Skip to content

alexnm/re-ducks

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

36 Commits
 
 
 
 

Repository files navigation

Building on the duck legacy

Before starting, read more about the original ducks modular approach proposal. When trying to embrace this idea in medium-large scale codebases I noticed that the single duck file becomes harder and harder to maintain and read.

So I want to propose an extended approach that works great when you go beyond a todo-app.

To recap, a duck:

  • MUST export default a function called reducer()
  • MUST export its action creators as functions
  • MUST have action types in the form npm-module-or-app/reducer/ACTION_TYPE
  • MAY export its action types as UPPER_SNAKE_CASE, if an external reducer needs to listen for them, or if it is a published reusable library

Enter re-ducks

Instead of duck files, we use duck folders.

Here's how a duck folder would look like:

duck/
├── actions.js
├── index.js
├── operations.js
├── reducers.js
├── selectors.js
├── tests.js
├── types.js
├── utils.js

NOTE: Each concept from your app will have a similar folder.

Codebase examples

React/Redux ecommerce example - WIP

React/Redux TodoApp + Flow

General rules for a duck folder

A duck folder:

  • MUST contain the entire logic for handling only ONE concept in your app, ex: product, cart, session, etc.
  • MUST have an index.js file that exports according to the original duck rules.
  • MUST keep code with similar purpose in the same file, ex: reducers, selectors, actions, etc.
  • MUST contain the tests related to the duck.

This structure does not require any libraries or abstractions other than redux and redux-thunk at a minimum (but can work with redux-saga or redux-observables)

NOTE: I'm using export default in most of the cases, don't want to get into details there, you can compose your module as you wish.

Types

Let's start from defining the constants we will use as redux action types. In order to keep the naming simple, let's call the file types.js, because constants.js is a bit too generic.

Our examples will model a real duck.

const QUACK = "app/duck/QUACK";
const SWIM = "app/duck/SWIM";

export {
    QUACK,
    SWIM
};

Actions

It's important to be consistent when defining actions, so let's always export functions from this file, we don't care if the action needs any input from the outside to build the payload or not.

import * as types from "./types";

const quack = ( ) => ( {
    type: types.QUACK
} );

const swim = ( distance ) => ( {
    type: types.SWIM,
    payload: {
        distance
    }
} );

export {
    swim,
    quack
};

NOTE: Trying to impose a bit of structure to the actions object, the type/payload approach is pretty popular.

Operations

In a simple application, you can easily dispatch simple actions and use the reducers to manage the state. However, in a more complex app you need to use some sort of middleware to handle more complex interactions. In our case, we use redux-thunk.

The operations file define the interface for our duck. You can reason about it like this: 1 operation = X actions dispatched. This makes each operation function either a thunk in case it needs to dispatch multiple actions, or simply a link to an action already defined in actions.js.

This separation should work with whatever middleware/lib you are using for handling chained/linked/delayed operations.

import * as actions from "./actions";

// This is a link to an action defined in actions.js.
const simpleQuack = actions.quack;

// This is a thunk which dispatches multiple actions from actions.js
const complexQuack = ( distance ) => ( dispatch ) => {
    dispatch( actions.quack( ) ).then( ( ) => {
        dispatch( actions.swim( distance ) );
        dispatch( /* any action */ );
    } );
}

export {
    simpleQuack,
    complexQuack
};

NOTE: redux-observables uses the idea of epics which I also like. Feel free to suggest names for this file by dropping a line on twitter.

Reducers

It's a good practice to keep your state shape in a comment above the reducers, just to have an overview.

In case the state shape is more complex, you should break the reducers into multiple smaller functions that deal with a slice of the state, then combine them at the end.

import { combineReducers } from "redux";
import * as types from "./types";

/* State Shape
{
    quacking: bool,
    distance: number
}
*/

const quackReducer = ( state = false, action ) => {
    switch( action.type ) {
        case types.QUACK: return true;
        /* ... */
        default: return state;
    }
}

const distanceReducer = ( state = 0, action ) => {
    switch( action.type ) {
        case types.SWIM: return state + action.payload.distance;
        /* ... */
        default: return state;
    }
}

const reducer = combineReducers( {
    quacking: quackReducer,
    distance: distanceReducer
} );

export default reducer;

NOTE: Let's keep it simple for now with switch statements and abstract later.

Selectors

In case your state shape is more complex you need selectors in order to map parts of the state to your props or in order to derive some data for your components from the current state.

These are the functions like: getVisibleTodos, isUserAuthenticated, etc. that take the current app state and return some derived data.

function checkIfDuckIsInRange( duck ) {
    return duck.distance > 1000;
}

export {
    checkIfDuckIsInRange
};

NOTE: Selector functions will be used outside the duck folder, so they are part of the interface of the duck.

Index

This file, from a module perspective, behaves as the duck file from the original proposal.

  • It exports, as default, the reducer function of the duck.
  • It exports, as named export, the selectors and the operations.
  • Optionally, it exports the types if they are needed in other ducks.
import reducer from "./reducers";

import * as duckSelectors from "./selectors";
import * as duckOperations from "./operations";
import * as duckTypes from "./types";

export {
    duckSelectors,
    duckOperations,
    duckTypes
};

export default reducer;

Tests

One of the main advantages of redux is that you can easily do unit tests for your reducers, action creators and selectors. And with a small effort, you can do the same for the operations.

Ultimately, the split proposed here also helps you see what you need to test inside each duck. This example is using mocha and expect.js.

import expect from "expect.js";
import reducer from "./reducers";
import * as actions from "./actions";

describe( "duck reducer", function( ) {
    describe( "quack", function( ) {
        const quack = actions.quack( );
        const initialState = false;

        const result = reducer( initialState, quack );

        it( "should quack", function( ) {
            expect( result ).to.be( true ) ;
        } );
    } );
} );

A word on abstractions!

I've been working almost 2 years now in the React ecosystem and I found that abstractions are good when you design them for your specific needs and are your worst enemies when you use them blindly. So while I understand the idea of reducing boilerplate, writing less code, reusing structures, keep in mind that any abstraction that you take for granted will eventually become a burden for your project.

What I like about the ducks approach is that it does not enforce any abstraction on you. It's simply a better way of organizing your app. You have the complete freedom to build your own helper functions or use utility packages that have those functions.

=======================================

Read more about the reasoning behind re-ducks

Hope you find something useful in this! Ping me on twitter, i'd be more than happy to hear your thoughts!

Alex M

About

An attempt to extend the original proposal for redux modular architecture: https://github.com/erikras/ducks-modular-redux

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •