisism is not fair as the code in question is actually necessary, but definitely Redux app can often feel very verbose.

In October I attended CycleConf, the yearly conference for the CycleJS framework. There I got to play with Cycle State, a new API where each component gets the current state as an input stream, and returns a stream of reducers to calculate the next state.

CycleJS is of course very (very) different from a Redux-driven app, but I realised that the same pattern can be applied to Redux nonetheless! And doing so means less boilerplate and better grouping of related logic.

In this post we'll explore a TypeScript-driven example, where we'll first write "traditional" Redux code, and then refactor it to the pattern in question.

For brevity's sake I'll be cutting some corners, so neither logic nor modelling will be optimal with regards to perfomance and convenience.

The data model

Imagine that we have an app with users and posts:

type User = {
name: string;
postIds: string[];
};

type Post = {
title: string;
content: string;
authorId: string;
};

They each live in a top-level prop in our app state:

type UserState = { [ID: string]: User };
type PostState = { [ID: string]: Post };

type AppState = {
users: UserState;
posts: PostState;
// ...probably more stuff here too
};

Creating actions

We'll be using the popular action format where all additional data are collected in a payload prop:

type ActionBlueprint<T, P> = {
type: T;
payload: P;
};

For our experiment, let's make up two actions - we want to delete posts and add posts!

type NewPostPayload = {
title: string;
content: string;
authorId: string;
postId: string;
};

type NewPostAction = ActionBlueprint<"NEW_POST", NewPostPayload>;

type DeletePostPayload = {
postId: string;
};

type DeletePostAction = ActionBlueprint<"DELETE_POST", DeletePostPayload>;

We'll also make an AppAction type that's simply the union of all existing actions in our app:

type AppAction = NewPostAction | DeletePostAction; // ...

Creating a new post action could be done something like this:

const NewPostActionCreator = (
title: string,
content: string,
authorId: string,
postId: string
): NewPostAction => ({
type: "NEW_POST",
payload: {
title,
content,
authorId,
postId
}
});

And here's how we create delete post actions:

const DeletePostActionCreator = (postId: PostId): DeletePostAction => ({
type: "DELETE_POST",
payload: { postId }
});

Reducer, take one

In a traditional Redux app we'd create a subreducer for appState.posts:

const postReducer = (postState: PostState, action: AppAction): PostState => {
switch (action.type) {
case "NEW_POST":
return {
...postState,
[action.payload.postId]: {
title: action.payload.title,
content: action.payload.content,
authorId: action.payload.authorId
}
};
case "DELETE_POST": {
const { [action.payload.postId]: DELETE_ME, ...keepUs } = postState;
return keepUs;
}
default:
return postState;
}
};

The subreducer for appState.users also has to deal with both actions, in order to keep the user postlists up to date:

const userReducer = (userState: UserState, action: AppAction): UserState => {
switch (action.type) {
case "NEW_POST": {
const author = userState[action.payload.authorId];
return {
...userState,
[author.userId]: {
...author,
postIds: author.postIds.concat(action.payload.postId)
}
};
}
case "DELETE_POST":
return Object.keys(userState).reduce(
(mem, userId) => {
const user = userState[userId];
return {
...mem,
[userId]: {
...user,
postIds: user.postIds.filter(pid => pid !== action.payload.postId)
}
};
},
{} as UserState
);
default:
return userState;
}
};

The reducer for our entire app state is merely just putting the subreducers together:

type Reducer = (state: AppState, action: AppAction) => AppState;

const reducer: Reducer = (state, action) => ({
users: userReducer(currentState.users, action),
posts: postReducer(currentState.posts, action)
});

Normally you'd use Redux.combineReducers, but that amounts to the same thing.

Intermission

So far the traditional approach. What's wrong with it?

Well, some would say we've written a fair bit of boilerplate. I'm not sure I fully agree, but definitely it feels verbose.

A worse thing is the shattering of related logic. If you're new to the codebase and want to read up on how post deletion works, you have to look in three places:

  • the delete action creator
  • the user reducer
  • the post reducer

And in reality it is worse - you have to look in the action creator and in ALL subreducers, because you don't really know which ones might have interest in the action!

Actions 2.0

The pattern we're exploring in this post has a very simple solution to this problem - we move all related reducing into the action objects!

An action, apart from the payload, may therefore also include reducers for whatever state domain it wants to affect! Thus we extend our action type accordingly:

type ActionBlueprint<T, P> = {
type: T;
payload: P;
reduceUsers?(userState: UserState, payload: P): UserState;
reducePosts?(postState: PostState, payload: P): PostState;
};

Here's what the DeletePostActionCreator could look like in this modern take:

const DeletePostActionCreator = (postId: string): DeletePostAction => ({
type: "DELETE_POST",
payload: { postId },
reducePosts(postState, payload) {
const { [payload.postId]: DELETE_ME, ...keepUs } = postState;
return keepUs;
},
reduceUsers(userState, payload) {
return Object.keys(userState).reduce(
(mem, userId) => {
const user = userState[userId];
return {
...mem,
[userId]: {
...user,
postIds: user.postIds.filter(pid => pid !== payload.postId)
}
};
},
{} as UserState
);
}
});

The logic inside the minireducers is what used to live inside individual case:s in the old subreducers.

And the NewPostActionCreator for completion's sake:

const NewPostActionCreator = (
title: string,
content: string,
authorId: string,
postId: string
): NewPostAction => ({
type: "NEW_POST",
payload: {
title,
content,
authorId,
postId
},
reducePosts(postState, payload) {
return {
...postState,
[payload.postId]: {
title: payload.title,
content: payload.content,
authorId: payload.authorId
}
};
},
reduceUsers(userState, payload) {
const author = userState[payload.authorId];
return {
...userState,
[author.userId]: {
...author,
postIds: author.postIds.concat(payload.postId)
}
};
}
});

Note that we don't pass in the full action object to the reducers. Why would we? We are inside a specific creator, so we know what type it is!

Reducer 2.0

What about the reducers in this brave new world? Well, first off there are no subreducers anymore. The various cases from their switches instead lives in the individual action creators, as you just saw.

The main reducer is simply a matter of applying whatever reducers the action in question happen to have:

type Reducer = (
state: AppState,
action: ActionBlueprint<string, any>
) =>
AppState;

const reducer: Reducer = (state, action) => ({
users: action.reduceUsers
? action.reduceUsers(state.users, action.payload)
: state.users,
posts: action.reducePosts
? action.reducePosts(state.posts, action.payload)
: state.posts
});

Taking score

So, did we reduce the amount of boilerplate? I think so! The string matching with action.type is completely gone, and the Redux-typical switches are nowhere to be seen.

But the big win in my mind is the colocation of related logic. Each action creator file will contain the full truth about how that particular action affects the state.

A third, less obvious win is that we get a clearer separation between actions and thunks. Normally it is very common for the former to be created inline in the latter, but now that the actions carry the reducers it is more obvious that they are their own thing.

Actions 3.0

But, waitaminue - why are we splitting up the reducing by state domain? Why don't we just do everything in a single reducer for the full state in one fell swoop?

type ActionBlueprint<T, P> = {
type: T;
payload: P;
reducer: (state: AppState, payload: P) => AppState;
};

The NewPostActionCreator can then be streamlined to this...

const NewPostActionCreator = (
title: string,
content: string,
authorId: UserId,
postId: PostId
): NewPostAction => ({
type: "NEW_POST",
payload: {
postId,
title,
content,
authorId
},
reducer(state, payload) {
const author = state.users[payload.authorId];
return {
...state,
posts: {
...state.posts,
[payload.postId]: {
title: payload.title,
content: payload.content,
authorId: payload.authorId
}
},
users: {
...state.users,
[author.userId]: {
...author,
posts: author.posts.concat(payload.postId)
}
}
};
}
});

...and DeletePostActionCreator becomes very short and sweet:

const DeletePostActionCreator3 = (postId: PostId): DeletePostAction => ({
type: "DELETE_POST",
payload: { postId },
reducer(state, payload) {
const { [payload.postId]: postToDelete, ...postsToKeep } = state.posts;
const author = state.users[postToDelete.authorId];
return {
...state,
posts: postsToKeep,
users: {
...state.users,
[postToDelete.authorId]: {
...author,
posts: author.posts.filter(pid => pid !== payload.postId)
}
}
};
}
});

Another cute advantage is that we can now make our action type work for any app by adding the app state type to the generic parameters:

type ActionBlueprint<T, P, S> = {
type: T;
payload: P;
reducer: (state: S, payload: P) => S;
};

type NewPostAction = ActionBlueprint<"NEW_POST", NewPostPayload, AppState>;
type DeletePostAction = ActionBlueprint<
"DELETE_POST",
DeletePostPayload,
AppState
>;

Reducer 3.0

Now that each action just has one reducer, the "real" reducer becomes even simpler to define:

const reducer: Reducer = (state, action) =>
action.reducer ? action.reducer(state, action.payload) : state;

We still need to check that a reducer exists on the action, since Redux fires its own actions that won't have a built-in reducer.

Wrapping up

I really like how putting the reducing logic inside the action groups all related code together in one place. And you'll find that employing this pattern means less cruft and boilerplate.

The only downside I can see with the pattern we just explored is that it might surprise developers used to the traditional approach. Other than that it is all wins in my book!

What do you think?