We really needed this for Flight before as well but we got away with it
because Blocks were lazy but with the removal of Blocks, we'll need this
to ensure that we can lazily stream in part of the content.
Luckily LazyComponent isn't really just a Component. It's just a generic
type that can resolve into anything kind of like a Promise.
So we can use that to resolve elements just like we can components.
This allows keys and props to become lazy as well.
To accomplish this, we suspend during reconciliation. This causes us to
not be able to render siblings because we don't know if the keys will
reconcile. For initial render we could probably special case this and
just render a lazy component fiber.
Throwing in reconciliation didn't work correctly with direct nested
siblings of a Suspense boundary before but it does now so it depends
on new reconciler.
* Bug: Spawning hydration in response to Idle update
Adds a test that fails in the new fork.
* Fix typos related to Idle priority
These are just silly mistakes that weren't caught by any of our tests.
There's a lot of duplication in the Lanes module right now. It's also
not super stable as we continue to refine our heuristics. Hopefully the
final state is simpler and less prone to these types of mistakes.
Facebook currently relies on being able to hydrate hidden HTML. So
skipping those trees is a regression.
We don't have a proper solution for this in the new API yet. So I'm
reverting it to match the old behavior.
Now the server renderer will treat LegacyHidden the same as a fragment,
with no other special behavior. We can only get away with this because
we assume that every instance of LegacyHidden is accompanied by a host
component wrapper. In the hidden mode, the host component is given a
`hidden` attribute, which ensures that the initial HTML is not visible.
To support the use of LegacyHidden as a true fragment, without an extra
DOM node, we will have to hide the initial HTML in some other way.
* Warn if MutableSource snapshot is a function
useMutableSource does not properly support snapshots that are functions. In part this is because of how it is implemented internally (the function gets mistaken for a state updater function). To fix this we could just wrap another function around the returned snapshot, but this pattern seems problematic to begin with- because the function that gets returned might itself close over mutable values, which would defeat the purpose of using the hook in the first place.
This PR proposes adding a new DEV warning if the snapshot returned is a function. It does not change the behavior (meaning that a function could still work in some cases- but at least the current behavior prevents passing around a closure that may later become stale unless you're really intentional about it e.g. () => () => {...}).
* Replaced .warn with .error
* useMutableSource hydration support
* Remove unnecessary ReactMutableSource fork
* Replaced root.registerMutableSourceForHydration() with mutableSources option
* Response to PR feedback:
1. Moved mutableSources root option to hydrationOptions object
2. Only initialize root mutableSourceEagerHydrationData if supportsHydration config is true
3. Lazily initialize mutableSourceEagerHydrationData on root object
This is a bit gross but I need to be able to access it without importing
the renderer.
There might be a better way but I need this to unblock internal bugfix.
Deferring a hidden tree is only supported in Concurrent Mode.
The missing check leads to an infinite loop when an update is scheduled
inside a hidden tree, because the pending work never gets reset.
This "accidentally" worked in the old reconciler because the heurstic
we used to detect offscreen trees was if `childExpirationTime`
was `Never`.
In the new reconciler, we check the tag instead. Which means we also
need to check the mode, like we do in the begin phase.
We should move this check out of the hot path. It shouldn't have been
in the hot path of the old reconciler, either.
Probably by moving `resetChildLanes` into the switch statement
in ReactFiberCompleteWork.
Need this to unblock www. Not sure yet how we'll support this properly
long term.
While adding this, I noticed that the normal "hidden" mode of
LegacyHidden doesn't work properly because it doesn't toggle the
visibility of newly inserted nodes. This is fine for now since we only
use it via a userspace abstraction that wraps the children in an
additional node. But implementing this correctly is required for us
to start using it like a fragment, without the wrapper node.
* Add LegacyHidden to server renderer
When the tree is hidden, the server renderer renders nothing. The
contents will be completely client rendered.
When the tree is visible it acts like a fragment.
The future streaming server renderer may want to pre-render these trees
and send them down in chunks, as with Suspense boundaries.
* Force client render, even at Offscreen pri
The motivation for doing this is to make it impossible for additional
uses of pre-rendering to sneak into www without going through the
LegacyHidden abstraction. Since this feature was already disabled in
the new fork, this brings the two closer to parity.
The LegacyHidden abstraction itself still needs to opt into
pre-rendering somehow, so rather than totally disabling the feature, I
updated the `hidden` prop check to be obnoxiously specific. Before, you
could set it to any truthy value; now, you must set it to the string
"unstable-do-not-use-legacy-hidden".
The node will still be hidden in the DOM, since any truthy value will
cause the browser to apply a style of `display: none`.
I will have to update the LegacyHidden component in www to use the
obnoxious string prop. This doesn't block merge, though, since the
behavior is gated by a dynamic flag. I will update the component before
I enable the flag.
* Failing useMutableSource test
If a source is mutated after initial read but before subscription is set
up, it should still entangle all pending mutations even if snapshot of
new subscription happens to match.
Test case illustrates how not doing this can lead to tearing.
* Fix useMutableSource tearing bug
Fix is to move the entanglement call outside of the block that checks
if the snapshot has changed.
* useMutableSource: "Entangle" instead of expiring
A lane is said to be entangled with another when it's not allowed to
render in a batch that does not also include the other lane.
This commit implements entanglement for `useMutableSource`. If a source
is mutated in between when it's read in the render phase, but before
it's subscribed to in the commit phase, we must account for whether the
same source has pending mutations elsewhere. The old subscriptions must
not be allowed to re-render without also including the new subscription
(and vice versa), to prevent tearing.
In the old reconciler, we did this by synchronously flushing all the
pending subscription updates. This works, but isn't ideal. The new
reconciler can entangle the updates without de-opting to sync.
In the future, we plan to use this same mechanism for other features,
like skipping over intermediate useTransition states.
* Use clz instead of ctrz to pick an arbitrary lane
Should be slightly faster since most engines have built-in support.
* Expose LegacyHidden type
I will use this internally at Facebook to migrate away from
<div hidden />. The end goal is to migrate to the Offscreen type, but
that has different semantics. This is an incremental step.
* Disable <div hidden /> API in new fork
Migrates to the unstable_LegacyHidden type instead. The old fork does
not support the new component type, so I updated the tests to use an
indirection that picks the correct API. I will remove this once the
LegacyHidden (and/or Offscreen) type has landed in both implementations.
* Add gated warning for `<div hidden />` API
Only exists so we can detect callers in www and migrate them to the new
API. Should not visible to anyone outside React Core team.
* Detect and prevent render starvation, per lane
If an update is CPU-bound for longer than expected according to its
priority, we assume it's being starved by other work on the main thread.
To detect this, we keep track of the elapsed time using a fixed-size
array where each slot corresponds to a lane. What we actually store is
the event time when the lane first became CPU-bound.
Then, when receiving a new update or yielding to the main thread, we
check how long each lane has been pending. If the time exceeds a
threshold constant corresponding to its priority, we mark it as expired
to force it to synchronously finish.
We don't want to mistake time elapsed while an update is IO-bound
(waiting for data to resolve) for time when it is CPU-bound. So when a
lane suspends, we clear its associated event time from the array. When
it receives a signal to try again, either a ping or an update, we assign
a new event time to restart the clock.
* Store as expiration time, not start time
I originally stored the start time because I thought I could use this
in the future to also measure Suspense timeouts. (Event times are
currently stored on each update object for this purpose.) But that
won't work because in the case of expiration times, we reset the clock
whenever the update becomes IO-bound. So to replace the per-update
field, I'm going to have to track those on the room separately from
expiration times.
There is a worry that `useOpaqueIdentifier` might run out of unique IDs if running for long enough. This PR moves the unique ID counter so it's generated per server renderer object instead. For people who render different subtrees, this PR adds a prefix option to `renderToString`, `renderToStaticMarkup`, `renderToNodeStream`, and `renderToStaticNodeStream` so identifiers can be differentiated for each individual subtree.
In the new reconciler, I made a change to how render phase updates
work. (By render phase updates, I mean when a component updates
another component during its render phase. Or when a class component
updates itself during the render phase. It does not include when
a hook updates its own component during the render phase. Those have
their own semantics. So really I mean anything triggers the "`setState`
in render" warning.)
The old behavior is to give the update the same "thread" (expiration
time) as whatever is currently rendering. So if you call `setState` on a
component that happens later in the same render, it will flush during
that render. Ideally, we want to remove the special case and treat them
as if they came from an interleaved event.
Regardless, this pattern is not officially supported. This behavior is
only a fallback. The flag only exists until we can roll out the
`setState` warnning, since existing code might accidentally rely on the
current behavior.
* Don't attempt to render the children of a dehydrated Suspense boundary
The DehydratedFragment tag doesn't exist so doing so throws.
This can happen if we schedule childExpirationTime on the boundary and
bail out.
* Warn if scheduling work on a component before it is committed
Adds several new experimental APIs to aid with automated testing.
Each of the methods below accepts an array of "selectors" that identifies a path (or paths) through a React tree. There are four basic selector types:
* Component: Matches Fibers with the specified React component type
* Role: Matches Host Instances matching the (explicit or implicit) accessibility role.
* Test name: Matches Host Instances with a data-testname attribute.
* Text: Matches Host Instances that directly contain the specified text.
* There is also a special lookahead selector type that enables further matching within a path (without actually including the path in the result). This selector type was inspired by the :has() CSS pseudo-class. It enables e.g. matching a <section> that contained a specific header text, then finding a like button within that <section>.
API
* findAllNodes(): Finds all Host Instances (e.g. HTMLElement) within a host subtree that match the specified selector criteria.
* getFindAllNodesFailureDescription(): Returns an error string describing the matched and unmatched portions of the selector query.
* findBoundingRects(): For all React components within a host subtree that match the specified selector criteria, return a set of bounding boxes that covers the bounds of the nearest (shallowed) Host Instances within those trees.
* observeVisibleRects(): For all React components within a host subtree that match the specified selector criteria, observe if it’s bounding rect is visible in the viewport and is not occluded.
* focusWithin(): For all React components within a host subtree that match the specified selector criteria, set focus within the first focusable Host Instance (as if you started before this component in the tree and moved focus forwards one step).
We've been shipping unprefixed experimental APIs (like `createRoot` and
`useTransition`) to the Experimental release channel, with the rationale
that because these APIs do not appear in any stable release, we're free
to change or remove them later without breaking any downstream projects.
What we didn't consider is that downstream projects might be tempted to
use feature detection:
```js
const useTransition = React.useTransition || fallbackUseTransition;
```
This pattern assumes that the version of `useTransition` that exists in
the Experimental channel today has the same API contract as the final
`useTransition` API that we'll eventually ship to stable.
To discourage feature detection, I've added an `unstable_` prefix to
all of our unstable APIs.
The Facebook builds still have the unprefixed APIs, though. We will
continue to support those; if we make any breaking changes, we'll
migrate the internal callers like we usually do. To make testing easier,
I added the `unstable_`-prefixed APIs to the www builds, too. That way
our tests can always use the prefixed ones without gating on the
release channel.
See PR #18796 for more information.
All of the changes I've made in this commit are behind the
`enableNewReconciler` flag. Merging this to master will not affect the
open source builds or the build that we ship to Facebook.
The only build that is affected is the `ReactDOMForked` build, which is
deployed to Facebook **behind an experimental flag (currently disabled
for all users)**. We will use this flag to gradually roll out the new
reconciler, and quickly roll it back if we find any problems.
Because we have those protections in place, what I'm aiming for with
this initial PR is the **smallest possible atomic change that lands
cleanly and doesn't rely on too many hacks**. The goal has not been to
get every single test or feature passing, and it definitely is not to
implement all the features that we intend to build on top of the new
model. When possible, I have chosen to preserve existing semantics and
defer changes to follow-up steps. (Listed in the section below.)
(I did not end up having to disable any tests, although if I had, that
should not have necessarily been a merge blocker.)
For example, even though one of the primary goals of this project is to
improve our model for parallel Suspense transitions, in this initial
implementation, I have chosen to keep the same core heuristics for
sequencing and flushing that existed in the ExpirationTimes model: low
priority updates cannot finish without also finishing high priority
ones.
Despite all these precautions, **because the scope of this refactor is
inherently large, I do expect we will find regressions.** The flip side
is that I also expect the new model to improve the stability of the
codebase and make it easier to fix bugs when they arise.
* Unhide Suspense trees without entanglement
When a Suspense boundary is in its fallback state, you cannot switch
back to the main content without also finishing any updates inside the
tree that might have been skipped. That would be a form of tearing.
Before we fixed this in #18411, the way this bug manifested was that a
boundary was suspended by an update that originated from a child
component (as opposed to props from a parent). While the fallback was
showing, it received another update, this time at high priority. React
would render the high priority update without also including the
original update. That would cause the fallback to switch back to the
main content, since the update that caused the tree to suspend was no
longer part of the render. But then, React would immediately try to
render the original update, which would again suspend and show the
fallback, leading to a momentary flicker in the UI.
The approach added in #18411 is, when receiving a high priority update
to a Suspense tree that's in its fallback state is to bail out, keep
showing the fallback and finish the update in the rest of the tree.
After that commits, render again at the original priority. Because low
priority expiration times are inclusive of higher priority expiration
times, this ensures that all the updates are committed together.
The new approach in this commit is to turn `renderExpirationTime` into a
context-like value that lives on the stack. Then, when unhiding the
Suspense boundary, we can push a new `renderExpirationTime` that is
inclusive of both the high pri update and the original update that
suspended. Then the boundary can be unblocked in a single render pass.
An advantage of the old approach is that by deferring the work of
unhiding, there's less work to do in the high priority update.
The key advantage of the new approach is that it solves the consistency
problem without having to entangle the entire root.
* Create internal LegacyHidden type
This only exists so we can clean up the internal implementation of
`<div hidden={isHidden} />`, which is not a stable feature. The goal
is to move everything to the new Offscreen type instead. However,
Offscreen has different semantics, so before we can remove the legacy
API, we have to migrate our internal usage at Facebook. So we'll need
to maintain both temporarily.
In this initial commit, I've only added the type. It's not used
anywhere. The next step is to use it to implement `hidden`.
* Use LegacyHidden to implement old hidden API
If a host component receives a `hidden` prop, we wrap its children in
an Offscreen fiber. This is similar to what we do for Suspense children.
The LegacyHidden type happens to share the same implementation as the
new Offscreen type, for now, but using separate types allows us to fork
the behavior later when we implement our planned changes to the
Offscreen API.
There are two subtle semantic changes here. One is that the children of
the host component will have their visibility toggled using the same
mechanism we use for Offscreen and Suspense: find the nearest host node
children and give them a style of `display: none`. We didn't used to do
this in the old API, because the `hidden` DOM attribute on the parent
already hides them. So with this change, we're actually "overhiding" the
children. I considered addressing this, but I figure I'll leave it as-is
in case we want to expose the LegacyHidden component type temporarily
to ease migration of Facebook's internal callers to the Offscreen type.
The other subtle semantic change is that, because of the extra fiber
that wraps around the children, this pattern will cause the children
to lose state:
```js
return isHidden ? <div hidden={true} /> : <div />;
```
The reason is that I didn't want to wrap every single host component
in an extra fiber. So I only wrap them if a `hidden` prop exists. In
the above example, that means the children are conditionally wrapped
in an extra fiber, so they don't line up during reconciliation, so
they get remounted every time `isHidden` changes.
The fix is to rewrite to:
```js
return <div hidden={isHidden} />;
```
I don't anticipate this will be a problem at Facebook, especially since
we're only supposed to use `hidden` via a userspace wrapper component.
(And since the bad pattern isn't very React-y, anyway.)
Again, the eventual goal is to delete this completely and replace it
with Offscreen.
* Root API should clear non-empty roots before mounting
Legacy render-into-subtree API removes children from a container before rendering into it. The root API did not do this previously, but just left the children around in the document.
This commit adds a new FiberRoot flag to clear a container's contents before mounting. This is done during the commit phase, to avoid multiple, observable mutations.
SuspenseList progressively renders items even if the list is CPU bound,
i.e. it isn't waiting for missing data. It does this by showing a
fallback for the remaining items, committing the items in that have
already finished, then starting a new render to continue working on
the rest.
When it schedules that subsequent render, it uses a slightly lower
priority than the current render: `renderExpirationTime - 1`.
This commit changes it to reschedule at `renderExpirationTime` instead.
I don't know what the original motivation was for bumping the expiration
time slightly lower. The comment says that the priorities of the two
renders are the same (which makes sense to me) so I imagine it was
motivated by some implementation detail. I don't think it's necessary
anymore, though perhaps it was when it was originally written. If it is
still necessary, we should write a test case that illustrates why.
Changes the internal fiber structure of the Suspense component. When a
Suspense boundary can't finish rendering and switches to a fallback, we
wrap the "primary" tree in a Fragment fiber and hide all its DOM nodes.
Then we mount the fallback tree into a separate Fragment fiber. Both
trees will render into the same parent DOM node (since React fragments
aren't part of the host tree), but the wrappers ensure that the children
in each tree are reconciled separately.
The old implementation would try to be clever and only add the fragment
wrapper when the fallback was in place, to save memory. This "worked"
but was prone to regressions, since this is the only such place in the
codebase where we wrap existing nodes in a new node. (In other words,
it's a form of reparenting, which we don't implement elsewhere).
Since the original implementation, we've also added lots of additional
requirements to the Suspense component that have led to an explosion in
complexity, like limited support in Legacy Mode (with very different
semantics) and progressive hydration.
We're planning to add even more features to the Suspense boundary, so
we're going to sacrifice a bit more memory for a simpler implementation
that is less prone to regressions.
This ended up removing a lot of weird hacks and edge cases, but there
are still plenty left over. Most of the remaining complexity is related
to Legacy mode. That's the next thing we should aim to drop support for.
Because this is a risky change, I've only changed this in the new
reconciler. It blocks some other features, but as of now we're not
planning to implement those in the old reconciler. If that changes, this
should cherry-pick to the other implementation without much effort.
* Add LanePriority type
React's internal scheduler has more priority levels than the external
Scheduler package. Let's use React as the source of truth for tracking
the priority of updates so we have more control. We'll still fall back
to Scheduler in the default case. In the future, we should consider
removing `runWithPriority` from Scheduler and replacing the valid use
cases with React-specific APIs.
This commit adds a new type, called a LanePriority to disambiguate from
the Scheduler one.
("Lane" refers to another type that I'm planning. It roughly translates
to "thread." Each lane will have a priority associated with it.)
I'm not actually using the lane anywhere, yet. Only setting stuff up.
* Remove expiration times train model
In the old reconciler, expiration times are computed by applying an
offset to the current system time. This has the effect of increasing
the priority of updates as time progresses. Because we also use
expiration times as a kind of "thread" identifier, it turns out this
is quite limiting because we can only flush work sequentially along
the timeline.
The new model will use a bitmask to represent parallel threads that
can be worked on in any combination and in any order.
In this commit, expiration times and the linear timeline are still in
place, but they are no longer based on a timestamp. Effectively, they
are constants based on their priority level.
* Stop using ExpirationTime to represent timestamps
Follow up to the previous commit. This converts the remaining places
where we were using the ExpirationTime type to represent a timestamp,
like Suspense timeouts.
* Fork Dependencies and PendingInteractionMap types
These contain expiration times
* Make ExpirationTime an opaque type
ExpirationTime is currently just an alias for the `number` type, for a
few reasons. One is that it predates Flow's opaque type feature. Another
is that making it opaque means we have to move all our comparisons and
number math to the ExpirationTime module, and use utility functions
everywhere else.
However, this is actually what we want in the new system, because the
Lanes type that will replace ExpirationTime is a bitmask with a
particular layout, and performing operations on it will involve more
than just number comparisions and artihmetic. I don't want this logic to
spread ad hoc around the whole codebase.
The utility functions get inlined by Closure so it doesn't matter
performance-wise.
I automated most of the changes with JSCodeshift, with only a few manual
tweaks to stuff like imports. My goal was to port the logic exactly to
prevent subtle mistakes, without trying to simplify anything in the
process. I'll likely need to audit many of these sites again when I
replace them with the new type, though, especially the ones
in ReactFiberRoot.
I added the codemods I used to the `scripts` directory. I won't merge
these to master. I'll remove them in a subsequent commit. I'm only
committing them here so they show up in the PR for future reference.
I had a lot of trouble getting Flow to pass. Somehow it was not
inferring the correct type of the constants exported from the
ExpirationTime module, despite being annotated correctly.
I tried converting them them to constructor functions — `NoWork`
becomes `NoWork()` — and that made it work. I used that to unblock me,
and fixed all the other type errors. Once there were no more type
errors, I tried converting the constructors back to constants. Started
getting errors again.
Then I added a type constraint everywhere a constant was referenced.
That fixed it. I also figured out that you only have to add a constraint
when the constant is passed to another function, even if the function is
annotated. So this indicates to me that it's probably a Flow bug. I'll
file an issue with Flow.
* Delete temporary codemods used in previous commit
I only added these to the previous commit so that I can easily run it
again when rebasing. When the stack is squashed, it will be as if they
never existed.
Sebastian Markbåge
Brian Vaughn
Andrew Clark
Dominic Gannaway
Sebastian SilbermannandBrian Vaughn
Sophie Alpert
Luna Ruan
Fernando Lores
Dan Abramov
Moti ZilbermanandSebastian Markbage