Stacked on #30981. Same as #30967 but for effects.
This logs a tree of components using `performance.measure()`.
In addition to the previous render phase this logs one tree for each
commit phase:
- Mutation Phase
- Layout Effect
- Passive Unmounts
- Passive Mounts
I currently skip the Before Mutation phase since the snapshots are so
unusual it's not worth creating trees for those.
The mechanism is that I reuse the timings we track for
`enableProfilerCommitHooks`. I track first and last effect timestamp
within each component subtree. Then on the way up do we log the entry.
This means that we don't include overhead to find our way down to a
component and that we don't need to add any additional overhead by
reading timestamps.
To ensure that the entries get ordered correctly we need to ensure that
the start time of each parent is slightly before the inner one.
Stacked on #30979.
The problem with the previous approach is that it recursively walked the
tree up to propagate the resulting time from recording a layout effect.
Instead, we keep a running count of the effect duration on the module
scope. Then we reset it when entering a nested Profiler and then we add
its elapsed count when we exit the Profiler.
This also fixes a bug where we weren't previously including unmount
times for some detached trees since they couldn't bubble up to find the
profiler.
Stacked on #30960 and #30966. Behind the enableComponentPerformanceTrack
flag.
This is the first step of performance logging. This logs the start and
end time of a component render in the passive effect phase. We use the
data we're already tracking on components when the Profiler component or
DevTools is active in the Profiling or Dev builds. By backdating this
after committing we avoid adding more overhead in the hot path. By only
logging things that actually committed, we avoid the costly unwinding of
an interrupted render which was hard to maintain in earlier versions.
We already have the start time but we don't have the end time. That's
because `actualStartTime + actualDuration` isn't enough since
`actualDuration` counts the actual CPU time excluding yields and
suspending in the render.
Instead, we infer the end time to be the start time of the next sibling
or the complete time of the whole root if there are no more siblings. We
need to pass this down the passive effect tree. This will mean that any
overhead and yields are attributed to this component's span. In a follow
up, we'll need to start logging these yields to make it clear that this
is not part of the component's self-time.
In follow ups, I'll do the same for commit phases. We'll also need to
log more information about the phases in the top track. We'll also need
to filter out more components from the trees that we don't need to
highlight like the internal Offscreen components. It also needs polish
on colors etc.
Currently, I place the components into separate tracks depending on
which lane currently committed. That way you can see what was blocking
Transitions or Suspense etc. One problem that I've hit with the new
performance.measure extensions is that these tracks show up in the order
they're used which is not the order of priority that we use. Even when
you add fake markers they have to actually be within the performance run
since otherwise the calls are noops so it's not enough to do that once.
However, I think this visualization is actually not good because these
trees end up so large that you can't see any other lanes once you expand
one. Therefore, I think in a follow up I'll actually instead switch to a
model where Components is a single track regardless of lane since we
don't currently have overlap anyway. Then the description about what is
actually rendering can be separate lanes.
<img width="1512" alt="Screenshot 2024-09-15 at 10 55 55 PM"
src="https://github.com/user-attachments/assets/5ca3fa74-97ce-40c7-97f7-80c1dd7d6470">
<img width="1512" alt="Screenshot 2024-09-15 at 10 56 27 PM"
src="https://github.com/user-attachments/assets/557ad65b-4190-465f-843c-0bc6cbb9326d">
We used to queue a separate third passive phase to invoke onPostCommit
but this is unnecessary. We can just treat it as a plain passive effect.
This means it is interleaved with other passive effects but we only need
to know the duration of the things below us which is already done at
this point.
I also extracted the user space call to onPostCommit into
ReactCommitEffects. Same as onCommit. It's now covered by
runWithFiberInDEV and catches.
Nit: I don't trust flags in hot code. While it can take somewhat longer
to compile two functions and JIT them. After that they don't need to
check branches. Also makes it clearer the purpose.
Insertion effects do not unmount when a subtree is removed while
offscreen.
Current behavior for an insertion effect is if the component goes
- *visible -> removed:* calls insertion effect cleanup
- *visible -> offscreen -> removed:* insertion effect cleanup is never
called
This makes it so we always call insertion effect cleanup when removing
the component.
Likely also fixes https://github.com/facebook/react/issues/26670
---------
Co-authored-by: Rick Hanlon <rickhanlonii@fb.com>
At some point this trick was added to initialize the value first to NaN
and then replace them with zeros and negative ones.
This is to address the issue noted in
https://github.com/facebook/react/issues/14365 where these hidden
classes can be initialized to SMIs at first and then deopt when we
realize they're actually doubles.
However, this fix has been long broken and has deopted the profiling
build for years because closure compiler optimizes out the first write.
I'm not sure because I haven't A/B-tested this in the JIT yet but I
think we can use negative zero and -1.1 as the initial values instead
since they're not simple integers. Negative zero `===` zero (but not
Object.is) so is the same as far as our code is concerned. The negative
value is just `< 0` comparisons.
When a component suspends and is replaced by a fallback, we should start
prerendering the fallback immediately, even before any new data is
received. During the retry, we can enter prerender mode directly if
we're sure that no new data was received since we last attempted to
render the boundary.
To do this, when completing the fallback, we leave behind a pending
retry lane on the Suspense boundary. Previously we only did this once a
promise resolved, but by assigning a lane during the complete phase, we
will know that there's speculative work to be done.
Then, upon committing the fallback, we mark the retry lane as suspended
— but only if nothing was pinged or updated in the meantime. That allows
us to immediately enter prerender mode (i.e. render without skipping any
siblings) when performing the retry.
If something suspends in the shell — i.e. we won't replace the suspended
content with a fallback — we might as well prerender the siblings during
the current render pass, instead of spawning a separate prerender pass.
This is implemented by setting the "is prerendering" flag to true
whenever we suspend in the shell. But only if we haven't already skipped
over some siblings, because if so, then we need to schedule a separate
prerender pass regardless.
This reverts #19603.
Before:
<img width="724" alt="Screenshot 2024-08-28 at 12 07 29 AM"
src="https://github.com/user-attachments/assets/0613088f-c013-4f1c-92c3-fbdae8c1f109">
After:
<img width="771" alt="Screenshot 2024-08-28 at 12 08 13 AM"
src="https://github.com/user-attachments/assets/eef21bee-d11f-4f0a-9147-053a163f720f">
Consensus seems to be that while the purple on is a bit clearer and
easier to read. The purple is not on brand so it doesn't look like
React. It looks ugly. It's distracting (too eye catching). Taking away
attention from other tabs in an unfair way.
It also gets worse with more tabs added. We plan on both adding another
tab and panes inside other tabs (elements/sources) soon. Each needs to
be marked somehow as part of React but spelling it out is too long.
Putting inside a second tab means two clicks and takes away real-estate
from our extension and doesn't solve the problem with extension panes in
other tabs. We also plan on adding multiple different tracks to the
Performance tab which also needs a name other than just React and
spelling out React as a prefix is too long. The Emoji is too
distracting. So it seems best to uniformly apply the symbol - albeit it
might just look like a dot to many.
Dark mode looks close to on brand:
<img width="1089" alt="Screenshot 2024-08-28 at 12 32 50 AM"
src="https://github.com/user-attachments/assets/7175a540-4241-4c26-9e4d-4d367873af57">
Fixes the bug that @alexmckenley and @mofeiZ found where setState-in-render can reset useMemoCache and cause an infinite loop. The bug was that renderWithHooksAgain() was not resetting hook state when rerendering (so useMemo values were preserved) but was resetting the updateQueue. This meant that the entire memo cache was cleared on a setState-in-render.
The fix here is to call a new helper function to clear the update queue. It nulls out other properties, but for memoCache it just sets the index back to zero.
ghstack-source-id: fc0947ce21
Pull Request resolved: https://github.com/facebook/react/pull/30889
Stacked on #30881.
Move `runWithFiberInDEV` from the recursive part of the commit phase and
instead wrap each call into user space. These should really map 1:1 with
where we're using `try/catch` since that's where we're calling into user
space.
The goal of this is to avoid the extra stack frames added by
`enableOwnerStacks` in the recursive parts to avoid stack overflow. This
way we only have a couple of extra at the end of the stack instead of a
couple of extra at every depth of the tree.
This is mostly just moves and same code extracted into utility
functions.
This is to help clarify what needs to be wrapped in try/catch and
runWithFiberInDEV. I'll do the runWithFiberInDEV changes in a follow up.
This leaves ReactCommitWork mostly to do matching on the tag and the
recursive loops.
Adds the concept of a "prerender". These special renders are spawned
whenever something suspends (and we're not already prerendering).
The purpose is to move speculative rendering work into a separate phase
that does not block the UI from updating. For example, during a
transition, if something suspends, we should not speculatively prerender
siblings that will be replaced by a fallback in the UI until *after* the
fallback has been shown to the user.
### Based on
- #30761
- #30759
---
`use` has an optimization where in some cases it can suspend the work
loop during the render phase until the data has resolved, rather than
unwind the stack and lose context. However, the current implementation
is not compatible with sibling prerendering. So I've temporarily
disabled it until the sibling prerendering has been refactored. We will
add it back in a later step.
This lets us track what a Component might suspend on from DevTools. We
could already collect this by replaying a component's Hooks but that
would be expensive to collect from a whole tree.
The thenables themselves might contain useful information but mainly
we'd want access to the `_debugInfo` on the thenables which might
contain additional information from the server.
https://github.com/facebook/react/blob/19bd26beb689e554fceb0b929dc5199be8cba594/packages/shared/ReactTypes.js#L114
In a follow up we should really do something similar in Flight to
transfer `use()` on the debugInfo of that Server Component.
## Summary
suspenseCallback feature has proved to be useful for FB Web. Let's look
at enabling the feature for the React Native build.
## How did you test this change?
Will sync the react changes with a React Native build and will verify
that performance logging is correctly notified of suspense promises
during the suspense callback.
This is a refactor of the fix in #27505.
When a transition update is scheduled by a popstate event, (i.e. a back/
forward navigation) we attempt to render it synchronously even though
it's a transition, since it's likely the previous page's data is cached.
In #27505, I changed the implementation so that it only "upgrades" the
priority of the transition for a single attempt. If the attempt
suspends, say because the data is not cached after all, from then on it
should be treated as a normal transition.
But it turns out #27505 did not work as intended, because it relied on
marking the root with pending synchronous work (root.pendingLanes),
which was never cleared until the popstate update completed.
The test scenarios I wrote accidentally worked for a different reason
related to suspending the work loop, which I'm currently in the middle
of refactoring.
Persistent renderers used the `Update` effect flag to check if a subtree
needs to be cloned. In some cases, that causes extra renders, such as
when a layout effect is triggered which only has an effect on the JS
side, but doesn't update the host components.
It's been a bit tricky to find the right places where this needs to be
set and I'm not 100% sure I got all the cases even though the tests
passed.
We use static dependency injection. We shouldn't use this dynamic
dependency injection we do for DevTools internals. There's also meta
programming like spreading and stuff that isn't needed.
This moves the config from `injectIntoDevTools` to the FiberConfig so it
can be statically resolved.
Closure Compiler has some trouble generating optimal code for this
anyway so ideally we'd refactor this further but at least this is better
and saves a few bytes and avoids some code paths (when minified).
This is not used by DevTools since it has its own implementation of it.
This function is getting removed since `findDOMNode` is getting removed
so we shouldn't keep around extra bytes unnecessarily.
There is also `findHostInstancesForRefresh` which should really be
implemented on the `react-refresh` side. Not using an injection but
that's a heavier lift and only affects `__DEV__`.
**This API is not intended to ship. This is a temporary unstable hook
for internal performance profiling.**
This PR exposes `unstable_useContextWithBailout`, which takes a compare
function in addition to Context. The comparison function is run to
determine if Context propagation and render should bail out earlier.
`unstable_useContextWithBailout` returns the full Context value, same as
`useContext`.
We can profile this API against `useContext` to better measure the cost
of Context value updates and gather more data around propagation and
render performance.
The bailout logic and test cases are based on
https://github.com/facebook/react/pull/20646
Additionally, this implementation allows multiple values to be compared
in one hook by returning a tuple to avoid requiring additional Context
consumer hooks.
This is a major nit but this avoids an extra stack frame when we're
replaying logs.
Normally the `printToConsole` frame doesn't show up because it'd be
ignore listed.
<img width="421" alt="Screenshot 2024-07-25 at 11 49 39 AM"
src="https://github.com/user-attachments/assets/81334c2f-e19e-476a-871e-c4db9dee294e">
When you expand to show ignore listed frames a ton of other frames show
up.
<img width="516" alt="Screenshot 2024-07-25 at 11 49 47 AM"
src="https://github.com/user-attachments/assets/2ab8bdfb-464c-408d-9176-ee2fabc114b6">
The annoying thing about this frame is that it's at the top of the stack
where as typically framework stuff ends up at the bottom and something
you can ignore. The user space stack comes first.
With this fix there's no longer any `printToConsole` frame.
<img width="590" alt="Screenshot 2024-07-25 at 12 09 09 PM"
src="https://github.com/user-attachments/assets/b8365d53-31f3-43df-abce-172d608d3c9c">
Am I wiling to eat the added complexity and slightly slower performance
for this nit? Definitely.
Following https://github.com/facebook/react/pull/30436
Concurrent by default strategy has been unshipped. Here we clean up the
`allowConcurrentByDefault` path and related logic/tests.
For now, this keeps the `concurrentUpdatesByDefaultOverride` argument in
`createContainer` and `createHydrationContainer` and ignores the value
to prevent more breaking changes to `react-reconciler` in the RC stage.
We still filter them before passing from server to client in Flight
Server but when presenting a native stack, we don't need to filter them.
That's left to ignore listing in the presentation.
The stacks are pretty clean regardless thanks to the bottom stack
frames.
We can also unify the owner stack formatters into one shared module
since Fizz/Flight/Fiber all do the same thing. DevTools currently does
the same thing but is forked so it can support multiple versions.
Concurrent by default has been unshipped! Let's clean it up.
Here we remove `forceConcurrentByDefaultForTesting`, which allows us to
run tests against both concurrent strategies. In the next PR, we'll
remove the actual concurrent by default code path.
Stacked on #30427.
Most hooks and such are called inside renders which already have these
on the stack but life-cycles that call out on them are useful to cut off
too.
Typically we don't create JSX in here so they wouldn't be part of owner
stacks anyway but they can be apart of plain stacks such as the ones
prefixes to console logs or printed by error dialogs.
This lets us cut off any React internals below. This should really be
possible using just ignore listing too ideally.
At this point we should maybe just build a Babel plugin that lets us
annotate a function to need to have this name.
Stacked on #30401.
Previously we were transferring the original V8 stack trace string to
the client and then parsing it there. However, really the server is the
one that knows what format it is and it should be able to vary by server
environment.
We also don't use the raw string anymore (at least not in
enableOwnerStacks). We always create the native Error stacks.
The string also made it unclear which environment it is and it was
tempting to just use it as is.
Instead I parse it on the server and make it a structured stack in the
transfer format. It also makes it clear that it needs to be formatted in
the current environment before presented.
Stacked on https://github.com/facebook/react/pull/30400 and
https://github.com/facebook/react/pull/30369
Previously we were using fake evals to recreate a stack for console
replaying and thrown errors. However, for owner stacks we just used the
raw string that came from the server.
This means that the format of the owner stack could include different
formats. Like Spidermonkey format for the client components and V8 for
the server components. This means that this stack can't be parsed
natively by the browser like when printing them as error like in
https://github.com/facebook/react/pull/30289. Additionally, since
there's no source file registered with that name and no source mapping
url, it can't be source mapped.
Before:
<img width="1329" alt="before-firefox"
src="https://github.com/user-attachments/assets/cbe03f9c-96ac-48fb-b58f-f3a224a774f4">
Instead, we need to create a fake stack like we do for the other things.
That way when it's printed as an Error it gets source mapped. It also
means that the format is consistently in the native format of the
current browser.
After:
<img width="753" alt="after-firefox"
src="https://github.com/user-attachments/assets/b436f1f5-ca37-4203-b29f-df9828c9fad3">
So this is nice because you can just take the result from
`captureOwnerStack()` and append it to an `Error` stack and print it
natively. E.g. this is what React DevTools will do.
If you want to parse and present it yourself though it's a bit awkward
though. The `captureOwnerStack()` API now includes a bunch of
`rsc://React/` URLs. These don't really have any direct connection to
the source map. Only the browser knows this connection from the eval.
You basically have to strip the prefix and then manually pass the
remainder to your own `findSourceMapURL`.
Another awkward part is that since Safari doesn't support eval sourceURL
exposed into `error.stack` - it means that `captureOwnerStack()` get an
empty location for server components since the fake eval doesn't work
there. That's not a big deal since these stacks are already broken even
for client modules for many because the `eval-source-map` strategy in
Webpack doesn't work in Safari for this same reason.
A lot of this refactoring is just clarifying that there's three kind of
ReactComponentInfo fields:
- `stack` - The raw stack as described on the original server.
- `debugStack` - The Error object containing the stack as represented in
the current client as fake evals.
- `debugTask` - The same thing as `debugStack` but described in terms of
a native `console.createTask`.
Ideally we wouldn't need to filter out React internals and it'd just be
covered by ignore listing by any downstream tool. E.g. a framework using
captureOwnerStack could have its own ignore listing. Printed owner
stacks would get browser source map ignore-listing. React DevTools could
have its own ignore list for internals. However, it's nice to be able to
provide nice owner stacks without a bunch of noise by default.
Especially on the server since they have to be serialized.
We currently call each function that calls into user space and track its
stack frame. However, this needs code for checking each one and doesn't
let us work across bundles.
Instead, we can name each of these frame something predictable by giving
the function a name.
Unfortunately, it's a common practice to rename functions or inline them
in compilers. Even if we didn't, others downstream from us or a dev-mode
minifier could. I use this `.bind()` trick to avoid minifying these
functions and ensure they get a unique name added to them in all
browsers. It's not 100% fool proof since a smart enough compiler could
also discover that the `this` value is not used and strip out the
function and then inline it but nobody does this yet at least.
This lets us find the bottom stack easily from stack traces just by
looking for the name.
Stacked on #30308.
This is now a noop module so we can stop applying the transform of
console.error using the Babel plugin in the mainline builds. I'm keeping
the transform for RN/WWW for now although it might be nice if the
transform moved into those systems as it gets synced instead of keeping
it upstream.
In jest tests we're already not running the forks for RN/WWW so we don't
need it at all there.
React transpiles some of its own `console.error` calls into a helper
that appends component stacks to those calls. However, this doesn't
cover user space `console.error` calls - which includes React helpers
that React has moved into third parties like createClass and prop-types.
The idea is that any user space component can add a warning just like
React can which is why React DevTools adds them too if they don't
already exist. Having them appended in both places is tricky because now
you have to know whether to remove them from React's logs.
Similarly it's often common for server-side frameworks to forget to
cover the `console.error` logs from other sources since React DevTools
isn't active there. However, it's also annoying to get component stacks
clogging the terminal - depending on where the log came from.
In the future `console.createTask()` will cover this use case natively
and when available we don't append them at all.
The new strategy relies on either:
- React DevTools existing to add them to React logs as well as third
parties.
- `console.createTask` being supported and surfaced.
- A third party framework showing the component stack either in an Error
Dialog or appended to terminal output.
For a third party to be able to implement this they need to be able to
get the component stack. To get the component stack from within a
`console.error` call you need to use the `React.captureOwnerStack()`
helper which is only available in `enableOwnerStacks` flag. However,
it's possible to polyfill with parent stacks using internals as a stop
gap. There's a question of whether React 19 should just go out with
`enableOwnerStacks` to expose this but regardless I think it's best it
doesn't include component stacks from the runtime for consistency.
In practice it's not really a regression though because typically either
of the other options exists and error dialogs don't implement
`console.error` overrides anyway yet. SSR terminals might miss them but
they'd only have them in DEV warnings to begin with an a subset of React
warnings. Typically those are either going to happen on the client
anyway or replayed.
Our tests are written to assert that component stacks work in various
scenarios all over the place. To ensure that this keeps working I
implement a "polyfill" that is similar to that expected a server
framework might do - in `assertConsoleErrorDev` and `toErrorDev`.
This PR doesn't yet change www or RN since they have their own forks of
consoleWithStackDev for now.
The full stack is the current execution stack (`new Error().stack`) +
the current owner stack (`React.captureOwnerStack()`).
The idea with the top frame was that when we append it to console.error
we'd include both since otherwise the true reason would be obscured
behind the little `>` to expand. So we'd just put both stack front and
center. By adding this into getCurrentStack it was easy to use the same
filtering. I never implemented in Fizz or Flight though.
However, with the public API `React.captureOwnerStack()` it's not
necessary to include the current stack since you already have it and
you'd have filtering capabilities in user space too.
Since I'm removing the component stacks from React itself we no longer
need this. It's expected that maybe RDT or framework polyfill would
include this same technique though.
While the goal is to remove legacy context completely, I think we can
already land the removal of legacy context for function components. I
didn't even know this feature existed until reading the code recently.
The win is just a couple of property lookups on function renders, but it
trims down the API already as the full removal will likely still take a
bit more time.
www: Starting with enabled test renderer and a feature flag for
production rollout.
RN: Not enabled, will follow up on this.
Object literals should be faster at least on React Native with Hermes as
the JS engine.
It might also be interesting to confirm the old comments in this file
from years ago are even still valid. Creating an object from a literal
should be a simpler operation.
It's a bit unfortunate that this introduces a bunch of copied code, but
since we rearely update the fields on fibers, this seems like an okay
tradeoff for a hot code path. An alternative would be some sort of macro
system, but that doesn't seem worth the extra complexity.
Sebastian MarkbågeandGitHub
Andrew ClarkandGitHub
Joe Savona
mofeiZandGitHub
Benoit GirardandGitHub
Sebastian SilbermannandGitHub
Jack PopeandGitHub
Tom ShermanandGitHub