We use the stack of a Promise as the start of the I/O instead of the
actual I/O since that can symbolize the start of the operation even if
the actual I/O is batched, deduped or pooled. It can also group multiple
I/O operations into one.
We want the deepest possible Promise since otherwise it would just be
the Component's Promise.
However, we don't really need deeper than the boundary between first
party and third party. We can't just take the outer most that has third
party things on the stack though because third party can have callbacks
into first party and then we want the inner one. So we take the inner
most Promise that depends on I/O that has a first party stack on it.
The realization is that for the purposes of determining whether we have
a first party stack we need to ignore async stack frames. They can
appear on the stack when we resume third party code inside a resumption
frame of a first party stack.
<img width="832" alt="Screenshot 2025-07-08 at 6 34 25 PM"
src="https://github.com/user-attachments/assets/1636f980-be4c-4340-ad49-8d2b31953436"
/>
---------
Co-authored-by: Sebastian Sebbie Silbermann <sebastian.silbermann@vercel.com>
This adds plumbing for opening a stream from the Flight Client to the
Flight Server so it can ask for more data on-demand. In this mode, the
Flight Server keeps the connection open as long as the client is still
alive and there's more objects to load. It retains any depth limited
objects so that they can be asked for later. In this first PR it just
releases the object when it's discovered on the server and doesn't
actually lazy load it yet. That's coming in a follow up.
This strategy is built on the model that each request has its own
channel for this. Instead of some global registry. That ensures that
referential identity is preserved within a Request and the Request can
refer to previously written objects by reference.
The fixture implements a WebSocket per request but it doesn't have to be
done that way. It can be multiplexed through an existing WebSocket for
example. The current protocol is just a Readable(Stream) on the server
and WritableStream on the client. It could even be sent through a HTTP
request body if browsers implemented full duplex (which they don't).
This PR only implements the direction of messages from Client to Server.
However, I also plan on adding Debug Channel in the other direction to
allow debug info (optionally) be sent from Server to Client through this
channel instead of through the main RSC request. So the `debugChannel`
option will be able to take writable or readable or both.
---------
Co-authored-by: Hendrik Liebau <mail@hendrik-liebau.de>
We added an experimental `prerender` API to flight. This change exposes
this API in stable channels prefixed as `unstable_prerender`. We have
high confidence this API should exist but because we have not yet
settled on how to handle resuming/replaying of RSC streams we may need
to change the API contract to suit future needs. This release will allow
us to get more usage out of the existing implemented functionality
without requiring you to use experimental builds which will open up
greater adoption and opportunity for feedback.
the `prerender` implementation is documented in the `react-server`
package. As with all RSC APIs implemented in bundler specific binding
packages these aren't intended to be called by end users but instead be
used by frameworks implementing React Server Components.
Previously `prerender` was exposed unprefixed and only in the
experimental channel. This PR renames the export across all channels to
`unstable_prerender` so users of this previously unprefixed api will
need to update to the unstable form. This isn't a breaking change
because it was only exposed in the experimental channel which does not
follow semver. The reason we don't expose it under both names is that
users may feature detect the unprefixed form and then when we finally do
ship it as unprefixed we may change the function signature and break
this code. Changing the name now is much safer.
Prerendering in flight is similar to prerendering in Fizz. Instead of
receiving a result (the stream) immediately a promise is returned which
resolves to the stream when the prerender is complete. The promise will
reject if the flight render fatally errors otherwise it will resolve
when the render is completed or is aborted.
This lets us ensure that we use the original V8 format and it lets us
skip source mapping. Source mapping every call can be expensive since we
do it eagerly for server components even if an error doesn't happen.
In the case of an error being thrown we don't actually always do this in
practice because if a try/catch before us touches it or if something in
onError touches it (which the default console.error does), it has
already been initialized. So we have to be resilient to thrown errors
having other formats.
These are not as perf sensitive since something actually threw but if
you want better perf in these cases, you can simply do something like
`onError(error) { console.error(error.message) }` instead.
The server has to be aware whether it's looking up original or compiled
output. I currently use the file:// check to determine if it's referring
to a source mapped file or compiled file in the fixture. A bundled app
can more easily check if it's a bundle or not.
This lets you click a stack frame on the client and see the Server
source code inline.
<img width="871" alt="Screenshot 2024-06-01 at 11 44 24 PM"
src="https://github.com/facebook/react/assets/63648/581281ce-0dce-40c0-a084-4a6d53ba1682">
<img width="840" alt="Screenshot 2024-06-01 at 11 43 37 PM"
src="https://github.com/facebook/react/assets/63648/00dc77af-07c1-4389-9ae0-cf1f45199efb">
We could do some logic on the server that sends a source map url for
every stack frame in the RSC payload. That would make the client
potentially config free. However regardless we need the config to
describe what url scheme to use since that’s not built in to the bundler
config. In practice you likely have a common pattern for your source
maps so no need to send data over and over when we can just have a
simple function configured on the client.
The server must return a source map, even if the file is not actually
compiled since the fake file is still compiled.
The source mapping strategy can be one of two models depending on if the
server’s stack traces (`new Error().stack`) are source mapped back to
the original (`—enable-source-maps`) or represents the location in
compiled code (like in the browser).
If it represents the location in compiled code it’s actually easier. You
just serve the source map generated for that file by the tooling.
If it is already source mapped it has to generate a source map where
everything points to the same location (as if not compiled) ideally with
a segment per logical ast node.
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.
If an error happens before the shell, we need to handle it. In this case
we choose the strategy of rendering a blank document and client
rendering the app. Which will intentionally have a hydration mismatch.
If a Server Action is passed to useFormState, the action may be
submitted before it has hydrated. This will trigger a full page
(MPA-style) navigation. We can transfer the form state to the next page
by comparing the key path of the hook instance.
`ReactServerDOMServer.decodeFormState` is used by the server to extract
the form state from the submitted action. This value can then be passed
as an option when rendering the new page. It must be passed during both
SSR and hydration.
```js
const boundAction = await decodeAction(formData, serverManifest);
const result = await boundAction();
const formState = decodeFormState(result, formData, serverManifest);
// SSR
const response = createFromReadableStream(<App />);
const ssrStream = await renderToReadableStream(response, { formState })
// Hydration
hydrateRoot(container, <App />, { formState });
```
If the `formState` option is omitted, then the state won't be
transferred to the next page. However, it must be passed in both places,
or in neither; misconfiguring will result in a hydration mismatch.
(The `formState` option is currently prefixed with `experimental_`)
Currently, since we use a module cache for async modules, it doesn't
automatically get updated when the module registry gets updated (HMR).
This technique ensures that if Webpack replaces the module (HMR) then
we'll get the new Promise when we require it again.
This technique doesn't work for ESM and probably not Vite since ESM will
provide a new Promise each time you call `import()` but in the
Webpack/CJS approach this Promise is an entry in the module cache and
not a promise for the entry.
I tried to replicate the original issue in the fixture but it's tricky
to replicate because 1) we can't really use async modules the same way
without compiling both server and client 2) even then I'm not quite sure
how to repro the HMR issue.
This automatically exposes `$$FORM_ACTIONS` on Server References coming
from Flight. So that when they're used in a form action, we can encode
the ID for the server reference as a hidden field or as part of the name
of a button.
If the Server Action is a bound function it can have complex data
associated with it. In this case this additional data is encoded as
additional form fields.
To process a POST on the server there's now a `decodeAction` helper that
can take one of these progressive posts from FormData and give you a
function that is prebound with the correct closure and FormData so that
you can just invoke it.
I updated the fixture which now has a "Server State" that gets
automatically refreshed. This also lets us visualize form fields.
There's no "Action State" here for showing error messages that are not
thrown, that's still up to user space.
This adds `encodeReply` to the Flight Client and `decodeReply` to the
Flight Server.
Basically, it's a reverse Flight. It serializes values passed from the
client to the server. I call this a "Reply". The tradeoffs and
implementation details are a bit different so it requires its own
implementation but is basically a clone of the Flight Server/Client but
in reverse. Either through callServer or ServerContext.
The goal of this project is to provide the equivalent serialization as
passing props through RSC to client. Except React Elements and
Components and such. So that you can pass a value to the client and back
and it should have the same serialization constraints so when we add
features in one direction we should mostly add it in the other.
Browser support for streaming request bodies are currently very limited
in that only Chrome supports it. So this doesn't produce a
ReadableStream. Instead `encodeReply` produces either a JSON string or
FormData. It uses a JSON string if it's a simple enough payload. For
advanced features it uses FormData. This will also let the browser
stream things like File objects (even though they're not yet supported
since it follows the same rules as the other Flight).
On the server side, you can either consume this by blocking on
generating a FormData object or you can stream in the
`multipart/form-data`. Even if the client isn't streaming data, the
network does. On Node.js busboy seems to be the canonical library for
this, so I exposed a `decodeReplyFromBusboy` in the Node build. However,
if there's ever a web-standard way to stream form data, or if a library
wins in that space we can support it. We can also just build a multipart
parser that takes a ReadableStream built-in.
On the server, server references passed as arguments are loaded from
Node or Webpack just like the client or SSR does. This means that you
can create higher order functions on the client or server. This can be
tokenized when done from a server components but this is a security
implication as it might be tempting to think that these are not fungible
but you can swap one function for another on the client. So you have to
basically treat an incoming argument as insecure, even if it's a
function.
I'm not too happy with the naming parity:
Encode `server.renderToReadableStream` Decode: `client.createFromFetch`
Decode `client.encodeReply` Decode: `server.decodeReply`
This is mainly an implementation details of frameworks but it's annoying
nonetheless. This comes from that `renderToReadableStream` does do some
"rendering" by unwrapping server components etc. The `create` part comes
from the parity with Fizz/Fiber where you `render` on the server and
`create` a root on the client.
Open to bike-shedding this some more.
---------
Co-authored-by: Josh Story <josh.c.story@gmail.com>
We always look up these references in a map so it doesn't matter what
their value is. It could be a hash for example.
The loaders now encode a single $$id instead of filepath + name.
This changes the react-client-manifest to have a single level. The value
inside the map is still split into module id + export name because
that's what gets looked up in webpack.
The react-ssr-manifest is still two levels because that's a reverse
lookup.
Builds on #26257.
To do this we need access to a manifest for which scripts and CSS are
used for each "page" (entrypoint).
The initial script to bootstrap the app is inserted with
`bootstrapScripts`. Subsequent content are loaded using the chunks
mechanism built-in.
The stylesheets for each pages are prepended to each RSC payload and
rendered using Float. This doesn't yet support styles imported in
components that are also SSR:ed nor imported through Server Components.
That's more complex and not implemented in the node loader.
HMR doesn't work after reloads right now because the SSR renderer isn't
hot reloaded because there's no idiomatic way to hot reload ESM modules
in Node.js yet. Without killing the HMR server. This leads to hydration
mismatches when reloading the page after a hot reload.
Notably this doesn't show serializing the stream through the HTML like
real implementations do. This will lead to possible hydration mismatches
based on the data. However, manually serializing the stream as a string
isn't exactly correct due to binary data. It's not the idiomatic way
this is supposed to work. This will all be built-in which will make this
automatic in the future.
This proxies requests through the global server instead of requesting
RSC responses from the regional server. This is a bit closer to
idiomatic, and closer to SSR.
This also wires up HMR using the Middleware technique instead of server.
This will be an important part of RSC compatibility because there will
be a `react-refresh` aspect to the integration.
This convention uses `Accept` header to branch a URL between HTML/RSC
but it could be anything really. Special headers, URLs etc. We might be
more opinionated about this in the future but now it's up to the router.
Some fixes for Node 16/17 support in the loader and fetch polyfill.
This lets us put it in the same server that would be serving this
content in a more real world scenario.
I also de-CRA:ified this a bit by simplifying pieces we don't need.
I have more refactors coming for the SSR pieces but since many are
eyeing these fixtures right now I figured I'd push earlier.
The design here is that there are two servers:
- Global - representing a "CDN" which will also include the SSR server.
- Regional - representing something close to the data with low waterfall
costs which include the RSC server.
This is just an example.
These are using the "unbundled" strategy for the RSC server just to show
a simple case, but an implementation can use a bundled SSR server.
A smart SSR bundler could also put RSC and SSR in the same server and
even the same JS environment. It just need to ensure that the module
graphs are kept separately - so that the `react-server` condition is
respected. This include `react` itself. React will start breaking if
this isn't respected because the runtime will get the wrong copy of
`react`. Technically, you don't need the *entire* module graph to be
separated. It just needs to be any part of the graph that depends on a
fork. Like if "Client A" -> "foo" and "Server B" -> "foo", then it's ok
for the module "foo" to be shared. However if "foo" -> "bar", and "bar"
is forked by the "react-server" condition, then "foo" also needs to be
duplicated in the module graph so that it can get two copies of "bar".
Sebastian MarkbågeandSebastian Sebbie Silbermann
Josh Story
Andrew Clark
