* s/flushPassiveEffects/unstable_flushWithoutYielding
a first crack at flushing the scheduler manually from inside act(). uses unstable_flushWithoutYielding(). The tests that changed, mostly replaced toFlushAndYield(...) with toHaveYielded(). For some tests that tested the state of the tree before flushing effects (but still after updates), I replaced act() with bacthedUpdates().
* ugh lint
* pass build, flushPassiveEffects returns nothing now
* pass test-fire
* flush all work (not just effects), add a compatibility mode
of note, unstable_flushWithoutYielding now returns a boolean much like flushPassiveEffects
* umd build for scheduler/unstable_mock, pass the fixture with it
* add a comment to Shcduler.umd.js for why we're exporting unstable_flushWithoutYielding
* run testsutilsact tests in both sync/concurrent modes
* augh lint
* use a feature flag for the missing mock scheduler warning
I also tried writing a test for it, but couldn't get the scheduler to unmock. included the failing test.
* Update ReactTestUtilsAct-test.js
- pass the mock scheduler warning test,
- rewrite some tests to use Scheduler.yieldValue
- structure concurrent/legacy suites neatly
* pass failing tests in batchedmode-test
* fix pretty/lint/import errors
* pass test-build
* nit: pull .create(null) out of the act() call
PR #15650 is a bugfix but it's technically a semantic change that could
cause regressions. I don't think it will be an issue, since the
previous behavior was both broken and incoherent, but out of an
abundance of caution, let's wrap it in a flag so we can easily revert
it if necessary.
* Failing test for false positive warning
* Flush passive effects before discrete events
Currently, we check for pending passive effects inside the `setState`
method before we add additional updates to the queue, in case those
pending effects also add things to the queue.
However, the `setState` method is too late, because the event that
caused the update might not have ever fired had the passive effects
flushed before we got there.
This is the same as the discrete/serial events problem. When a serial
update comes in, and there's already a pending serial update, we have to
do it before we call the user-provided event handlers. Because the event
handlers themselves might change as a result of the pending update.
This commit moves the `flushPassiveEffects` call to before the discrete
event handlers are called, and removes it from the `setState` method.
Non-discrete events will not cause passive effects to flush, which is
fine, since by definition they are not order dependent.
If React finishes rendering a tree, delays committing it (e.g.
Suspense), then subsequently starts over or renders a new tree, the
pending tree is no longer valid. That's because rendering a new work-in
progress mutates the old one in place.
The current structure of the work loop makes this hard to reason about
because, although `renderRoot` and `commitRoot` are separate functions,
they can't be interleaved. If they are interleaved by accident, it
either results in inconsistent render output or invariant violations
that are hard to debug.
This commit adds an invariant that throws if the new tree is the same as
the old one. This won't prevent all bugs of this class, but it should
catch the most common kind.
To implement the invariant, I store the finished tree on a field on the
root. We already had a field for this, but it was only being used for
the unstable `createBatch` feature.
A more rigorous way to address this type of problem could be to unify
`renderRoot` and `commitRoot` into a single function, so that it's
harder to accidentally interleave the two phases. I plan to do something
like this in a follow-up.
* Add Batched Mode
React has an unfortunate quirk where updates are sometimes synchronous
-- where React starts rendering immediately within the call stack of
`setState` — and sometimes batched, where updates are flushed at the
end of the current event. Any update that originates within the call
stack of the React event system is batched. This encompasses most
updates, since most updates originate from an event handler like
`onClick` or `onChange`. It also includes updates triggered by lifecycle
methods or effects. But there are also updates that originate outside
React's event system, like timer events, network events, and microtasks
(promise resolution handlers). These are not batched, which results in
both worse performance (multiple render passes instead of single one)
and confusing semantics.
Ideally all updates would be batched by default. Unfortunately, it's
easy for components to accidentally rely on this behavior, so changing
it could break existing apps in subtle ways.
One way to move to a batched-by-default model is to opt into Concurrent
Mode (still experimental). But Concurrent Mode introduces additional
semantic changes that apps may not be ready to adopt.
This commit introduces an additional mode called Batched Mode. Batched
Mode enables a batched-by-default model that defers all updates to the
next React event. Once it begins rendering, React will not yield to
the browser until the entire render is finished.
Batched Mode is superset of Strict Mode. It fires all the same warnings.
It also drops the forked Suspense behavior used by Legacy Mode, in favor
of the proper semantics used by Concurrent Mode.
I have not added any public APIs that expose the new mode yet. I'll do
that in subsequent commits.
* Suspense in Batched Mode
Should have same semantics as Concurrent Mode.
* Use RootTag field to configure type of root
There are three types of roots: Legacy, Batched, and Concurrent.
* flushSync should not flush batched work
Treat Sync and Batched expiration times separately. Only Sync updates
are pushed to our internal queue of synchronous callbacks.
Renamed `flushImmediateQueue` to `flushSyncCallbackQueue` for clarity.
* Don't consider "Never" expiration as part of most recent event time
This doesn't happen with deprioritization since those are not "updates"
by themselves so they don't go into this accounting.
However, they are real updates if they were scheduled as Idle pri using
the scheduler explicitly. It's unclear what suspense should do for these
updates. For offscreen work, we probably want them to commit immediately.
No point in delay them since they're offscreen anyway. However if this is
an explicit but very low priority update that might not make sense.
So maybe this means that these should have different expiration times?
In this PR I just set the suspense to the lowest JND.
However, we don't want is for these things to commit earlier in case
they got batched in with other work so I also ensured that they're not
accounted for in in the workInProgressRootMostRecentEventTime calculation
at all. This makes them commit immediately if they're by themselves, or
after the JND of whatever they were batched in with.
Ultimately, I think that we should probably never schedule anything at
Never that isn't truly offscreen so this should never happen.
However, that begs the question what happens with very low pri work that
suspends. Do we always work at that level first?
* Adjust test to account for the new shorter suspense time
* Warn when suspending at wrong priority
Adds a warning when a user-blocking update is suspended.
Ideally, all we would need to do is check the current priority level.
But we currently have no rigorous way to distinguish work that was
scheduled at user- blocking priority from work that expired a bit and
was "upgraded" to a higher priority. That's because we don't schedule
separate callbacks for every level, only the highest priority level per
root. The priority of subsequent levels is inferred from the expiration
time, but this is an imprecise heuristic.
However, we do store the last discrete pending update per root. So we
can reliably compare to that one. (If we broaden this warning to include
high pri updates that aren't discrete, then this won't be sufficient.)
My rationale is that it's better for this warning to have false
negatives than false positives.
Potential follow-ups:
- Bikeshed more on the message. I don't like what I landed on that much
but I think it's good enough to start.
- Include the names of the components that updated. (The ideal place to
fire the warning is during the setState call but we don't know if
something will suspend until the next update. Maybe we could be clever
and warn during a subsequent update to the same component?)
* Move Suspense priority check to throwException
Bugfix for `inferPriorityFromExpirationTime` function. It happened to
work in our existing tests because we use virtual time.
Flow would have caught this if expiration times were an opaque type. We
should consider that in the future. (The downside of opaque types is
that all operations would have to go through helper functions, which may
or may not get inlined by Closure.)
Fixes a bug where the timeout passed to `scheduleCallback` represented
an absolute timestamp, instead of the amount of time until that
timestamp is reached. The solution is to subtract the current time
from the expiration.
The bug wasn't caught by other tests because we use virtual times that
default to 0, and most tests don't advance time.
I also moved the `initialTimeMs` offset to the
`SchedulerWithReactIntegration` module so that we don't have to remember
to subtract the offset every time. (We should consider upstreaming this
to the Scheduler package.)
The implementation is wrong, but also it's not that useful for
debugging. Implementing it properly would involve tracking more
information than we do currently. Perhaps including the priority
of the callback in the message would be helpful, but not sure. For now
I'll just remove it.
Moves the cancelTimeout call to right before creating a new work-in-
progress root. Fixes a class of bugs where a pending commit is not
cancelled, causing an incomplete tree to accidentally commit.
In the interest of fixing downstream bugs quickly, I'm landing this
without a test case; I'll add one in a follow up.
via #15319
This solves 2 specific problems -
- using the 'wrong' act() doesn't silence the warning
- using the wrong act logs a warning
It does this by using an empty object on the reconciler as the identity of ReactShouldWarnActingUpdates.current. We also add this check when calling createContainer() to catch the common failure of this happening right in the beginning.
make a proper fixture for act()
made fixtures/act for act(). specifically, this lets us tests how act behaves when you use the wrong act() for the wrong renderer. also mvoes dom/../act-dom.html to this fixture.
cleanup fixtures/dom/.gitignore
verify that it 'works' with art+dom
verify that it 'works' with art+test
augh prettier
tweak warning messages
* Track the earliest event time in this render
Rebase
* Track the time of the fallback being shown as an event time
When we switch back from fallback to content, we made progress and we track
the time from when we showed the fallback in the first place as the
last time we made progress.
* Don't retry if synchronous
* Only suspend when we switch to fallback mode
This ensures that we don't resuspend unnecessarily if we're just retrying
the same exact boundary again. We can still unnecessarily suspend
for nested boundaries.
* Rename timedOutAt to fallbackExpirationTime
* Account for suspense in devtools suspense test
We use bitwise operations to compute expiration times, which means they
need to be smaller than 31 bits. So we measure times relative to module
initialization, similar to `performance.now`.
This was already working in the old fiber scheduler, but we didn't have
a test for it.
When an async update expires, React renders at the expiration time that
corresponds to the current time, not at the original update's expiration
time. That way, all the expired work in the tree is flushed in a
single batch.
This is implemented inside `renderRoot` by comparing the next render
expiration time to the current time. If the current time is later,
`renderRoot` will restart at the later time.
Because of poor factoring, the check is currently performed right before
entering the work loop. But the work loop is usually entered multiple
times in a single callback: each time a component throws or suspends.
This led to an infinite loop where React would detect that an update
expired, restart at the current time, make a bit of progress, suspend,
check for expired work again, and start the loop again.
I fixed this by moving the expired work check to the beginning of
`renderRoot`, so that it is not performed every time something suspends.
This isn't ideal, because you could technically still fall into a loop
if more than 10ms lapse in between exiting `renderRoot` and entering it
again. The proper fix is to lift the check outside of `renderRoot`
entirely so that the function can restart without checking for expired
work again. Since this is exceedingly unlikely (and this whole thing is
still behind a flag), I'll do the better fix in an already-planned
follow up to fork `renderRoot` into separate functions for sync and
async work.
* Allow DevTools to toggle Suspense state
* Change API to overrideSuspense
This lets detect support for overriding Suspense from DevTools.
* Add ConcurrentMode test
* Newlines
* Remove unnecessary change
* Naming changes
This doesn't rely on checking the tag. When the alternate of a parent
is missing, it assumes it's a fragment indirection and moves onto the
next parent fiber.
* Rewrite ReactFiberScheduler
Adds a new implementation of ReactFiberScheduler behind a feature flag.
We will maintain both implementations in parallel until the new one
is proven stable enough to replace the old one.
The main difference between the implementations is that the new one is
integrated with the Scheduler package's priority levels.
* Conditionally add fields to FiberRoot
Some fields only used by the old scheduler, and some by the new.
* Add separate build that enables new scheduler
* Re-enable skipped test
If synchronous updates are scheduled by a passive effect, that work
should be flushed synchronously, even if flushPassiveEffects is
called inside batchedUpdates.
* Passive effects have same priority as render
* Revert ability to cancel the current callback
React doesn't need this anyway because it never schedules callbacks if
it's already rendering.
* Revert change to FiberDebugPerf
Turns out this isn't neccessary.
* Fix ReactFiberScheduler dead code elimination
Should initialize to nothing, then assign the exports conditionally,
instead of initializing to the old exports and then reassigning to the
new ones.
* Don't yield before commit during sync error retry
* Call Scheduler.flushAll unconditionally in tests
Instead of wrapping in enableNewScheduler flag.
