-
Notifications
You must be signed in to change notification settings - Fork 46.8k
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
[Fizz] Add FormatContext and Refactor Work #21103
Conversation
Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it.
We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works.
Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
Comparing: a7c5726...6dcd49a Critical size changesIncludes critical production bundles, as well as any change greater than 2%:
Significant size changesIncludes any change greater than 0.2%: Expand to show
|
Yeah I like this. I had a similar thought when working on the lazy context implementation. Right now, we have to eagerly propagate context to suspended trees (children of the Fiber that throws) because during the retry the stack of which providers changed is no longer available. To make that lazy, we'd need to store it as continuation instead. (Which is also how Cache retries work, so we could unify the implementations further). Kinda funny because that's basically what a "fiber" already is :D The lazy context implementation works by backtracking the Fiber return path to check which contexts changed. Can't do that during the retry because of the alternate pooling mechanism, but maybe post-alternate refactor we can literally just use the previous Fiber as the continuation. |
Summary: This sync includes the following changes: - **[c9aab1c9d](facebook/react@c9aab1c9d )**: react-refresh@0.10.0 //<Dan Abramov>// - **[516b76b9a](facebook/react@516b76b9a )**: [Fast Refresh] Support callthrough HOCs ([#21104](facebook/react#21104)) //<Dan Abramov>// - **[0853aab74](facebook/react@0853aab74 )**: Log all errors to console.error by default ([#21130](facebook/react#21130)) //<Sebastian Markbåge>// - **[d1294c9d4](facebook/react@d1294c9d4 )**: Add global onError handler ([#21129](facebook/react#21129)) //<Sebastian Markbåge>// - **[64983aab5](facebook/react@64983aab5 )**: Remove redundant setUpdatePriority call ([#21127](facebook/react#21127)) //<Andrew Clark>// - **[634cc52e6](facebook/react@634cc52e6 )**: Delete dead variable: currentEventWipLanes ([#21123](facebook/react#21123)) //<Andrew Clark>// - **[1102224bb](facebook/react@1102224bb )**: Fix: flushSync changes priority inside effect ([#21122](facebook/react#21122)) //<Andrew Clark>// - **[dbe98a5aa](facebook/react@dbe98a5aa )**: Move sync task queue to its own module ([#21109](facebook/react#21109)) //<Andrew Clark>// - **[3ba5c8737](facebook/react@3ba5c8737 )**: Remove Scheduler indirection ([#21107](facebook/react#21107)) //<Andrew Clark>// - **[46b68eaf6](facebook/react@46b68eaf6 )**: Delete LanePriority type ([#21090](facebook/react#21090)) //<Andrew Clark>// - **[dcd13045e](facebook/react@dcd13045e )**: Use Lane to track root callback priority ([#21089](facebook/react#21089)) //<Andrew Clark>// - **[5f21a9fca](facebook/react@5f21a9fca )**: Clean up host pointers in level 2 of clean-up flag ([#21112](facebook/react#21112)) //<Andrew Clark>// - **[32d6f39ed](facebook/react@32d6f39ed )**: [Fizz] Support special HTML/SVG/MathML tags to suspend ([#21113](facebook/react#21113)) //<Sebastian Markbåge>// - **[a77dd13ed](facebook/react@a77dd13ed )**: Delete enableDiscreteEventFlushingChange ([#21110](facebook/react#21110)) //<Andrew Clark>// - **[048ee4c0c](facebook/react@048ee4c0c )**: Use `act` in fuzz tester to flush expired work ([#21108](facebook/react#21108)) //<Andrew Clark>// - **[556644e23](facebook/react@556644e23 )**: Fix plurals ([#21106](facebook/react#21106)) //<Sebastian Markbåge>// - **[8b741437b](facebook/react@8b741437b )**: Rename SuspendedWork to Task ([#21105](facebook/react#21105)) //<Sebastian Markbåge>// - **[38a1aedb4](facebook/react@38a1aedb4 )**: [Fizz] Add FormatContext and Refactor Work ([#21103](facebook/react#21103)) //<Sebastian Markbåge>// - **[1b7e471b9](facebook/react@1b7e471b9 )**: React Native New Architecture: Support passing nativeViewTag to getInspectorDataForViewAtPoint callback, for React DevTools compat ([#21080](facebook/react#21080)) //<Joshua Gross>// - **[4a99c5c3a](facebook/react@4a99c5c3a )**: Use highest priority lane to detect interruptions ([#21088](facebook/react#21088)) //<Andrew Clark>// - **[77be52729](facebook/react@77be52729 )**: Remove LanePriority from computeExpirationTime ([#21087](facebook/react#21087)) //<Andrew Clark>// - **[3221e8fba](facebook/react@3221e8fba )**: Remove LanePriority from getBumpedLaneForHydration ([#21086](facebook/react#21086)) //<Andrew Clark>// - **[05ec0d764](facebook/react@05ec0d764 )**: Entangled expired lanes with SyncLane ([#21083](facebook/react#21083)) //<Andrew Clark>// - **[03ede83d2](facebook/react@03ede83d2 )**: Use EventPriority to track update priority ([#21082](facebook/react#21082)) //<Andrew Clark>// - **[a63f0953b](facebook/react@a63f0953b )**: Delete SyncBatchedLane ([#21061](facebook/react#21061)) //<Ricky>// - **[fa868d6be](facebook/react@fa868d6be )**: Make opaque EventPriority type a Lane internally ([#21065](facebook/react#21065)) //<Andrew Clark>// - **[eb58c3909](facebook/react@eb58c3909 )**: react-hooks/exhaustive-deps: Handle optional chained methods as dependency ([#20204](facebook/react#20204)) ([#20247](facebook/react#20247)) //<Ari Perkkiö>// - **[7b84dbd16](facebook/react@7b84dbd16 )**: Fail build on deep requires in npm packages ([#21063](facebook/react#21063)) //<Dan Abramov>// - **[2c9d8efc8](facebook/react@2c9d8efc8 )**: Add react-reconciler/constants entry point ([#21062](facebook/react#21062)) //<Dan Abramov>// - **[d0eaf7829](facebook/react@d0eaf7829 )**: Move priorities to separate import to break cycle ([#21060](facebook/react#21060)) //<Andrew Clark>// - **[435cff986](facebook/react@435cff986 )**: [Fizz] Expose callbacks in options for when various stages of the content is done ([#21056](facebook/react#21056)) //<Sebastian Markbåge>// - **[25bfa287f](facebook/react@25bfa287f )**: [Experiment] Add feature flag for more aggressive memory clean-up of deleted fiber trees ([#21039](facebook/react#21039)) //<Benoit Girard>// - **[8fe7810e7](facebook/react@8fe7810e7 )**: Remove already completed comment ([#21054](facebook/react#21054)) //<Sebastian Markbåge>// - **[6c3202b1e](facebook/react@6c3202b1e )**: [Fizz] Use identifierPrefix to avoid conflicts within the same response ([#21037](facebook/react#21037)) //<Sebastian Markbåge>// - **[dcdf8de7e](facebook/react@dcdf8de7e )**: Remove discrete lanes and priorities ([#21040](facebook/react#21040)) //<Andrew Clark>// - **[ca99ae97b](facebook/react@ca99ae97b )**: Replace some flushExpired callsites ([#20975](facebook/react#20975)) //<Ricky>// - **[1fafac002](facebook/react@1fafac002 )**: Use SyncLane for discrete event hydration ([#21038](facebook/react#21038)) //<Andrew Clark>// Changelog: [General][Changed] - React Native sync for revisions 6d3ecb7...c9aab1c jest_e2e[run_all_tests] Reviewed By: JoshuaGross Differential Revision: D27436763 fbshipit-source-id: da79a41e26bffdcdacd293178062edf098e9b58a
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
* Add format context * Let the Work node hold all working state for the recursive loop Stacks are nice and all but there's a cost to maintaining each frame both in terms of stack size usage and writing to it. * Move current format context into work * Synchronously render children of a Suspense boundary We don't have to spawn work and snapshot the context. Instead we can try to render the boundary immediately in case it works. * Lazily create the fallback work Instead of eagerly create the fallback work and then immediately abort it. We can just avoid creating it if we finish synchronously.
I'll need the equivalent of HostContext for keeping track of which namespace we're rendering in and some other things. There are also some contextual things already and we need to add some more for "New" Context and Legacy Context.
There are a couple of ways we could deal with Context. We could pass it on the stack to every function (that's what I was doing before). We could store it on something "thread local" in module scope like how "New" Context works.
In Fizz we don't need to yield in the middle of rendering so we can use the stack and we only need one "thread local". However, we might need to be able to switch between various contexts and pick up a context. That's what SuspendedWork is. A heap allocation for where to pick back up.
There's also a downside to passing everything recursively in the stack. 1) We use extra stack memory which risks running out. 2) We have to write to the stack for each call. Typically the same thing over and over. 3) We also have to save/restore it every time we context switch.
So instead, I opted to just keep the "SuspendedWork" as the place where these contextual things are stored and we can temporarily override them along the stack where we need to. It's just one extra hidden class check I think to get it from there rather than the stack so should be strictly faster.
The "New" Context implementation does need restore/reset but that's a separate later issue and dependent on implementation details.
I also took this opportunity to synchronously render the content of Suspense boundaries since we can cheaply take advantage of already being in the right context.
Note: We could actually implement Fiber like this too. We'd just create SuspendedWork if we run out of time same as what we do when we suspend. In other words, it would use a recursive stack up until a point and then create a continuation there.