- Start Date: 2018-03-10
- RFC PR: #33
- React Issue: facebook/react#7678
Add new "commit" phase lifecycle, getSnapshotBeforeUpdate
, that gets called before mutations are made. Any value returned by this lifecycle will be passed as the third parameter to componentDidUpdate
.
This lifecycle is important for async rendering, where there may be delays between "render" phase lifecycles (e.g. componentWillUpdate
and render
) and "commit" phase lifecycles (e.g. componentDidUpdate
).
Consider the use case of preserving scroll position within a list as its contents are updated. The way this is typically done is to read scrollHeight
during render (componentWillUpdate
) and then adjust it after the update has been committed (componentDidUpdate
).
Unfortunately this approach does not work with async rendering, because there might be a delay between these lifecycles during which the user continues scrolling. The only way to ensure an accurate scroll position is read would be to force a synchronous render.
The solution is to introduce a new lifecycle that gets called during the commit phase before mutations have been made to e.g. the DOM. For example:
type Snapshot = number | null;
class ScrollingList extends React.Component<Props, State, Snapshot> {
listRef = React.createRef();
getSnapshotBeforeUpdate(
prevProps: Props,
prevState: State
): Snapshot {
// Are we adding new items to the list?
// Capture the current height of the list so we can adjust scroll later.
if (prevProps.list.length < this.props.list.length) {
return this.listRef.value.scrollHeight;
}
return null;
}
componentDidUpdate(
prevProps: Props,
prevState: State,
snapshot: Snapshot
) {
// If we have a snapshot value, then we've just added new items.
// Adjust scroll so these new items don't push the old ones out of view.
if (snapshot !== null) {
this.listRef.value.scrollTop +=
this.listRef.value.scrollHeight - snapshot;
}
}
render() {
return (
<div ref={this.listRef}>{/* ...contents... */}</div>
);
}
}
This lifecycle provides a way for asynchronously rendered components to accurately read values from the host environment (e.g. the DOM) before it is mutated.
The example above describes one use case in which this could be useful. Others might involve text selection and cursor position, audio/video playback position, etc.
Add a new effect type, Snapshot
, and update ReactFiberClassComponent
to assign this type when updating components that define the new getSnapshotBeforeUpdate
lifecycle.
During the commitAllHostEffects
traversal, call getSnapshotBeforeUpdate
for any fiber tagged with the new Snapshot
effect type. Store return value on the instance (as __reactInternalSnapshotBeforeUpdate
) and later pass to componentDidUpdate
during commitLifeCycles
.
Add DEV warnings for the following conditions:
- Undefined return values for
getSnapshotBeforeUpdate
- Components that define
getSnapshotBeforeUpdate
without also definingcomponentDidUpdate
Flow will also need to be updated to add a third Snapshot
type parameter to React.Component
to ensure consistency for the return type fo getSnapshotBeforeUpdate
adn the new parameter passed to componentDidUpdate
. This new type parameter will be declared like so:
// If there is a State type:
class Example extends React.Component<Props, State, Snapshot> {}
// If there is no State type:
class Example extends React.Component<Props, State = void, Snapshot> {}
It would be possible for react-lifecycles-compat to polyfill this new method for older, synchronous versions of React using the componentWillUpdate
lifecycle. It would require a couple of hacks though:
- The polyfilled
componentWillUpdate
method would need to temporarily mutate instance props (this.props
andthis.state
) before calling the newgetSnapshotFromUdate
method in order to maintain next/prev semantics. - The polyfill would need to mutate the component's
prototype
to decoratecomponentDidUpdate
in order to add the newsnapshot
parameter. This would not work in all cases (e.g. methods that get attached to the instance in the constructor rather than as part of the prototype).
Regardless, I think it's probably reasonable to follow the precedent set by getDerivedStateFromProps
and not call the call unsafe legacy lifecycles componentWillMount
, componentWillReceiveProps
, or componentWillUpdate
for any component that defines the new getSnapshotBeforeUpdate
method.
A DEV warning can be added for components that define both getSnapshotBeforeUpdate
and any of the unsafe legacy lifecycles.
Each new lifecycle adds complexity and makes the component API harder for beginners to understand. Although this lifecycle is important, it will probably not be used often, and so I think the impact is minimal.
A new commit-phase lifecycle is necessary. The signature does not have to match the one proposed by this RFC however. Below are some alternatives that were considered.
The most recently-added lifecycle, getDerivedStateFromProps
, was a static method in order to prevent unsafe access of instance properties. That concern is less relevant in this case though, because this lifecycle is called during the commit phase.
class ScrollingList extends React.Component<Props, State> {
state = {
listHasGrown: false,
listRef: React.createRef(),
prevList: this.props.list
};
static getDerivedStateFromProps(
nextProps: Props,
prevState: State
): $Shape<State> | null {
if (nextProps.list !== prevState.prevList) {
return {
listHasGrown:
nextProps.list.length > prevState.prevList.length,
prevList: nextProps.list
};
} else if (prevState.listHasGrown) {
return {
listHasGrown: false
};
}
return null;
}
static getSnapshotBeforeUpdate(
prevProps: Props,
prevState: State
): Snapshot | null {
if (prevState.listHasGrown) {
return prevState.listRef.value.scrollHeight;
}
return null;
}
// ...
}
This approach was not chosen because of the added complexity of storing additional values (including refs) in state
.
The proposed lifecycle will be the first commit phase lifecycle with a meaningful return value and the first lifecycle whose return value is passed as a parameter to another lifecycle. Likewise, the new parameter for componentDidUpdate
will be the first passed to a lifecycle that isn't some form of Props
or State
. This adds some complexity to the API, since it requires a more nuanced understanding the relationship between getSnapshotBeforeUpdate
and componentDidUpdate
.
An alternative would be to scrap the return value in favor of storing snapshot values on the instance. This has the added benefit of not requiring any changes to be made to Flow.
class ScrollingList extends React.Component<Props, State> {
listRef = React.createRef();
listScrollHeight = null;
getSnapshotBeforeUpdate(
prevProps: Props,
prevState: State
) {
if (prevProps.list.length < this.props.list.length) {
this.listScrollHeight = this.listRef.value.scrollHeight;
}
}
componentDidUpdate(prevProps: Props, prevState: State) {
if (this.listScrollHeight !== null) {
this.listRef.value.scrollTop +=
this.listRef.value.scrollHeight - snapshot;
this.listScrollHeight = null;
}
}
// ...
}
Ultimately, the team voted against this approach because it encourages mutations and may invite other side-effects in a lifecycle that is intended to be used for a very specific purpose.
Since this lifecycle- and async rendering in general- is new functionality, adoption will be organic. Documentation and dev-mode warnings have already been created to encourage people to move away from render phase lifecycles like componentWillUpdate
in favor of commit phase lifecycles.
Lifecycle documentation on the website. Add a before and after example (like the one above) to the Update on Async Rendering blog post "recipes".
None presently.