-
Notifications
You must be signed in to change notification settings - Fork 47k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Don't bail out on referential equality of Consumer's props.children function #12470
Don't bail out on referential equality of Consumer's props.children function #12470
Conversation
I’m struggling to understand what you’re trying to do a bit. Do you expect React to propagate the context update even though the context Why are both provider and consumer in one component? Is this representative of the real code you’re writing? Can you provide a real world example? I’m not sure I see how the code in your test case relates to the code in the description. |
|
||
renderConsumer = (context) => { | ||
ReactNoop.yield('App#renderConsumer'); | ||
return <span prop={this.state.value} />; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@gaearon The issue is not so much about context propagating, but that this instance method that reads from state
to be rendered and updated when setState
is called.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The following is pretty close to actual code that experienced this.
// @flow
import * as React from 'react';
import { login } from 'api';
import { BetaFeatureConsumer } from 'BetaFeatures';
import { SingleSignOn } from 'SingleSignOn';
type P = {};
type S = {
username: string,
password: string,
loading: boolean,
errors: *,
};
export default class Login extends React.Component<P, S> {
state = {
username: '',
password: '',
loading: false,
errors: null,
};
onChange = (event: SyntheticEvent<HTMLInputElement>) => {
const { name, value } = event.currentTarget;
this.setState({ [name]: value });
};
onSubmit = (event: SyntheticEvent<*>) => {
event.preventDefault();
this.setState({ loading: true, errors: null });
login(this.state.username, this.state.password).then(data => {
this.setState({ loading: false });
if (data.hasError) {
this.setState({ errors: { ...this.state.errors, ...data.errors } });
}
});
};
_render = v => {
return (
<form onSubmit={this.onSubmit}>
<input
label="Username"
name="username"
onChange={this.onChange}
value={this.state.username}
/>
<input
label="Password"
name="password"
onChange={this.onChange}
value={this.state.password}
type="password"
/>
<button type="submit" disabled={this.state.loading}>
Submit
</button>
{v.singleSignOn && <SingleSignOn onComplete={this.onSubmit} />}
</form>
);
};
render() {
return (
<BetaFeatureConsumer>
{this._render}
</BetaFeatureConsumer>
);
}
}
this.state
is updated unexpectedly
I see. Here's an example: const Ctx = React.createContext();
class Thing extends React.Component {
state = {counter: 0};
renderValue = (value) => {
return <h1>Hello, {value}, {this.state.counter}</h1>;
}
render() {
return (
<React.Fragment>
<Ctx.Consumer>
{this.renderValue}
{/* it would work with v => this.renderValue(v) */}
</Ctx.Consumer>
<button onClick={() =>
this.setState(prevState => ({ counter: prevState.counter + 1 }))
}>
+
</button>
</React.Fragment>
);
}
}
ReactDOM.render(
<Ctx.Provider value="hello"><Thing /></Ctx.Provider>,
document.getElementById('container')
); It's confusing that |
Yes. That exactly. |
@@ -995,8 +995,18 @@ export default function<T, P, I, TI, HI, PI, C, CC, CX, PL>( | |||
renderExpirationTime, | |||
); | |||
} else if (oldProps !== null && oldProps.children === newProps.children) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If we want to fix it, it seems that we should remove this whole bailout branch.
Normally we bail out of children equality because it is very unusual to cache React elements but not impossible. E.g. this pattern would still fail to update: class ... {
...
consumer = <Ctx.Consumer>{value => <h1>Hello, {value}, {this.state.counter}</h1>}</Ctx.Consumer>;
render() {
return consumer;
}
} There is an argument to be made that since it is common to cache closures, we shouldn't apply the same bailout strategy for render props. It's pretty sketchy though. I think in the future, we'll just want to move away from bound closures on instances and instead always provide new closures. So the tradeoff is whether we want to optimize for future best practices or current. If we allow this pattern now, we can't really easily unallow it later. |
This reverts commit 8686c0f.
this.state
is updated unexpectedlyb6f1a2a
to
f7ce2f9
Compare
} | ||
// There is no bailout on `children` equality because we expect people | ||
// to often pass a bound method as a child, but it may reference | ||
// `this.state` (and thus may need to re-render on `setState`). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(The same is true for this.props
)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This change looks good to me.
thank you! 👍 |
…unction (facebook#12470) * Test case for React Context bailing out unexpectedly * This is 💯% definitely not the correct fix at all * Revert "This is 💯% definitely not the correct fix at all" This reverts commit 8686c0f. * Formatting + minor tweaks to the test * Don't bail out on consumer child equality * Tweak the comment * Pretty lint * Silly Dan
I was playing around with the new context API and ran into some unexpected bail outs. The exact scenario was something like the following:
The test case here is similar but without any DOM dependencies. I would expect this to “just work” but alas. My only thoughts for “solving” this is checking if the function exists on the instance before optimistically bailing out and assuming it is pure.