Permalink
Switch branches/tags
Nothing to show
Find file Copy path
db43126 Apr 13, 2018
2 contributors

Users who have contributed to this file

@j4chou @acdlite
401 lines (318 sloc) 13.4 KB

Summary

Introduce a new version of context that addresses existing limitations.

Basic example

type Theme = 'light' | 'dark';
// Pass a default theme to ensure type correctness
const ThemeContext: Context<Theme> = React.createContext('light');

class ThemeToggler extends React.Component {
  state = {theme: 'light'};
  render() {
    return (
      // Pass the current context value to the Provider's `value` prop.
      // Changes are detected using strict comparison (Object.is)
      <ThemeContext.Provider value={this.state.theme}>
        <button
          onClick={() =>
            this.setState(state => ({
              theme: state.theme === 'light' ? 'dark' : 'light',
            }))
          }>
          Toggle theme
        </button>
        {this.props.children}
      </ThemeContext.Provider>
    );
  }
}

class Title extends React.Component {
  render() {
    return (
      // The Consumer uses a render prop API. Avoids conflicts in the
      // props namespace.
      <ThemeContext.Consumer>
        {theme => (
          <h1 style={{color: theme === 'light' ? '#000' : '#fff'}}>
            {this.props.children}
          </h1>
        )}
      </ThemeContext.Consumer>
    );
  }
}

Motivation

Typically, data in a React application is passed top-down (parent to child) via props. But sometimes it's useful to pass values through multiple levels of abstraction without involving each intermediate. Examples include a locale, or a UI theme. Many components may rely on those but you don't want to have to pass a locale prop and a theme prop through every level of the tree.

Context in React provides a mechanism for a child component to access a value in an ancestor component. In this document, we'll refer to the ancestor as the provider and the child as the consumer.

Drawbacks of the existing version of context

shouldComponentUpdate blocks context changes

The main flaw with context today is how it interacts with shouldComponentUpdate. If an intermediate component bails out using shouldComponentUpdate, and there are no pending updates further down the tree, React will assume there are no changes and reuse the entire subtree. If the subtree contains a context consumer, the consumer will not receive the latest context. In other words, context changes will not propagate through a component whose shouldComponentUpdate return false.

shouldComponentUpdate is a fairly common optimization in React applications. A shared component or open source library can't assume that applications won't use it. In practice, this means that context alone is not a reliable way to broadcast changes.

Shifts complexity to user space

Today, developers circumvent the shouldComponentUpdate problem using subscriptions:

  • The provider acts as an event emitter. It keeps track of the most recent context value, and a list of subscribers to be notified whenever it changes.
  • The consumer accesses the provider's event emitter using the context API. (This usage is fine because the event emitter itself does not change).
  • The consumer registers an event listener with the provider.
  • When the provider emits a change event, the consumer is notified and calls setState on itself to schedule a re-render.

Subscriptions are widely used by open source libraries such as Redux and React Broadcast. It works, but it has some clear drawbacks:

  • Not ergonomic. Given how common the context use case is, it shouldn't be so difficult to implement properly.
  • Start-up cost. Setting up subscriptions for every consumer is costly, especially since they aren't used during the initial mount.
  • Encourages mutation and other non-idiomatic patterns that could cause bugs in async mode.
  • The same code ends up being duplicated by every library, increasing bundle sizes.

The meta problem is that the ownership and responsibility for a core feature has been shifted from the framework to its users.

Main goals of this proposal

  • Zero cost (or close to it) for initial mount, commit, and unmount, trading-off update cost as necessary.
  • Easy-to-use API.
  • Statically typable.
  • Encourage async-friendly practices, like immutability.
  • Discourage non-ideal practices, like event emitters and mutation.
  • Eliminate duplicated complexity in userland code.

Detailed design

Introduces new component types: Provider and Consumer.

type Provider<T> = React.Component<{
  value: T,
  children?: React.Node,
}>;

type Consumer<T> = React.Component<{
  children: (value: T) => React.Node,
}>;

Providers and consumers come in pairs—for each provider, there is a corresponding consumer.

A provider-consumer pair is created using React.createContext():

type Context<T> = {
  Provider: Provider<T>,
  Consumer: Consumer<T>,
};

interface React {
  createContext<T>(defaultValue: T): Context<T>;
}

createContext requires a default value to ensure type correctness.

Note that an arbitrary provider and arbitrary consumer cannot necessarily be used in conjunction, even if their value type is the same. They must be the result of the same createContext() call.

The provider accepts a context value as a prop. Any matching consumer in the provider's subtree can access it, regardless of how deeply it's nested.

render() {
  return (
    <Provider value={this.state.contextValue}>
      {this.props.children}
    </Provider>
  );
}

To update the context value, the parent re-renders and passes a different value. Changes to context are detected using Object.is comparison. (Referred to as strict comparision in this proposal, though Object.is is different from === in some edge cases.) This is meant to encourage the use of immutable or persistent data structures. In the typical scenario, context is updated by calling setState on the provider's parent.

The consumer uses a render prop API:

render() {
  return (
    <Consumer>
      {contextValue => <Child arbitraryProp={contextValue} />}
    </Consumer>
  )
}

Notice how in the above example, the context value can be passed to any arbitrary prop on the child component. The advantage of the render prop API is that we avoid clobbering the prop namespace.

If a consumer is rendered without a matching provider as its ancestor, it receives the default value passed to createContext, ensuring type safety.

Implementation

TODO.

Drawbacks

Relies on strict comparison of context values

This proposal uses strict (reference) comparison to detect changes to context values. This is partly to encourage the use of immutable or persistent data structures. But many common data sources rely on mutation. For example, certain implementations of Flux, or even newer libraries like Relay Modern.

However, there are inherent problems with mutation when combined with async rendering, mostly related to tearing. For architectures that rely on mutation, developers will either decide some level of tearing is acceptable, or evolve to better support async. Regardless, these problems are not exclusive to the context API.

An escape hatch for libraries that rely on mutation is to clone the outer container. (Or even just alternate between two copies.) React will detect a new object reference and trigger a change.

Only one provider type per consumer

The proposed API only allows for a consumer to read values from a single provider type, unlike the current API, which allows a consumer to read from an arbitrary number of provider types.

The solution is to use compose consumers:

<FooConsumer>
  {foo => (
    <BarConsumer>
      {bar => (
        // Render using both foo and bar
        <Child foo={foo} bar={bar} />
      )}
    </BarConsumer>
  )}
</FooConsumer>;

Most abstractions around context already use similar patterns.

Alternatives

setContext

Instead of relying on reference equality to detect changes to context, we could instead use a setContext API that works like setState. However, setting aside the implementation overhead, this API would only be valuable when combined with mutation, which we're specifically aiming to discourage.

Passing context to shouldComponentUpdate

One argument is that we could avoid the shouldComponentUpdate problem by passing context as an argument to that method, compare the incoming context to the previous context, and return true if they are different. The problem is that, unlike props or state, we have no type information. The type of the context object depends on the component's position in the React tree. You could perform a shallow comparison of both objects, but that only works if we assume the values are immutable. And if we're going to assume the values are immutable, React might as well do the comparison automatically.

Class-based API

Instead of render props, we could use a class-based API similar to the one we have today:

class ThemeToggler extends React.Component {
  state = {theme: 'light'};
  getChildContext() {
    return this.state.theme;
  }
  render() {
    return (
      <>
        <button
          onClick={() =>
            this.setState(state => ({
              theme: state.theme === 'light' ? 'dark' : 'light',
            }))
          }>
          Toggle theme
        </button>
        {this.props.children}
      </>
    );
  }
}

class Title extends React.Component {
  static contextType = ThemeContext;
  componentDidUpdate(prevProps, prevState, prevContext) {
    if (this.context !== prevContext) {
      alert('Theme changed!');
    }
  }
  render() {
    return (
      <h1 style={{color: this.context.theme === 'light' ? '#000' : '#fff'}}>
        {this.props.children}
      </h1>
    );
  }
}

The advantage of this API is you'd have easier access to context inside lifecycle methods, possibly avoiding the need for an extra component in the tree.

However, although increasing the depth of a React tree incurs some overhead, the advantage of using special component types is that it's faster to scan the tree for consumers, because we can quickly skip over the other types. With the class-based API, we'd have to check every class component, which is slightly slower. This is likely enough to offset the cost of the extra component.

Unlike the class API, the render prop API also has the advantage of being sufficiently different from the existing context API that we could support both versions during a transitional period without too much confusion.

Algebraic effects

In the past, we've considered modeling context in terms of algebraic effects. Our conclusion was that memoizing effects (in order to skip over subtrees without reevaluating them) made this approach intractable.

How we teach this

This would be our first API that uses render props. They are an increasingly popular pattern in third-party React libraries. However, there could be learning curve for beginners.

Other than that, the use case is one that most React developers will be familiar with. We'll update the context documentation to use the new API.

Plan for adoption

Before releasing, reach out to prominent third-party library maintainers.

Initially, we will release the new API alongside the existing context API as a minor update. The APIs are different enough to avoid much confusion.

Publish a migration guide. Make it clear that although the APIs are different, the new version of context provides a superset of functionality. Sending down static values is still supported.

Replacing subscription-based patterns in favor of context's built-in change propgation may require a larger effort. However, as a first step, developers can migrate to the new API without abandoning subscriptions. Or, they can keep using subscriptions even with the new API, if that makes the most sense for their use case.

Coordinate with library authors to migrate to the new API. Once the major libraries are ready, and we've allowed enough time for users to migrate, we will deprecate the old API in a minor update.

In the following major release, we'll remove the old API.

Unresolved questions

Should we warn if a consumer is rendered without a matching provider?

There are valid use cases for relying on the default context value. In many or most cases, it's likely a developer mistake. We could print a warning in development when a consumer is rendered without a matching provider. To supress the warning, the developer passes true to allowDetached.

render() {
  return (
    // If there's no provider, this renders with the default theme.
    // `allowDetached` suppresses the development warning
    <ThemeContext.Consumer allowDetached={true}>
      {theme => (
        <h1 style={{color: theme === 'light' ? '#000' : '#fff'}}>
          {this.props.children}
        </h1>
      )}
    </ThemeContext.Consumer>
  );
}

Add displayName argument to createContext for better debugging

For warnings and React DevTools, it would help if providers and consumers had a displayName. The question is whether it should be required. We could make it optional, and use a Babel transform to add the name automatically. This is the strategy we use for createClass.

Other

  • Should the consumer use children as a render prop, or a named prop?
  • How quickly should we deprecate and remove the existing context API?
  • Need to run benchmarks to determine how fast this will be.
  • Heuristics for caching.
  • A high-priority version of this feature for animations. (May submit as a separate proposal.)