`encodeReply` throws "React Element cannot be passed to Server Functions
from the Client without a temporary reference set" when a React element
is the root value of a `serializeModel` call (either passed directly or
resolved from a promise), even when a temporary reference set is
provided.
The cause is that `resolveToJSON` hits the `REACT_ELEMENT_TYPE` switch
case before reaching the `existingReference`/`modelRoot` check that
regular objects benefit from. The synthetic JSON root created by
`JSON.stringify` is never tracked in `writtenObjects`, so
`parentReference` is `undefined` and the code falls through to the
throw. This adds a `modelRoot` check in the `REACT_ELEMENT_TYPE` case,
following the same pattern used for promises and plain objects.
The added `JSX as root model` test also uncovered a pre-existing crash
in the Flight Client: when the JSX element round-trips back, it arrives
as a frozen object (client-created elements are frozen in DEV), and
`Object.defineProperty` for `_debugInfo` fails because frozen objects
are non-configurable. The same crash can occur with JSX exported as a
client reference. For now, we're adding `!Object.isFrozen()` guards in
`moveDebugInfoFromChunkToInnerValue` and `addAsyncInfo` to prevent the
crash, which means debug info is silently dropped for frozen elements.
The proper fix would likely be to clone the element so each rendering
context gets its own mutable copy with correct debug info.
closes#34984closes#35690
Requires full error message in assert helpers.
Some of the error messages we asset on add a native javascript stack
trace, which would be a pain to add to the messages and maintain. This
PR allows you to just add `\n in <stack>` placeholder to the error
message to denote a native stack trace is present in the message.
---
Note: i vibe coded this so it was a pain to backtrack this to break this
into a stack, I tried and gave up, sorry.
When dealing with optimistic state, a common problem is not knowing the
id of the thing we're waiting on. Items in lists need keys (and single
items should often have keys too to reset their state). As a result you
have to generate fake keys. It's a pain to manage those and when the
real item comes in, you often end up rendering that with a different
`key` which resets the state of the component tree. That in turns works
against the grain of React and a lot of negatives fall out of it.
This adds a special `optimisticKey` symbol that can be used in place of
a `string` key.
```js
import {optimisticKey} from 'react';
...
const [optimisticItems, setOptimisticItems] = useOptimistic([]);
const children = savedItems.concat(
optimisticItems.map(item =>
<Item key={optimisticKey} item={item} />
)
);
return <div>{children}</div>;
```
The semantics of this `optimisticKey` is that the assumption is that the
newly saved item will be rendered in the same slot as the previous
optimistic items. State is transferred into whatever real key ends up in
the same slot.
This might lead to some incorrect transferring of state in some cases
where things don't end up lining up - but it's worth it for simplicity
in many cases since dealing with true matching of optimistic state is
often very complex for something that only lasts a blink of an eye.
If a new item matches a `key` elsewhere in the set, then that's favored
over reconciling against the old slot.
One quirk with the current algorithm is if the `savedItems` has items
removed, then the slots won't line up by index anymore and will be
skewed. We might be able to add something where the optimistic set is
always reconciled against the end. However, it's probably better to just
assume that the set will line up perfectly and otherwise it's just best
effort that can lead to weird artifacts.
An `optimisticKey` will match itself for updates to the same slot, but
it will not match any existing slot that is not an `optimisticKey`. So
it's not an `any`, which I originally called it, because it doesn't
match existing real keys against new optimistic keys. Only one
direction.
This PR ensures that server components are reliably included in the
DevTools component tree, even if debug info is received delayed, e.g.
when using a debug channel. The fix consists of three parts:
- We must not unset the debug chunk before all debug info entries are
resolved.
- We must ensure that the "RSC Stream" IO debug info entry is pushed
last, after all other entries were resolved.
- We need to transfer the debug info from blocked element chunks onto
the lazy node and the element.
Ideally, we wouldn't even create a lazy node for blocked elements that
are at the root of the JSON payload, because that would basically wrap a
lazy in a lazy. This optimization that ensures that everything around
the blocked element can proceed is only needed for nested elements.
However, we also need it for resolving deduped references in blocked
root elements, unless we adapt that logic, which would be a bigger lift.
When reloading the Flight fixture, the component tree is now displayed
deterministically. Previously, it would sometimes omit synchronous
server components.
<img width="306" height="565" alt="complete"
src="https://github.com/user-attachments/assets/db61aa10-1816-43e6-9903-0e585190cdf1"
/>
---------
Co-authored-by: Sebastian Markbage <sebastian@calyptus.eu>
One thing that can suspend is the downloading of the RSC stream itself.
This tracks an I/O entry for each Promise (`SomeChunk<T>`) that
represents the request to the RSC stream. As the value we use the
`Response` for `createFromFetch` (or the `ReadableStream` for
`createFromReadableStream`). The start time is when you called those.
Since we're not awaiting the whole stream, each I/O entry represents the
part of the stream up until it got unblocked. However, in a production
environment with TLS packets and buffering in practice the chunks
received by the client isn't exactly at the boundary of each row. It's a
bit longer into larger chunks. From testing, it seems like multiples of
16kb or 64kb uncompressed are common. To simulate a production
environment we group into roughly 64kb chunks if they happen in rapid
sequence. Note that this might be too small to give a good idea because
of the throttle many boundaries might be skipped anyway so this might
show too many.
The React DevTools will see each I/O entry as separate but dedupe if an
outer boundary already depends on the same chunk. This deduping makes it
so that small boundaries that are blocked on the same chunk, don't get
treated as having unique suspenders. If you have a boundary with large
content, then that content will likely be in a separate chunk which is
not in the parent and then it gets marked as.
This is all just an approximation. The goal of this is just to highlight
that very large boundaries will very likely suspend even if they don't
suspend on any I/O on the server. In practice, these boundaries can
float around a lot and it's really any Suspense boundary that might
suspend but some are more likely than others which this is meant to
highlight.
It also just lets you inspect how many bytes needs to be transferred
before you can show a particular part of the content, to give you an
idea that it's not just I/O on the server that might suspend.
If you don't use the debug channel it can be misleading since the data
in development mode stream will have a lot more data in it which leads
to more chunking.
Similarly to "client references" these I/O infos don't have an "env"
since it's the client that has the I/O and so those are excluded from
flushing in the Server performance tracks.
Note that currently the same Response can appear many times in the same
Instance of SuspenseNode in DevTools when there are multiple chunks. In
a follow up I'll show only the last one per Response at any given level.
Note that when a separate debugChannel is used it has its own I/O entry
that's on the `_debugInfo` for the debug chunks in that channel.
However, if everything works correctly these should never leak into the
DevTools UI since they should never be propagated from a debug chunk to
the values waited by the runtime. This is easy to break though.
This creates a debug info object for the React.lazy call when it's
called on the client. We have some additional information we can track
for these since they're created by React earlier.
We can track the stack trace where `React.lazy` was called to associate
it back to something useful. We can track the start time when we
initialized it for the first time and the end time when it resolves. The
name from the promise if available.
This data is currently only picked up in child position and not
component position. The component position is in a follow up.
<img width="592" height="451" alt="Screenshot 2025-08-08 at 2 49 33 PM"
src="https://github.com/user-attachments/assets/913d2629-6df5-40f6-b036-ae13631379b9"
/>
This begs for ignore listing in the front end since these stacks aren't
filtered on the server.
This ensures that if the name is set manually after the declaration,
then we get that name when we log the value. For example Node.js
`Response` is declared as `_Response` and then later assigned a new
name.
We should probably really serialize all static enumerable properties but
"name" is non-enumerable so it's still a special case.
There's a lot of overlap between `enableComponentPerformanceTrack` and
`enableAsyncDebugInfo` because they both rely on timing information. The
former is mainly emit timestamps for how long server components and
awaits took. The latter how long I/O took.
`enableAsyncDebugInfo` is currently primarily for the component
performance track but its meta data is useful for other debug tools too.
This promotes that flag to stable.
However, `enableComponentPerformanceTrack` needs more work due to
performance concerns with Chrome DevTools so I need to separate them.
This keeps doing most of the timing tracking on the server but doesn't
emit the per-server component time stamps when
`enableComponentPerformanceTrack` is false.
Chrome DevTools Extensions has a silly problem where they block access
to load Resources from all protocols except [an allow
list](eb970fbc64/front_end/models/extensions/ExtensionServer.ts (L60)).
https://issues.chromium.org/issues/416196401
Even though these are `eval()` and not actually loaded from the network
they're blocked. They can really be any string. We just have to pick one
of:
```js
'http:', 'https:', 'file:', 'data:', 'chrome-extension:', 'about:'
```
That way React DevTools extensions can load this content to source map
them.
Webpack has the same issue with its `webpack://` and
`webpack-internal://` urls.
When we have a debug channel open that can ask for more objects. That
doesn't close until all lazy objects have been explicitly asked for. If
you GC an object before the lazy references inside of it before asking
for or releasing the objects, then it'll never close.
This ensures that if there are no more PendingChunk and no more
ResolvedModelChunk then we can close the connection.
There's two sources of retaining the Response object. On one side we
have a handle to it from the stream coming from the server. On the other
side we have a handle to it from ResolvedModelChunk to ask for more data
when we lazily parse a model.
This PR makes a weak handle from the stream to the Response. However, it
keeps a strong reference alive whenever we're waiting on a pending chunk
because then the stream might be the root if the only listeners are the
callbacks passed to the promise and no references to the promise itself.
The pending chunks count can end up being zero even if we might get more
data because the references might be inside lazy chunks. In this case
the lazy chunks keeps the Response alive. When the lazy chunk gets
parsed it can find more chunks that then end up pending to keep the
response strongly alive until they resolve.
This is using the same trick as #30798 but for runtime code too. It's
essential zero cost.
This lets us include a source location for parent stacks of Server
Components when it has an owned child's location. Either from JSX or
I/O.
Ironically, a Component that throws an error will likely itself not get
the stack because it won't have any JSX rendered yet.
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>
This adds better support for serializing class instances as Debug
values.
It adds a new marker on the object `{ "": "$P...", ... }` which
indicates which constructor's prototype to use for this object's
prototype. It doesn't encode arbitrary prototypes and it doesn't encode
any of the properties on the prototype. It might get some of the
properties from the prototype by virtue of `toString` on a `class`
constructor will include the whole class's body.
This will ensure that the instance gets the right name in logs.
Additionally, this now also invokes getters if they're enumerable on the
prototype. This lets us reify values that can only be read from native
classes.
---------
Co-authored-by: Hendrik Liebau <mail@hendrik-liebau.de>
We already support serializing the values of instrumented Promises as
debug values such as in console logs. However, we don't support plain
native promises.
This waits a microtask to see if we can read the value within a
microtask and if so emit it. This is so that we can still close the
connection.
Otherwise, we emit a "halted" row into its row id which replaces the old
"Infinite Promise" reference.
We could potentially wait until the end of the render before cancelling
so that if it resolves before we exit we can still include its value but
that would require a bit more work. Ideally we'd have a way to get these
lazily later anyway.
Stacked on #33539.
Stores dedupes of `renderConsoleValue` in a separate set. This allows us
to dedupe objects safely since we can't write objects using this
algorithm if they might also be referenced by the "real" serialization.
Also renamed it to `renderDebugModel` since it's not just for console
anymore.
On pages that have a high number of server components (e.g. common when
doing syntax highlighting), the debug outlining can produce extremely
large RSC payloads. For example a documentation page I was working on
had a 13.8 MB payload. I noticed that a majority of this was the source
code for the same function components repeated over and over again (over
4000 times) within `$E()` eval commands.
This PR deduplicates the same functions by serializing by reference,
similar to what is already done for objects. Doing this reduced the
payload size of my page from 13.8 MB to 4.6 MB, and resulted in only 31
evals instead of over 4000. As a result it reduced development page load
and hydration time from 4 seconds to 1.5 seconds. It also means the
deserialized functions will have reference equality just as they did on
the server.
Stacked on #33482.
There's a flaw with getting information from the execution context of
the ping. For the soft-deprecated "throw a promise" technique, this is a
bit unreliable because you could in theory throw the same one multiple
times. Similarly, a more fundamental flaw with that API is that it
doesn't allow for tracking the information of Promises that are already
synchronously able to resolve.
This stops tracking the async debug info in the case of throwing a
Promise and only when you render a Promise. That means some loss of data
but we should just warn for throwing a Promise anyway.
Instead, this also adds support for tracking `use()`d thenables and
forwarding `_debugInfo` from then. This is done by extracting the info
from the Promise after the fact instead of in the resolve so that it
only happens once at the end after the pings are done.
This also supports passing the same Promise in multiple places and
tracking the debug info at each location, even if it was already
instrumented with a synchronous value by the time of the second use.
Technically the async call graph spans basically all the way back to the
start of the app potentially, but we don't want to include everything.
Similarly we don't want to include everything from previous components
in every child component. So we need some heuristics for filtering out
data.
We roughly want to be able to inspect is what might contribute to a
Suspense loading sequence even if it didn't this time e.g. due to a race
condition.
One flaw with the previous approach was that awaiting a cached promise
in a sibling that happened to finish after another sibling would be
excluded. However, in a different race condition that might end up being
used so I wanted to include an empty "await" in that scenario to have
some association from that component.
However, for data that resolved fully before the request even started,
it's a little different. This can be things that are part of the start
up sequence of the app or externally cached data. We decided that this
should be excluded because it doesn't contribute to the loading sequence
in the expected scenario. I.e. if it's cached. Things that end up being
cache misses would still be included. If you want to test externally
cached data misses, then it's up to you or the framework to simulate
those. E.g. by dropping the cache. This also helps free up some noise
since static / cached data can be excluded in visualizations.
I also apply this principle to forwarding debug info. If you reuse a
cached RSC payload, then the Server Component render time and its awaits
gets clamped to the caller as if it has zero render/await time. The I/O
entry is still back dated but if it was fully resolved before we started
then it's completely excluded.
I noticed that the ThirdPartyComponent in the fixture was showing the
wrong stack and the `"use third-party"` is in the wrong location.
<img width="628" alt="Screenshot 2025-06-06 at 11 22 11 PM"
src="https://github.com/user-attachments/assets/f0013380-d79e-4765-b371-87fd61b3056b"
/>
When creating the initial JSX inside the third party server, we should
make sure that it has no owner. In a real cross-server environment you
get this by default by just executing in different context. But since
the fixture example is inside the same AsyncLocalStorage as the parent
it already has an owner which gets transferred. So we should make sure
that were we create the JSX has no owner to simulate this.
When we then parse a null owner on the receiving side, we replace its
owner/stack with the owner/stack of the call to `createFrom...` to
connect them. This worked fine with only two environments. The bug was
that when we did this and then transferred the result to a third
environment we took the original parsed stack trace. We should instead
parse a new one from the replaced stack in the current environment.
The second bug was that the `"use third-party"` badge ends up in the
wrong place when we do this kind of thing. Because the stack of the
thing entering the new environment is the call to `createFrom...` which
is in the old environment even though the component itself executes in
the new environment. So to see if there's a change we should be
comparing the current environment of the task to the owner's environment
instead of the next environment after the task.
After:
<img width="494" alt="Screenshot 2025-06-07 at 1 13 28 AM"
src="https://github.com/user-attachments/assets/e2e870ba-f125-4526-a853-bd29f164cf09"
/>
Alternative to #33421. The difference is that this also adds an
underscore between the "R" and the ID.
The reason we wanted to use special characters is because we use the
full spectrum of A-Z 0-9 in our ID generation so we can basically
collide with any common word (or anyone using a similar algorithm,
base64 or even base16). It's a little less likely that someone would put
`_R_` specifically unless you generate like two IDs separated by
underscore.
Typed errors is not a feature that Flight currently supports. However,
for presentation purposes, serializing a custom error name is something
we could support today.
With this PR, we're now transporting custom error names through the
server-client boundary, so that they are available in the client e.g.
for console replaying. One example where this can be useful is when you
want to print debug information while leveraging the fact that
`console.warn` displays the error stack, including handling of hiding
and source mapping stack frames. In this case you may want to show
`Warning: ...` or `Debug: ...` instead of `Error: ...`.
In prod mode, we still transport an obfuscated error that uses the
default `Error` name, to not leak any sensitive information from the
server to the client. This also means that you must not rely on the
error name to discriminate errors, e.g. when handling them in an error
boundary.
Stacked on #31715.
This adds profiling data for Server Components to the RSC stream (but
doesn't yet use it for anything). This is on behind
`enableProfilerTimer` which is on for Dev and Profiling builds. However,
for now there's no Profiling build of Flight so in practice only in DEV.
It's gated on `enableComponentPerformanceTrack` which is experimental
only for now.
We first emit a timeOrigin in the beginning of the stream. This provides
us a relative time to emit timestamps against for cross environment
transfer so that we can log it in terms of absolute times. Using this as
a separate field allows the actual relative timestamps to be a bit more
compact representation and preserves floating point precision.
We emit a timestamp before emitting a Server Component which represents
the start time of the Server Component. The end time is either when the
next Server Component starts or when we finish the task.
We omit the end time for simple tasks that are outlined without Server
Components.
By encoding this as part of the debugInfo stream, this information can
be forwarded between Server to Server RSC.
When we serialize debug info we should never error even though we don't
currently support everything being serialized. Since it's non-essential
dev information.
We already handle errors in the replacer but not when errors happen in
the JSON algorithm itself - such as cyclic errors.
We should ideally support cyclic objects but regardless we should
gracefully handle the errors.
When parsing stacks from third parties they may include invalid url
characters. So we need to encode them. Since these are expected to be
urls though we use just encodeURI instead of encodeURIComponent.
Normally we filter out stack frames with missing `filename` because they
can be noisy and not ignore listed. However, it's up to the
filterStackFrame function to determine whether to do it. This lets us
match `<anonymous>` stack frames in V8 parsing (they don't have line
numbers).
This was gated behind `enableOwnerStacks` since they share some code
paths but it's really part of `enableServerComponentLogs`.
This just includes the server-side regular stack on Error/replayed logs
but doesn't use console.createTask and doesn't include owner stacks.
We can't make a special getter to mark the boundary of deep
serialization (which can be used for lazy loading in the future) when
the parent object is a special object that we parse with
getOutlinedModel. Such as Map/Set and JSX.
This marks the objects that are direct children of those as not possible
to limit.
I don't love this solution since ideally it would maybe be more local to
the serialization of a specific object.
It also means that very deep trees of only Map/Set never get cut off.
Maybe we should instead override the `get()` and enumeration methods on
these instead somehow.
It's important to have it be a getter though because that's the
mechanism that lets us lazy-load more depth in the future.
renderModelDesctructive can sometimes be called direclty on Date values.
When this happens we don't first call toJSON on the Date value so we
need to explicitly handle the case where where the rendered value is a
Date instance as well. This change updates renderModelDesctructive to
account for sometimes receiving Date instances directly.
