Normally we take the renderClientElement path but this is an internal
fast path.
No tests because we don't run tests with console.createTask (which is
not easy since we test component stacks).
Ideally this would be covered by types but since the types don't
consider flags and DEV it doesn't really help.
We have three kinds of stacks that we send in the RSC protocol:
- The stack trace where a replayed `console.log` was called on the
server.
- The JSX callsite that created a Server Component which then later
called another component.
- The JSX callsite that created a Host or Client Component.
These stack frames disappear in native stacks on the client since
they're executed on the server. This evals a fake file which only has
one call in it on the same line/column as the server. Then we call
through these fake modules to "replay" the callstack. We then replay the
`console.log` within this stack, or call `console.createTask` in this
stack to recreate the stack.
The main concern with this approach is the performance. It adds
significant cost to create all these eval:ed functions but it should
eventually balance out.
This doesn't yet apply source maps to these. With source maps it'll be
able to show the server source code when clicking the links.
I don't love how these appear.
- Because we haven't yet initialized the client module we don't have the
name of the client component we're about to render yet which leads to
the `<...>` task name.
- The `(async)` suffix Chrome adds is still a problem.
- The VMxxxx prefix is used to disambiguate which is noisy. Might be
helped by source maps.
- The continuation of the async stacks end up rooted somewhere in the
bootstrapping of the app. This might be ok when the bootstrapping ends
up ignore listed but it's kind of a problem that you can't clear the
async stack.
<img width="927" alt="Screenshot 2024-05-28 at 11 58 56 PM"
src="https://github.com/facebook/react/assets/63648/1c9d32ce-e671-47c8-9d18-9fab3bffabd0">
<img width="431" alt="Screenshot 2024-05-28 at 11 58 07 PM"
src="https://github.com/facebook/react/assets/63648/52f57518-bbed-400e-952d-6650835ac6b6">
<img width="327" alt="Screenshot 2024-05-28 at 11 58 31 PM"
src="https://github.com/facebook/react/assets/63648/d311a639-79a1-457f-9a46-4f3298d07e65">
<img width="817" alt="Screenshot 2024-05-28 at 11 59 12 PM"
src="https://github.com/facebook/react/assets/63648/3aefd356-acf4-4daa-bdbf-b8c8345f6d4b">
This didn't actually fail before but I'm just adding an extra check.
Currently Client References are always "function" proxies so they never
fall into this branch. However, we do in theory support objects as
client references too depending on environment. We have checks
elsewhere. So this just makes that consistent.
This is necessary to simplify the component stack handling to make way
for owner stacks. It also solves some hacks that we used to have but
don't quite make sense. It also solves the problem where things like key
warnings get silenced in RSC because they get deduped. It also surfaces
areas where we were missing key warnings to begin with.
Almost every type of warning is issued from the renderer. React Elements
are really not anything special themselves. They're just lazily invoked
functions and its really the renderer that determines there semantics.
We have three types of warnings that previously fired in
JSX/createElement:
- Fragment props validation.
- Type validation.
- Key warning.
It's nice to be able to do some validation in the JSX/createElement
because it has a more specific stack frame at the callsite. However,
that's the case for every type of component and validation. That's the
whole point of enableOwnerStacks. It's also not sufficient to do it in
JSX/createElement so we also have validation in the renderers too. So
this validation is really just an eager validation but also happens
again later.
The problem with these is that we don't really know what types are valid
until we get to the renderer. Additionally, by placing it in the
isomorphic code it becomes harder to do deduping of warnings in a way
that makes sense for that renderer. It also means we can't reuse logic
for managing stacks etc.
Fragment props validation really should just be part of the renderer
like any other component type. This also matters once we add Fragment
refs and other fragment features. So I moved this into Fiber. However,
since some Fragments don't have Fibers, I do the validation in
ChildFiber instead of beginWork where it would normally happen.
For `type` validation we already do validation when rendering. By
leaving it to the renderer we don't have to hard code an extra list.
This list also varies by context. E.g. class components aren't allowed
in RSC but client references are but we don't have an isomorphic way to
identify client references because they're defined by the host config so
the current logic is flawed anyway. I kept the early validation for now
without the `enableOwnerStacks` since it does provide a nicer stack
frame but with that flag on it'll be handled with nice stacks anyway. I
normalized some of the errors to ensure tests pass.
For `key` validation it's the same principle. The mechanism for the
heuristic is still the same - if it passes statically through a parent
JSX/createElement call then it's considered validated. We already did
print the error later from the renderer so this also disables the early
log in the `enableOwnerStacks` flag.
I also added logging to Fizz so that key warnings can print in SSR logs.
Flight is a bit more complex. For elements that end up on the client we
just pass the `validated` flag along to the client and let the client
renderer print the error once rendered. For server components we log the
error from Flight with the server component as the owner on the stack
which will allow us to print the right stack for context. The factoring
of this is a little tricky because we only want to warn if it's in an
array parent but we want to log the error later to get the right debug
info.
Fiber/Fizz has a similar factoring problem that causes us to create a
fake Fiber for the owner which means the logs won't be associated with
the right place in DevTools.
This errors on the client normally but in the case the `type` is a
function - i.e. a Server Component - it wouldn't be transferred to error
on the client so you end up with a worse error message. So this just
implements the same check as ChildFiber.
Stacked on #28997.
We can use the technique of referencing an object by its row + property
name path for temporary references - like we do for deduping. That way
we don't need to generate an ID for temporary references. Instead, they
can just be an opaque marker in the slot and it has the implicit ID of
the row + path.
Then we can stash all objects, even the ones that are actually available
to read on the server, as temporary references. Without adding anything
to the payload since the IDs are implicit. If the same object is
returned to the client, it can be referenced by reference instead of
serializing it back to the client. This also helps preserve object
identity.
We assume that the objects are immutable when they pass the boundary.
I'm not sure if this is worth it but with this mechanism, if you return
the `FormData` payload from a `useActionState` it doesn't have to be
serialized on the way back to the client. This is a common pattern for
having access to the last submission as "default value" to the form
fields. However you can still control it by replacing it with another
object if you want. In MPA mode, the temporary references are not
configured and so it needs to be serialized in that case. That's
required anyway for hydration purposes.
I'm not sure if people will actually use this in practice though or if
FormData will always be destructured into some other object like with a
library that turns it into typed data, and back. If so, the object
identity is lost.
Instead of forcing an object to be outlined to be able to refer to it
later we can refer to it by the property path inside another parent
object.
E.g. this encodes such a reference as `'$123:props:children:foo:bar'`.
That way we don't have to preemptively outline object and we can dedupe
after the first time we've found it.
There's no cost on the client if it's not used because we're not storing
any additional information preemptively.
This works mainly because we only have simple JSON objects from the root
reference. Complex objects like Map, FormData etc. are stored as their
entries array in the look up and not the complex object. Other complex
objects like TypedArrays or imports don't have deeply nested objects in
them that can be referenced.
This solves the problem that we only dedupe after the third instance.
This dedupes at the second instance. It also solves the problem where
all nested objects inside deduped instances also are outlined.
The property paths can get pretty large. This is why a test on payload
size increased. We could potentially outline the reference itself at the
first dupe. That way we get a shorter ID to refer to in the third
instance.
This is the first step to experimenting with a new type of stack traces
behind the `enableOwnerStacks` flag - in DEV only.
The idea is to generate stacks that are more like if the JSX was a
direct call even though it's actually a lazy call. Not only can you see
which exact JSX call line number generated the erroring component but if
that's inside an abstraction function, which function called that
function and if it's a component, which component generated that
component. For this to make sense it really need to be the "owner" stack
rather than the parent stack like we do for other component stacks. On
one hand it has more precise information but on the other hand it also
loses context. For most types of problems the owner stack is the most
useful though since it tells you which component rendered this
component.
The problem with the platform in its current state is that there's two
ways to deal with stacks:
1) `new Error().stack`
2) `console.createTask()`
The nice thing about `new Error().stack` is that we can extract the
frames and piece them together in whatever way we want. That is great
for constructing custom UIs like error dialogs. Unfortunately, we can't
take custom stacks and set them in the native UIs like Chrome DevTools.
The nice thing about `console.createTask()` is that the resulting stacks
are natively integrated into the Chrome DevTools in the console and the
breakpoint debugger. They also automatically follow source mapping and
ignoreLists. The downside is that there's no way to extract the async
stack outside the native UI itself so this information cannot be used
for custom UIs like errors dialogs. It also means we can't collect this
on the server and then pass it to the client for server components.
The solution here is that we use both techniques and collect both an
`Error` object and a `Task` object for every JSX call.
The main concern about this approach is the performance so that's the
main thing to test. It's certainly too slow for production but it might
also be too slow even for DEV.
This first PR doesn't actually use the stacks yet. It just collects them
as the first step. The next step is to start utilizing this information
in error printing etc.
For RSC we pass the stack along across over the wire. This can be
concatenated on the client following the owner path to create an owner
stack leading back into the server. We'll later use this information to
restore fake frames on the client for native integration. Since this
information quickly gets pretty heavy if we include all frames, we strip
out the top frame. We also strip out everything below the functions that
call into user space in the Flight runtime. To do this we need to figure
out the frames that represents calling out into user space. The
resulting stack is typically just the one frame inside the owner
component's JSX callsite. I also eagerly strip out things we expect to
be ignoreList:ed anyway - such as `node_modules` and Node.js internals.
Stacked on #28798.
Add another AsyncLocalStorage to the FlightServerConfig. This context
tracks data on a per component level. Currently the only thing we track
is the owner in DEV.
AsyncLocalStorage around each component comes with a performance cost so
we only do it DEV. It's not generally a particularly safe operation
because you can't necessarily associate side-effects with a component
based on execution scope. It can be a lazy initializer or cache():ed
code etc. We also don't support string refs anymore for a reason.
However, it's good enough for optional dev only information like the
owner.
Rebasing and landing https://github.com/facebook/react/pull/28798
This PR was approved already but held back to give time for the sync.
Rebased and landing here without pushing to seb's remote to avoid
possibility of lost updates
---------
Co-authored-by: Sebastian Markbage <sebastian@calyptus.eu>
Stacked on #28853 and #28854.
React supports rendering `Iterable` and will soon support
`AsyncIterable`. As long as it's multi-shot since during an update we
may have to rerender with new inputs an loop over the iterable again.
Therefore the `Iterator` and `AsyncIterator` types are not supported
directly as a child of React - and really it shouldn't pass between
Hooks or components neither for this reason. For parity, that's also the
case when used in Server Components.
However, there is a special case when the component rendered itself is a
generator function. While it returns as a child an `Iterator`, the React
Element itself can act as an `Iterable` because we can re-evaluate the
function to create a new generator whenever we need to.
It's also very convenient to use generator functions over constructing
an `AsyncIterable`. So this is a proposal to special case the
`Generator`/`AsyncGenerator` returned by a (Async) Generator Function.
In Flight this means that when we render a Server Component we can
serialize this value as an `Iterable`/`AsyncIterable` since that's
effectively what rendering it on the server reduces down to. That way if
Fiber can receive the result in any position.
For SuspenseList this would also need another special case because the
children of SuspenseList represent "rows".
`<SuspenseList><Component /></SuspenseList>` currently is a single "row"
even if the component renders multiple children or is an iterator. This
is currently different if Component is a Server Component because it'll
reduce down to an array/AsyncIterable and therefore be treated as one
row per its child. This is different from `<SuspenseList><Component
/><Component /></SuspenseList>` since that has a wrapper array and so
this is always two rows.
It probably makes sense to special case a single-element child in
`SuspenseList` to represent a component that generates rows. That way
you can use an `AsyncGeneratorFunction` to do this.
For [`AsyncIterable`](https://github.com/facebook/react/pull/28847) we
encode `AsyncIterator` as a separate tag.
Previously we encoded `Iterator` as just an Array. This adds a special
encoding for this. Technically this is a breaking change.
This is kind of an edge case that you'd care about the difference but it
becomes more important to treat these correctly for the warnings here
#28853.
So that when we end up referring to it in more places, it's only one.
We don't do this same pattern for regular `Symbol.iterator` because we
also support the string `"@@iterator"` for backwards compatibility.
This adds support in Flight for serializing four kinds of streams:
- `ReadableStream` with objects as a model. This is a single shot
iterator so you can read it only once. It can contain any value
including Server Components. Chunks are encoded as is so if you send in
10 typed arrays, you get the same typed arrays out on the other side.
- Binary `ReadableStream` with `type: 'bytes'` option. This supports the
BYOB protocol. In this mode, the receiving side just gets `Uint8Array`s
and they can be split across any single byte boundary into arbitrary
chunks.
- `AsyncIterable` where the `AsyncIterator` function is different than
the `AsyncIterable` itself. In this case we assume that this might be a
multi-shot iterable and so we buffer its value and you can iterate it
multiple times on the other side. We support the `return` value as a
value in the single completion slot, but you can't pass values in
`next()`. If you want single-shot, return the AsyncIterator instead.
- `AsyncIterator`. These gets serialized as a single-shot as it's just
an iterator.
`AsyncIterable`/`AsyncIterator` yield Promises that are instrumented
with our `.status`/`.value` convention so that they can be synchronously
looped over if available. They are also lazily parsed upon read.
We can't do this with `ReadableStream` because we use the native
implementation of `ReadableStream` which owns the promises.
The format is a leading row that indicates which type of stream it is.
Then a new row with the same ID is emitted for every chunk. Followed by
either an error or close row.
`AsyncIterable`s can also be returned as children of Server Components
and then they're conceptually the same as fragment arrays/iterables.
They can't actually be used as children in Fizz/Fiber but there's a
separate plan for that. Only `AsyncIterable` not `AsyncIterator` will be
valid as children - just like sync `Iterable` is already supported but
single-shot `Iterator` is not. Notably, neither of these streams
represent updates over time to a value. They represent multiple values
in a list.
When the server stream is aborted we also close the underlying stream.
However, closing a stream on the client, doesn't close the underlying
stream.
A couple of possible follow ups I'm not planning on doing right now:
- [ ] Free memory by releasing the buffer if an Iterator has been
exhausted. Single shots could be optimized further to release individual
items as you go.
- [ ] We could clean up the underlying stream if the only pending data
that's still flowing is from streams and all the streams have cleaned
up. It's not very reliable though. It's better to do cancellation for
the whole stream - e.g. at the framework level.
- [ ] Implement smarter Binary Stream chunk handling. Currently we wait
until we've received a whole row for binary chunks and copy them into
consecutive memory. We need this to preserve semantics when passing
typed arrays. However, for binary streams we don't need that. We can
just send whatever pieces we have so far.
With the enableBinaryFlight flag on we should encode typed arrays and
blobs in the Reply direction too for parity.
It's already possible to pass Blobs inside FormData but you should be
able to pass them inside objects too.
We encode typed arrays as blobs and then unwrap them automatically to
the right typed array type.
Unlike the other protocol, I encode the type as a reference tag instead
of row tag. Therefore I need to rename the tags to avoid conflicts with
other tags in references. We are running out of characters though.
This is similar to #28771 but for isomorphic. We need a make over for
these dispatchers anyway so this is the first step. Also helps flush out
some internals usage that will break anyway.
It flattens the inner mutable objects onto the ReactSharedInternals.
We used to assume that outlined models are emitted before the reference
(which was true before Blobs). However, it still wasn't safe to assume
that all the data will be available because an "import" (client
reference) can be async and therefore if it's directly a child of an
outlined model, it won't be able to update in place.
This is a similar problem as the one hit by @unstubbable in #28669 with
elements, but a little different since these don't follow the same way
of wrapping.
I don't love the structuring of this code which now needs to pass a
first class mapper instead of just being known code. It also shares the
host path which is just an identity function. It wouldn't necessarily
pass my own review but I don't have a better one for now. I'd really
prefer if this was done at a "row" level but that ends up creating even
more code.
Add test for Blob in FormData and async modules in Maps.
We currently support FormData for Replies mainly for Form Actions. This
supports it in the other direction too which lets you return it from an
action as the response. Mainly for parity.
We don't really recommend that you just pass the original form data back
because the action is supposed to be able to clear fields and such but
you could potentially at least use this as the format and could clear
some fields.
We could potentially optimize this with a temporary reference if the
same object was passed to a reply in case you use it as a round trip to
avoid serializing it back again. That way the action has the ability to
override it to clear fields but if it doesn't you get back the same as
you sent.
#28755 adds support for Blobs when the `enableBinaryFlight` is enabled
which allows them to be used inside FormData too.
We currently support Blobs when passing from Client to Server so this
adds it in the other direction for parity - when `enableFlightBinary` is
enabled.
We intentionally only support the `Blob` type to pass-through, not
subtype `File`. That's because passing additional meta data like
filename might be an accidental leak. You can still pass a `File`
through but it'll appear as a `Blob` on the other side. It's also not
possible to create a faithful File subclass in all environments without
it actually being backed by a file.
This implementation isn't great but at least it works. It creates a few
indirections. This is because we need to be able to asynchronously emit
the buffers but we have to "block" the parent object from resolving
while it's loading.
Ideally, we should be able to create the Blob on the client early and
then stream in it lazily. Because the Blob API doesn't guarantee that
the data is available synchronously. Unfortunately, the native APIs
doesn't have this. We could implement custom versions of all the data
read APIs but then the blobs still wouldn't work with native APIs. So we
just have to wait until Blob accepts a stream in the constructor.
We should be able to stream each chunk early in the protocol though even
though we can't unblock the parent until they've all loaded. I didn't do
this yet mostly because of code structure and I'm lazy.
This implements the concept of a DEV-only "owner" for Server Components.
The owner concept isn't really super useful. We barely use it anymore,
but we do have it as a concept in DevTools in a couple of cases so this
adds it for parity. However, this is mainly interesting because it could
be used to wire up future owner-based stacks.
I do this by outlining the DebugInfo for a Server Component
(ReactComponentInfo). Then I just rely on Flight deduping to refer to
that. I refer to the same thing by referential equality so that we can
associate a Server Component parent in DebugInfo with an owner.
If you suspend and replay a Server Component, we have to restore the
same owner. To do that, I did a little ugly hack and stashed it on the
thenable state object. Felt unnecessarily complicated to add a stateful
wrapper for this one dev-only case.
The owner could really be anything since it could be coming from a
different implementation. Because this is the first time we have an
owner other than Fiber, I have to fix up a bunch of places that assumes
Fiber. I mainly did the `typeof owner.tag === 'number'` to assume it's a
Fiber for now.
This also doesn't actually add it to DevTools / RN Inspector yet. I just
ignore them there for now.
Because Server Components can be async the owner isn't tracked after an
await. We need per-component AsyncLocalStorage for that. This can be
done in a follow up.
Alternative to #28620.
Instead of emitting lazy references to not-yet-emitted models in the
Flight Server, this fixes the observed issue in
https://github.com/unstubbable/ai-rsc-test/pull/1 by adjusting the lazy
model resolution in the Flight Client to update stale blocked root
models, before assigning them as chunk values. In addition, the element
props are not outlined anymore in the Flight Server to avoid having to
also handle their staleness in blocked elements.
fixes#28595
Currently you can accidentally pass React Element to a Server Action. It
warns but in prod it actually works because we can encode the symbol and
otherwise it's mostly a plain object. It only works if you only pass
host components and no function props etc. which makes it potentially
error later. The first thing this does it just early hard error for
elements.
I made Lazy work by unwrapping though since that will be replaced by
Promises later which works.
Our protocol is not fully symmetric in that elements flow from Server ->
Client. Only the Server can resolve Components and only the client
should really be able to receive host components. It's not intended that
a Server can actually do something with them other than passing them to
the client.
In the case of a Reply, we expect the client to be stateful. It's
waiting for a response. So anything we can't serialize we can still pass
by reference to an in memory object. So I introduce the concept of a
TemporaryReferenceSet which is an opaque object that you create before
encoding the reply. This then stashes any unserializable values in this
set and encode the slot by id. When a new response from the Action then
returns we pass the same temporary set into the parser which can then
restore the objects. This lets you pass a value by reference to the
server and back into another slot.
For example it can be used to render children inside a parent tree from
a server action:
```
export async function Component({ children }) {
"use server";
return <div>{children}</div>;
}
```
(You wouldn't normally do this due to the waterfalls but for advanced
cases.)
A common scenario where this comes up accidentally today is in
`useActionState`.
```
export function action(state, formData) {
"use server";
if (errored) {
return <div>This action <strong>errored</strong></div>;
}
return null;
}
```
```
const [errors, formAction] = useActionState(action);
return <div>{errors}<div>;
```
It feels like I'm just passing the JSX from server to client. However,
because `useActionState` also sends the previous state *back* to the
server this should not actually be valid. Before this PR this actually
worked accidentally. You get a DEV warning but it used to work in prod.
Once you do something like pass a client reference it won't work tho. We
could perhaps make client references work by stashing where we got them
from but it wouldn't work with all possible JSX.
By adding temporary references to the action implementation this will
work again - on the client. It'll also be more efficient since we don't
send back the JSX content that you shouldn't introspect on the server
anyway.
However, a flaw here is that the progressive enhancement of this case
won't work because we can't use temporary references for progressive
enhancement since there's no in memory stash. What is worse is that it
won't error if you hydrate. ~It also will error late in the example
above because the first state is "undefined" so invoking the form once
works - it errors on the second attempt when it tries to send the error
state back again.~ It actually errors on the first invocation because we
need to eagerly serialize "previous state" into the form. So at least
that's better.
I think maybe the solution to this particular pattern would be to allow
JSX to serialize if you have no temporary reference set, and remember
client references so that client references can be returned back to the
server as client references. That way anything you could send from the
server could also be returned to the server. But it would only deopt to
serializing it for progressive enhancement. The consequence of that
would be that there's a lot of JSX that might accidentally seem like it
should work but it's only if you've gotten it from the server before
that it works. This would have to have pair them somehow though since
you can't take a client reference from one implementation of Flight and
use it with another.
The idea here is that host dispatchers are not bound to renders so we
need to be able to dispatch to them at any time. This updates the
implementation to chain these dispatchers so that each renderer can
respond to the dispatch. Semantically we don't always want every
renderer to do this for instance if Fizz handles a float method we don't
want Fiber to as well so each dispatcher implementation can decide if it
makes sense to forward the call or not. For float methods server
disaptchers will handle the call if they can resolve a Request otherwise
they will forward. For client dispatchers they will handle the call and
always forward. The choice needs to be made for each dispatcher method
and may have implications on correct renderer import order. For now we
just live with the restriction that if you want to use server and client
together (such as renderToString in the browser) you need to import the
server renderer after the client renderer.
Builds on top of #28384.
This prefixes each log with a badge similar to how we badge built-ins
like "ForwardRef" and "Memo" in the React DevTools. The idea is that we
can add such badges in DevTools for Server Components too to carry on
the consistency.
This puts the "environment" name in the badge which defaults to
"Server". So you know which source it is coming from.
We try to use the same styling as the React DevTools. We use light-dark
mode where available to support the two different color styles, but if
it's not available I use a fixed background so that it's always readable
even in dark mode.
In Terminals, instead of hard coding colors that might not look good
with some themes, I use the ANSI color code to flip
background/foreground colors in that case.
In earlier commits I had it on the end of the line similar to the
DevTools badges but for multiline I found it better to prefix it. We
could try various options tough.
In most cases we can use both ANSI and the `%c` CSS color specifier,
because node will only use ANSI and hide the other. Chrome supports both
but the color overrides ANSI if it comes later (and Chrome doesn't
support color inverting anyway). Safari/Firefox prints the ANSI, so it
can only use CSS colors.
Therefore in browser builds I exclude ANSI.
On the server I support both so if you use Chrome inspector on the
server, you get nice colors on both terminal and in the inspector.
Since Bun uses WebKit inspector and it prints the ANSI we can't safely
emit both there. However, we also can't emit just the color specifier
because then it prints in the terminal.
https://github.com/oven-sh/bun/issues/9021 So we just use a plain string
prefix for now with a bracket until that's fixed.
Screen shots:
<img width="758" alt="Screenshot 2024-02-21 at 12 56 02 AM"
src="https://github.com/facebook/react/assets/63648/4f887ffe-fffe-4402-bf2a-b7890986d60c">
<img width="759" alt="Screenshot 2024-02-21 at 12 56 24 AM"
src="https://github.com/facebook/react/assets/63648/f32d432f-f738-4872-a700-ea0a78e6c745">
<img width="514" alt="Screenshot 2024-02-21 at 12 57 10 AM"
src="https://github.com/facebook/react/assets/63648/205d2e82-75b7-4e2b-9d9c-aa9e2cbedf39">
<img width="489" alt="Screenshot 2024-02-21 at 12 57 34 AM"
src="https://github.com/facebook/react/assets/63648/ea52d1e4-b9fa-431d-ae9e-ccb87631f399">
<img width="516" alt="Screenshot 2024-02-21 at 12 58 23 AM"
src="https://github.com/facebook/react/assets/63648/52b50fac-bec0-471d-a457-1a10d8df9172">
<img width="956" alt="Screenshot 2024-02-21 at 12 58 56 AM"
src="https://github.com/facebook/react/assets/63648/0096ed61-5eff-4aa9-8a8a-2204e754bd1f">
When developing in an RSC environment, you should be able to work in a
single environment as if it was a unified environment. With thrown
errors we already serialize them and then rethrow them on the client.
Since by default we log them via onError both in Flight and Fizz, you
can get the same log in the RSC runtime, the SSR runtime and on the
client.
With console logs made in SSR renders, you typically replay the same
code during hydration on the client. So for example warnings already
show up both in the SSR logs and on the client (although not guaranteed
to be the same). You could just spend your time in the client and you'd
be fine.
Previously, RSC logs would not be replayed because they don't hydrate.
So it's easy to miss warnings for example.
With this approach, we replay RSC logs both during SSR so they end up in
the SSR logs and on the client. That way you can just stay in the
browser window during normal development cycles. You shouldn't have to
care if your component is a server or client component when working on
logical things or iterating on a product.
With this change, you probably should mostly ignore the Flight log
stream and just look at the client or maybe the SSR one. Unless you're
digging into something specific. In particular if you just naively run
both Flight and Fizz in the same terminal you get duplicates. I like to
run out fixtures `yarn dev:region` and `yarn dev:global` in two separate
terminals.
Console logs may contain complex objects which can be inspected. Ideally
a DevTools inspector could reach into the RSC server and remotely
inspect objects using the remote inspection protocol. That way complex
objects can be loaded on demand as you expand into them. However, that
is a complex environment to set up and the server might not even be
alive anymore by the time you inspect the objects. Therefore, I do a
best effort to serialize the objects using the RSC protocol but limit
the depth that can be rendered.
This feature is only own in dev mode since it can be expensive.
In a follow up, I'll give the logs a special styling treatment to
clearly differentiate them from logs coming from the client. As well as
deal with stacks.
Depends on:
- #28317
- #28320
---
Changes the behavior of the JSX runtime to pass through `ref` as a
normal prop, rather than plucking it from the props object and storing
on the element.
This is a breaking change since it changes the type of the receiving
component. However, most code is unaffected since it's unlikely that a
component would have attempted to access a `ref` prop, since it was not
possible to get a reference to one.
`forwardRef` _will_ still pluck `ref` from the props object, though,
since it's extremely common for users to spread the props object onto
the inner component and pass `ref` as a differently named prop. This is
for maximum compatibility with existing code — the real impact of this
change is that `forwardRef` is no longer required.
Currently, refs are resolved during child reconciliation and stored on
the fiber. As a result of this change, we can move ref resolution to
happen only much later, and only for components that actually use them.
Then we can remove the `ref` field from the Fiber type. I have not yet
done that in this step, though.
Same as #28327 but for Fizz.
One thing that's weird about this recoverable error is that we don't
send the regular stack for it, just the component stack it seems. This
is missing some potential information and if we move toward integrated
since stacks it would be one thing.
Also deals with symbols. Alternative to #28312.
We currently always normalize rejections or thrown values into `Error`
objects. Partly because in prod it'll be an error object and you
shouldn't fork behavior on knowing the value outside a digest. We might
want to even make the message always opaque to avoid being tempted and
then discover in prod that it doesn't work.
However, we do include the message in DEV.
If this is a non-Error object we don't know what the properties mean.
Ofc, we don't want to include too much information in the rendered
string, so we use the general `describeObjectForErrorMessage` helper.
Unfortunately it's pretty conservative about emitting values so it's
likely to exclude any embedded string atm. Could potentially expand it a
bit.
We could in theory try to serialize as much as possible and re-throw the
actual object to allow for inspection to be expanded inside devtools
which is what I plan on for consoles, but since we're normalizing to an
Error this is in conflict with that approach.
That way we can use it for debug information like component stacks and
DevTools. I used an extra stack argument in Child Fiber to track this as
it's flowing down since it's not just elements where we have this info
readily available but parent arrays and lazy can merge this into the
Fiber too. It's not great that this is a dev-only argument and I could
track it globally but seems more likely to make mistakes.
It is possible for the same debug info to appear for multiple child
fibers like when it's attached to a fragment or a lazy that resolves to
a fragment at the root. The object identity could be used in these
scenarios to infer if that's really one server component that's a parent
of all children or if each child has a server component with the same
name.
This is effectively a public API because you can use it to stash
information on Promises from a third-party service - not just Server
Components. I started outline the types for this for some things I was
planning to add but it's not final.
I was also planning on storing it from `use(thenable)` for when you
suspend on a Promise. However, I realized that there's no Hook instance
for those to stash it on. So it might need a separate data structure to
stash the previous pass over of `use()` that resets each render.
No tests yet since I didn't want to test internals but it'll be covered
once we have debugging features like component stacks.
A Flight Server can be a consumer of a stream from another Server. In
this case the meta data is attached to debugInfo properties on lazy,
Promises, Arrays or Elements that might in turn get forwarded to the
next stream. In this case we want to forward this debug information to
the client in the stream.
I also added a DEV only `environmentName` option to the Flight Server.
This lets you name the server that is producing the debug info so that
you can trace the origin of where that component is executing. This
defaults to `"server"`. DevTools could use this for badges or different
colors.
In #28123 I switched these to be lazy references. However that creates a
lazy wrapper even if they're synchronously available. We try to as much
as possible preserve the original data structure in these cases.
E.g. here in the dev outlining I only use a lazy wrapper if it didn't
complete synchronously:
https://github.com/facebook/react/pull/28272/files#diff-d4c9c509922b3671d3ecce4e051df66dd5c3d38ff913c7a7fe94abc3ba2ed72eR638
Unfortunately we don't have a data structure that tracks the status of
each emitted row. We could store the task in the map but then they
couldn't be GC:ed as they complete. We could maybe store the status of
each element but seems so heavy.
For now I just went back to direct reference which might be an issue
since it can suspend something higher up when deduped.
This adds a new DEV-only row type `D` for DebugInfo. If we see this in
prod, that's an error. It can contain extra debug information about the
Server Components (or Promises) that were compiled away during the
server render. It's DEV-only since this can contain sensitive
information (similar to errors) and since it'll be a lot of data, but
it's worth using the same stream for simplicity rather than a
side-channel.
In this first pass it's just the Server Component's name but I'll keep
adding more debug info to the stream, and it won't always just be a
Server Component's stack frame.
Each row can get more debug rows data streaming in as it resolves and
renders multiple server components in a row.
The data structure is just a side-channel and it would be perfectly fine
to ignore the D rows and it would behave the same as prod. With this
data structure though the data is associated with the row ID / chunk, so
you can't have inline meta data. This means that an inline Server
Component that doesn't get an ID otherwise will need to be outlined. The
way I outline Server Components is using a direct reference where it's
synchronous though so on the client side it behaves the same (i.e.
there's no lazy wrapper in this case).
In most cases the `_debugInfo` is on the Promises that we yield and we
also expose this on the `React.Lazy` wrappers. In the case where it's a
synchronous render it might attach this data to Elements or Arrays
(fragments) too.
In a future PR I'll wire this information up with Fiber to stash it in
the Fiber data structures so that DevTools can pick it up. This property
and the information in it is not limited to Server Components. The name
of the property that we look for probably shouldn't be `_debugInfo`
since it's semi-public. Should consider the name we use for that.
If it's a synchronous render that returns a string or number (text node)
then we don't have anywhere to attach them to. We could add a
`React.Lazy` wrapper for those but I chose to prioritize keeping the
data structure untouched. Can be useful if you use Server Components to
render data instead of React Nodes.
This used to be trivial but it's no longer trivial.
In Fizz and Fiber this is split into renderWithHooks and
finishFunctionComponent since they also support indeterminate
components.
Interestingly thanks to this unification we always call functions with
an arity of 2 which is a bit weird - with the second argument being
undefined in everything except forwardRef and legacy context consumers.
This makes Flight makes the same thing but we could also call it with an
arity of 1.
Since Flight errors early if you try to pass it a ref, and there's no
legacy context, the second arg is always undefined.
The practical change in this PR is that returning a Promise from a
forwardRef now turns it into a lazy. We previously didn't support async
forwardRef since it wasn't supported on the client. However, since
eventually this will be supported by child-as-a-promise it seems fine to
support it.
Every time we create a task we need to wait for it so we increase a ref
count. We can do this in `createTask`. This is in line with what Fizz
does too.
They differ in that Flight counts when they're actually flushed where as
Fizz decrements them when they complete.
Flight should probably count them when they complete so it's possible to
wait for the end before flushing for buffering purposes.
Server Context was never documented, and has been deprecated in
https://github.com/facebook/react/pull/27424.
This PR removes it completely, including the implementation code.
Notably, `useContext` is removed from the shared subset, so importing it
from a React Server environment would now should be a build error in
environments that are able to enforce that.
Conceptually a Server Component in the tree is the same as a Client
Component.
When we render a Server Component with a key, that key should be used as
part of the reconciliation process to ensure the children's state are
preserved when they move in a set. The key of a child should also be
used to clear the state of the children when that key changes.
Conversely, if a Server Component doesn't have a key it should get an
implicit key based on the slot number. It should not inherit the key of
its children since the children don't know if that would collide with
other keys in the set the Server Component is rendered in.
A Client Component also has an identity based on the function's
implementation type. That mainly has to do with the state (or future
state after a refactor) that Component might contain. To transfer state
between two implementations it needs to be of the same state type. This
is not a concern for a Server Components since they never have state so
identity doesn't matter.
A Component returns a set of children. If it returns a single child,
that's the same as returning a fragment of one child. So if you
conditionally return a single child or a fragment, they should
technically reconcile against each other.
The simple way to do this is to simply emit a Fragment for every Server
Component. That would be correct in all cases. Unfortunately that is
also unfortunate since it bloats the payload in the common cases. It
also means that Fiber creates an extra indirection in the runtime.
Ideally we want to fold Server Component aways into zero cost on the
client. At least where possible. The common cases are that you don't
specify a key on a single return child, and that you do specify a key on
a Server Component in a dynamic set.
The approach in this PR treats a Server Component that returns other
Server Components or Lazy Nodes as a sequence that can be folded away.
I.e. the parts that don't generate any output in the RSC payload.
Instead, it keeps track of their keys on an internal "context". Which
gets reset after each new reified JSON node gets rendered.
Then we transfer the accumulated keys from any parent Server Components
onto the child element. In the simple case, the child just inherits the
key of the parent.
If the Server Component itself is keyless but a child isn't, we have to
add a wrapper fragment to ensure that this fragment gets the implicit
key but we can still use the key to reset state. This is unusual though
because typically if you keyed something it's because it was already in
a fragment.
In the case a Server Component is keyed but forks its children using a
fragment, we need to key that fragment so that the whole set can move
around as one. In theory this could be flattened into a parent array but
that gets tricky if something suspends, because then we can't send the
siblings early.
The main downside of this approach is that switching between single
child and fragment in a Server Component isn't always going to reconcile
against each other. That's because if we saw a single child first, we'd
have to add the fragment preemptively in case it forks later. This
semantic of React isn't very well known anyway and it might be ok to
break it here for pragmatic reasons. The tests document this
discrepancy.
Another compromise of this approach is that when combining keys we don't
escape them fully. We instead just use a simple `,` separated concat.
This is probably good enough in practice. Additionally, since we don't
encode the implicit 0 index slot key, you can move things around between
parents which shouldn't really reconcile but does. This keeps the keys
shorter and more human readable.
Before, we used to reset the thenable state and extract the previous
state very early so that it's only the retried task that can possibly
consume it. This is nice because we can't accidentally consume that
state for any other node.
However, it does add a lot of branches of code that has to pass this
around. It also adds extra bytes on the stack per node. Even though it's
mostly just null.
This changes it so that where ever we can create a thenable state (e.g.
entering a component with hooks) we first extract this from the task.
The principle is that whatever could've created the thenable state in
the first place, must always be rerendered so it'll take the same code
paths to get there and so we'll always consume it.
This refactors the Flight render loop to behave more like Fizz with
similar naming conventions. So it's easier to apply similar techniques
across both. This is not necessarily better/faster - at least not yet.
This doesn't yet implement serialization by writing segments to chunks
but we probably should do that since the built-in parts that
`JSON.stringify` gets us isn't really much anymore (except serializing
strings). When we switch to that it probably makes sense for the whole
thing to be recursive.
Right now it's not technically fully recursive because each recursive
render returns the next JSON value to encode. So it's kind of like a
trampoline. This means we can't have many contextual things on the
stack. It needs to use the Server Context `__POP` trick. However, it
does work for things that are contextual only for one sequence of server
component abstractions in a row. Since those are now recursive.
An interesting observation here is that `renderModel` means that
anything can suspend while still serializing the outer siblings.
Typically only Lazy or Components would suspend but in principle a Proxy
can suspend/postpone too and now that is left serialized by reference to
a future value. It's only if the thing that we rendered was something
that can reduce to Lazy e.g. an Element that we can serialize it as a
lazy.
Similarly to how Suspense boundaries in Fizz can catch errors, anything
that can be reduced to Lazy can also catch an error rather than bubbling
it. It only errors when the Lazy resolves. Unlike Suspense boundaries
though, those things don't render anything so they're otherwise going to
use the destructive form. To ensure that throwing in an Element can
reuse the current task, this must be handled by `renderModel`, not for
example `renderElement`.
This wires up the use of `async_hooks` in the Node build (as well as the
Edge build when a global is available) in DEV mode only. This will be
used to track debug info about what suspended during an RSC pass.
Enabled behind a flag for now.
Postponing in a promise that is being serialized to the client from the
server should be possible however prior to this change Flight treated
this case like an error rather than a postpone. This fix adds support
for postponing in this position and adds a test asserting you can
successfully prerender the root if you unwrap this promise inside a
suspense boundary.
Now that we no longer support Server Context, we can now deduplicate
objects. It's not completely safe for useId but only in the same way as
it's not safe if you reuse elements on the client, so it's not a new
issue.
This also solves cyclic object references.
The issue is that we prefer to inline objects into a plain JSON format
when an object is not going to get reused. In this case the object
doesn't have an id. We could potentially serialize a reference to an
existing model + a path to it but it bloats the format and complicates
the client.
In a smarter flush phase like we have in Fizz we could choose to inline
or outline depending on what we've discovered so far before a flush. We
can't do that here since we use native stringify. However, even in that
solution you might not know that you're going to discover the same
object later so it's not perfect deduping anyway.
Instead, I use a heuristic where I mark previously seen objects and if I
ever see that object again, then I'll outline it. The idea is that most
objects are just going to be emitted once and if it's more than once
it's fairly likely you have a shared reference to it somewhere and it
might be more than two.
The third object gets deduplicated (or "detriplicated").
It's not a perfect heuristic because when we write the second object we
will have already visited all the nested objects inside of it, which
causes us to outline every nested object too even those weren't
reference more than by that parent. Not sure how to solve for that.
If we for some other reason outline an object such as if it suspends,
then it's truly deduplicated since it already has an id.
We only allow plain objects that can be faithfully serialized and
deserialized through JSON to pass through the serialization boundary.
It's a bit too expensive to do all the possible checks in production so
we do most checks in DEV, so it's still possible to pass an object in
production by mistake. This is currently exaggerated by frameworks
because the logs on the server aren't visible enough. Even so, it's
possible to do a mistake without testing it in DEV or just testing a
conditional branch. That might have security implications if that object
wasn't supposed to be passed.
We can't rely on only checking if the prototype is `Object.prototype`
because that wouldn't work with cross-realm objects which is
unfortunate. However, if it isn't, we can check wether it has exactly
one prototype on the chain which would catch the common error of passing
a class instance.
The jsx-runtime uses the ReactCurrentDispatcher from shared internals.
Recently this was moved to ReactServerSharedInternals which broke
jsx-runtime. This change moves it back to ReactSharedInternals until we
can come up with a new forking mechanism.
This lets a registered object or value be "tainted", which we block from
crossing the serialization boundary. It's only allowed to stay
in-memory.
This is an extra layer of protection against mistakes of transferring
data from a data access layer to a client. It doesn't provide perfect
protection, because it doesn't trace through derived values and
substrings. So it shouldn't be used as the only security layer but more
layers are better.
`taintObjectReference` is for specific object instances, not any nested
objects or values inside that object. It's useful to avoid specific
objects from getting passed as is. It ensures that you don't
accidentally leak values in a specific context. It can be for security
reasons like tokens, privacy reasons like personal data or performance
reasons like avoiding passing large objects over the wire.
It might be privacy violation to leak the age of a specific user, but
the number itself isn't blocked in any other context. As soon as the
value is extracted and passed specifically without the object, it can
therefore leak.
`taintUniqueValue` is useful for high entropy values such as hashes,
tokens or crypto keys that are very unique values. In that case it can
be useful to taint the actual primitive values themselves. These can be
encoded as a string, bigint or typed array. We don't currently check for
this value in a substring or inside other typed arrays.
Since values can be created from different sources they don't just
follow garbage collection. In this case an additional object must be
provided that defines the life time of this value for how long it should
be blocked. It can be `globalThis` for essentially forever, but that
risks leaking memory for ever when you're dealing with dynamic values
like reading a token from a database. So in that case the idea is that
you pass the object that might end up in cache.
A request is the only thing that is expected to do any work. The
principle is that you can derive values from out of a tainted
entry during a request. Including stashing it in a per request cache.
What you can't do is store a derived value in a global module level
cache. At least not without also tainting the object.
I do this by simply renaming the secret export name in the "subset"
bundle and this renamed version is what the FlightServer uses.
This requires us to be more diligent about always using the correct
instance of "react" in our tests so there's a bunch of clean up for
that.
We currently abort a stream either it's explicitly told to abort (e.g.
by an abortsignal). In this case we still finish writing what we have as
well as instructions for the client about what happened so it can
trigger fallback cases and log appropriately.
We also abort a request if the stream itself cancels. E.g. if you can't
write anymore. In this case we should not write anything to the outgoing
stream since it's supposed to be closed already now. However, we should
still abort the request so that more work isn't performed and so that we
can log the reason for it to the onError callback.
We should also not do any work after aborting.
There we need to stop the "flow" of bytes - so I call stopFlowing in the
cancel case before aborting.
The tests were testing this case but we had changed the implementation
to only start flowing at initial read (pull) instead of start like we
used to. As a result, it was no longer covering this case. We have to
call reader.read() in the tests to start the flow so that we need to
cancel it.
We also were missing a final assertion on the error logs and since we
were tracking them explicitly the extra error was silenced.
