When a server action performs a redirect, we currently initiate a `HEAD`
request to the targeted URL to verify if it has a proper RSC response.
If it does, it then invokes a GET and streams the response. This leads
to an extra request to the server which can be costly and poor for
performance. If the `GET` returns an invalid RSC response, we'll discard
the response. The client router will also see the invalid response which
will signal that it needs to perform an MPA navigation to the targeted
URL.
<!-- Thanks for opening a PR! Your contribution is much appreciated.
To make sure your PR is handled as smoothly as possible we request that
you follow the checklist sections below.
Choose the right checklist for the change(s) that you're making:
## For Contributors
### Improving Documentation
- Run `pnpm prettier-fix` to fix formatting issues before opening the
PR.
- Read the Docs Contribution Guide to ensure your contribution follows
the docs guidelines:
https://nextjs.org/docs/community/contribution-guide
### Adding or Updating Examples
- The "examples guidelines" are followed from our contributing doc
https://github.com/vercel/next.js/blob/canary/contributing/examples/adding-examples.md
- Make sure the linting passes by running `pnpm build && pnpm lint`. See
https://github.com/vercel/next.js/blob/canary/contributing/repository/linting.md
### Fixing a bug
- Related issues linked using `fixes #number`
- Tests added. See:
https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
### Adding a feature
- Implements an existing feature request or RFC. Make sure the feature
request has been accepted for implementation before opening a PR. (A
discussion must be opened, see
https://github.com/vercel/next.js/discussions/new?category=ideas)
- Related issues/discussions are linked using `fixes #number`
- e2e tests added
(https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs)
- Documentation added
- Telemetry added. In case of a feature if it's used or not.
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
## For Maintainers
- Minimal description (aim for explaining to someone not on the team to
understand the PR)
- When linking to a Slack thread, you might want to share details of the
conclusion
- Link both the Linear (Fixes NEXT-xxx) and the GitHub issues
- Add review comments if necessary to explain to the reviewer the logic
behind a change
### What?
### Why?
### How?
Closes NEXT-
Fixes #
-->
Closes NEXT-2956
Recently the serverActionReducer was updated to no longer use React's
thenable type to carry resolution/rejection information. However the
rejection reason was not updated so now when a server action fails we
were rejecting with `undefined` rather than the rejected reason. This
change updates the reject to use the rejection value.
Closes NEXT-2943
Addresses some feedback in #62821. This will re-allow implementations
like:
```js
'use server'
export const f = wrapper(async () => {})
```
Where `wrapper` creates a sync function that returns a promise. Although
it will still be silently converted to an async function under the hood.
Closes NEXT-2790
As mentioned in the new-added error messages, and the [linked
resources](https://react.dev/reference/react/use-server#:~:text=Because%20the%20underlying%20network%20calls%20are%20always%20asynchronous%2C%20%27use%20server%27%20can%20only%20be%20used%20on%20async%20functions.):
> Because the underlying network calls are always asynchronous, 'use
server' can only be used on async functions.
> https://react.dev/reference/react/use-server
It's a requirement that only async functions are allowed to be exported
and annotated with `'use server'`. Currently, we already have compiler
check so this will already error:
```js
'use server'
export function foo () {} // missing async
```
However, since exported values can be very dynamic the compiler can't
catch all mistakes like that. We also have a runtime check for all
exports in a `'use server'` function, but it only covers `typeof value
=== 'function'`.
This PR adds a stricter check for "use server" annotated values to also
make sure they're async functions (`value.constructor.name ===
'AsyncFunction'`).
That said, there are still cases like synchronously returning a promise
to make a function "async", but it's still very different by definition.
For example:
```js
const f = async () => { throw 1; return 1 }
const g = () => { throw 1; return Promise.resolve(1) }
```
Where `g()` can be synchronously caught (`try { g() } catch {}`) but
`f()` can't even if they have the same types. If we allow `g` to be a
Server Action, this behavior is no longer always true but depending on
where it's called (server or client).
Closes#62727.
### What?
Fixes https://github.com/vercel/next.js/issues/62525
Closes NEXT-2620
Since codes below do not consider searchParams, `redirect` in server
action removes all searchParams passed from client side.
93e4bb823c (diff-c809d50461027cdba7c092e564818b1172133d337abc5c513f829c94c8483dc6R186)
So I just add conditional branch for searchParams.
---
lint, prettier was applied.
Also I have done tests by commands below. and it was all passed.
```
pnpm testonly --testPathPattern "integration" -t "redirect"
```
---------
Co-authored-by: Zack Tanner <zacktanner@gmail.com>
With Server Actions, a module-level encryption can happen when you do:
```js
function wrapAction(value) {
return async function () {
'use server'
console.log(value)
}
}
const action = wrapAction('some-module-level-encryption-value')
```
...as that action will be created when requiring this module, and it
contains an encrypted argument from its closure (`value`). This
currently throws an error during build:
```
Error: Missing manifest for Server Actions. This is a bug in Next.js
at d (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/chunks/1772.js:1:15202)
at f (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/chunks/1772.js:1:16917)
at 714 (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/app/encryption/page.js:1:2806)
at t (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/webpack-runtime.js:1:127)
at 7940 (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/app/encryption/page.js:1:941)
at t (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/webpack-runtime.js:1:127)
at r (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/app/encryption/page.js:1:4529)
at /Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/app/encryption/page.js:1:4572
at t.X (/Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/webpack-runtime.js:1:1181)
at /Users/shu/Documents/git/next.js/test/e2e/app-dir/actions/.next/server/app/encryption/page.js:1:4542
```
Because during module require phase, the encryption logic can't run as
it doesn't have Server/Client references available yet (which are set
during the rendering phase).
Since both references are global singletons to the server and are
already loaded early, this fix makes sure that they're registered via
`setReferenceManifestsSingleton` before requiring the module.
Closes NEXT-2579
### Why?
I really dislike the way `.chain` works right now, it shouldn't mutate
the `BrowserInterface`, this PR changes it so it's just a pure chain
without weird side effects.
One example with the current version (before this PR):
```
const el = browser.elementByCss('#version-2')
await el.text()
// throws
await el.text()
```
### Additional Changes
- removes selenium (which is completely unused)
- updates playwright
- makes the playwright tracing not error all the time
### What
When attaching a rejection listener to a server action promise, in the case of network errors, the rejection handler would be skipped and it'd throw an error that crashes the application.
### Why
When we refactored these reducers to no longer suspend, it caused the rejection handling logic we have to no longer make sense. In this case we're working with a native promise that won't have a `status` property, so we'd re-throw the error and not call `reject`.
### How
This removes the special status handling logic and makes the rejection handler always call `reject` with the error. This will either be handled by user code or let the error bubble to an error boundary.
I also cleaned up some mutable code that is no longer needed now that these reducers aren't replayed.
Closes NEXT-1715
Closes NEXT-2323
Fixes#58638
When we apply `createActionProxy` twice to the same value, it currently
errors because we can't override the ID of it.
This PR makes sure that 1) when the value already have an ID defined, we
skip setting it again; 2) all the action IDs are being tracked even for
duplicated ones.
Note that it's technically impossible to "detect" the duplication here
(via static analyzation) and only set it once. For example:
```ts
'use server'
export async function foo () {}
export const bar = a_value_we_dont_know_yet ? foo : async () => {}
```
So, the compiler will always wrap `createActionProxy` for every exported
value and assume they're different Server Actions (hence different IDs).
With this fix, if we find that it's already defined before, we just
return the defined reference as they're strictly identical so the ID
does't matter.
Closes#54655, closes#61183.
Closes NEXT-2264
This PR fixes the cookie merging logic in Server Actions. Specifically,
when users do `cookies().set(...)` or `cookies.delete(...)` with a
`redirect()` to an internal route followed. Currently, we are just
concatenating the original cookies (request cookies) and the mutated
cookies. That introduces several issues, like it can't override or
delete an existing cookie.
Closes NEXT-2221
This PR fixes the issue where an inline Server Action gets exported. As
this isn't the designed use case for inline Server Actions (they're
supposed to be defined and used inside another closure, such as
components), we are not handling the export cases currently:
```ts
export async function action() {
"use server"
...
}
<form action={action}/>
```
...which gets compiled into:
```ts
export async function action() {} // No-op function
<form action={...actionReference...}/>
```
Note that everything works inside this module until you `import` that
action and use it in another module.
To tackle that, this PR changes how that works as described in
`server/27/output.js`.
Closes NEXT-2140
When sending post requests but it's not server action, skip logging
warning or calling non-existed server action. Instead we only log the
warning like missnig headers for server actions when it's a server
action and call the action handler when it's decoded as a function
Fixes#58152
Closes NEXT-1761
## What?
Currently there is a bug in Server Actions when you `fetch` as it uses
the same defaults (caching when not specified) as rendering, this causes
some issues as you want to read your writes in Server Actions.
This change adds the `no-store` default for Server Actions, you can
still override it by specifying `cache: 'force-cache'` for example, but
it defaults to `cache: 'no-store'`.
Fixes NEXT-1926
<!-- Thanks for opening a PR! Your contribution is much appreciated.
To make sure your PR is handled as smoothly as possible we request that
you follow the checklist sections below.
Choose the right checklist for the change(s) that you're making:
## For Contributors
### Improving Documentation
- Run `pnpm prettier-fix` to fix formatting issues before opening the
PR.
- Read the Docs Contribution Guide to ensure your contribution follows
the docs guidelines:
https://nextjs.org/docs/community/contribution-guide
### Adding or Updating Examples
- The "examples guidelines" are followed from our contributing doc
https://github.com/vercel/next.js/blob/canary/contributing/examples/adding-examples.md
- Make sure the linting passes by running `pnpm build && pnpm lint`. See
https://github.com/vercel/next.js/blob/canary/contributing/repository/linting.md
### Fixing a bug
- Related issues linked using `fixes #number`
- Tests added. See:
https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
### Adding a feature
- Implements an existing feature request or RFC. Make sure the feature
request has been accepted for implementation before opening a PR. (A
discussion must be opened, see
https://github.com/vercel/next.js/discussions/new?category=ideas)
- Related issues/discussions are linked using `fixes #number`
- e2e tests added
(https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs)
- Documentation added
- Telemetry added. In case of a feature if it's used or not.
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
## For Maintainers
- Minimal description (aim for explaining to someone not on the team to
understand the PR)
- When linking to a Slack thread, you might want to share details of the
conclusion
- Link both the Linear (Fixes NEXT-xxx) and the GitHub issues
- Add review comments if necessary to explain to the reviewer the logic
behind a change
### What?
### Why?
### How?
Closes NEXT-
Fixes #
-->
---------
Co-authored-by: Zack Tanner <zacktanner@gmail.com>
## What?
Adds two tests for server actions returning client components:
- (supported) Importing a server action from a server component. That
server action imports a client component and returns it.
- (not supported yet) Importing a server action from a client component.
The server action imports a client component and returns it.
The second case is not supported yet as it would effectively mean a
compilation loop:
`server` -> `client` -> `server` -> `client`
and if that last client component includes another server action it goes
even further:
`server` -> `client` -> `server` -> `client` -> `server` (and if that
server action includes anothe client component it goes further and
further)
Whereas currently it's only `server` -> `client` -> `server`, so it's
limited to that.
In the future we should be able to support this
server->client->server->client loop in Turbopack specifically because
Turbopack has a single module graph.
Importing the client component in a server action that is defined in the
server compiler (i.e. when created inline or when imported from a server
component) it does work correctly already.
<!-- Thanks for opening a PR! Your contribution is much appreciated.
To make sure your PR is handled as smoothly as possible we request that
you follow the checklist sections below.
Choose the right checklist for the change(s) that you're making:
## For Contributors
### Improving Documentation
- Run `pnpm prettier-fix` to fix formatting issues before opening the
PR.
- Read the Docs Contribution Guide to ensure your contribution follows
the docs guidelines:
https://nextjs.org/docs/community/contribution-guide
### Adding or Updating Examples
- The "examples guidelines" are followed from our contributing doc
https://github.com/vercel/next.js/blob/canary/contributing/examples/adding-examples.md
- Make sure the linting passes by running `pnpm build && pnpm lint`. See
https://github.com/vercel/next.js/blob/canary/contributing/repository/linting.md
### Fixing a bug
- Related issues linked using `fixes #number`
- Tests added. See:
https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
### Adding a feature
- Implements an existing feature request or RFC. Make sure the feature
request has been accepted for implementation before opening a PR. (A
discussion must be opened, see
https://github.com/vercel/next.js/discussions/new?category=ideas)
- Related issues/discussions are linked using `fixes #number`
- e2e tests added
(https://github.com/vercel/next.js/blob/canary/contributing/core/testing.md#writing-tests-for-nextjs)
- Documentation added
- Telemetry added. In case of a feature if it's used or not.
- Errors have a helpful link attached, see
https://github.com/vercel/next.js/blob/canary/contributing.md
## For Maintainers
- Minimal description (aim for explaining to someone not on the team to
understand the PR)
- When linking to a Slack thread, you might want to share details of the
conclusion
- Link both the Linear (Fixes NEXT-xxx) and the GitHub issues
- Add review comments if necessary to explain to the reviewer the logic
behind a change
### What?
### Why?
### How?
Closes NEXT-
Fixes #
-->
Closes NEXT-1873
### What?
When using a server action on an intercepted route, when submitting that
action, you'd expect it to correspond with the page you're currently on.
However if you have route interception set up, and you load the page
rather than the intercepted page, submitting the action would `POST` to
the intercepted page. This would result in a 404 error because the
action ID you're attempting to submit wouldn't be found on the requested
page.
### Why?
Interception routes rely on the `Next-Url` request header to determine
if an interception should occur via a rewrite. However, server actions
are submitted with this header as well, so the rewrite will be applied
to the `POST` request corresponding with a non-existent action, or an
action on the intercepted page.
### How?
When loading a page that has an intercepted route, `nextUrl` should be
consistent with URL derived from the flight router state tree. But when
an interception occurs via navigation, `nextUrl` will now deviate. I'm
using this to determine whether or not `Next-Url` should be forwarded
along in the `POST` request.
Closes NEXT-1436
Fixes#52591Fixes#49934
### What?
When handling a server action, in the non-progressive enhanced case,
React will attempt to parse the request body before verifying if a valid
server action is received. This results in an "Error: Connection Closed"
error being thrown, rather than ignoring the action and failing more
gracefully
### Why?
To support progressive enhancement with form actions, the `actionId`
value is added as a hidden input in the form, so the action ID from the
header shouldn't be verified until determining that we've reached the
non-PE case. ([React
ref](https://github.com/facebook/react/pull/26774)). However, in
https://github.com/vercel/next.js/pull/49187, support was added for a
URL encoded form (which is not currently used, as indicated on the PR).
Despite it not being used for server actions, it's currently possible to
trigger this codepath, ie by calling redirect in an action handler with
a 307/308 status code with some data in the URL. This would result in a
500 error.
### How?
React should not attempt to parse the URL encoded form data until after
we've verified the server action header for the non-PE case.
x-ref NEXT-1733
[Slack
context](https://vercel.slack.com/archives/C03S8ED1DKM/p1700674895218399?thread_ts=1700060786.749079&cid=C03S8ED1DKM)
### What?
A `redirect` that occurs during a fetch action will get a status code of
200, while the redirection logic is handled client-side.
### Why?
In this scenario, the redirect is handled by the client router, so no
`Location` is set on the action response. However for debugging /
logging purposes, it'd be useful to still return the same status code
used in other cases (see #58885)
### How?
Rather than selectively setting the status to 303 in the non-fetch
action case, this always applies it.
Closes NEXT-1745
### What?
Calling `redirect` or `permanentRedirect` with a route handler used by a server action will result in that POST request following the redirect. This could result in unexpected behavior, such as re-submitting an action (in the case where the redirected URL makes use of the same server action).
### Why?
By spec, 307 and 308 status codes will attempt to reuse the original request method & body on the redirected URL.
### How?
In all cases when calling a `redirect` handler inside of an action, we'll return a `303 See Other` response which is a typical status code when redirecting to a success / confirmation page as a result of a POST/PUT.
The other option would be to use 301 / 302 status codes, but since we're already doing a 303 status code [here](https://github.com/vercel/next.js/blob/canary/packages/next/src/server/app-render/action-handler.ts#L603), this aligns the behavior for the route handler case.
Closes NEXT-1733
See also: https://github.com/vercel/next.js/issues/51592#issuecomment-1810212676
[Slack x-ref](https://vercel.slack.com/archives/C03S8ED1DKM/p1700060786749079)
### What?
When the router action queue receives a bunch of async actions in quick succession, some of those requests are dropped, and as a result, anything observing pending transitions will be stuck in a pending state.
### Why?
When adding items to the action queue, the intended behavior is for new actions to be added to the end of the action queue, to be picked up by `runRemainingActions` once the in-flight action is processed. However, new actions are erroneously overwriting pending actions in the queue rather than appending them, as `actionQueue.last` might have a pending action attached to it.
### How?
This moves the assignment of `actionQueue.last` to always be in `dispatchAction`, rather than the function that processes the action, so that we always have a single spot where `last` is assigned and to prevent it from erroneously omitted/overwritten.
Fixes#59011
Utils `stringToUint8Array` and `arrayBufferToString` assume that the values are just arbitrary fixed width data. However that doesn't work when we do unicode concatenation (`actionId + arg`) which requires Text encoder/decoder to be used.
Closes#58463, closes#58579. In general any complex unicode characters will cause the same issue, for example emojis.
### Issue
In the client components world, when you're using `next/dynamic` with `ssr: false` to split chunks in pages of edge runtime, you could get the dynamic imported module still bundled in the server bundle for edge runtime. This could easily hit the bundle limit on edge runtime if you're loading a large size of non-SSR module.
This is caused by the whole chunk is still being included when we're creating the client entry. Since the client entry is imported eagerily, webpack will bundle all the modules under it, unless it's explicitly marked not being included.
### Fix
For client components, SSR rendering layer of bundle, non-SSR `next/dynamic` calls, we're transform the result of `dynamic()` call from to conditional import the dynamic loaded module.
From
```js
dynamic(() => import(...))
```
To
```js
dynamic(() => {
require.resolveWeak(...)
}, { ssr: false })
```
This will only be applied to SSR layer bundle client components non-SSR `next/dynamic` calls and only when webpack is bundling since turbopack doesn't need this. In this way, the server side will be stripped but it can still enter the module graph since we need to traverse if there's SA in client modules with using webpack API `require.resolveWeak`. And for client side bundle will still include the actual module.
Close NEXT-1703
This removes our current convention of throwing promises in reducers in
favor of returning promises that can be consumed by `use` instead. This
will help unblock some future improvements (batching, PPR)
Reducers that would typically throw a promise now return their promise.
This gets maintained by a mutable queue (initialized in hydrate) to
ensure actions are processed in-order. The queue is also responsible for
mutating state and passing it as an input to subsequent actions.
This PR does not modify reducer behavior to keep changes minimal, but
there's more cleanup that we can do in a follow-up PR to remove things
that previously assumed reducers would be replayed.
(I recommend reviewing with whitespace turned off)
---------
Co-authored-by: Tim Neutkens <tim@timneutkens.nl>
This new option specifies a list of host names that are considered safe, to accept as Server Action requests if they're different from the initial request origin. It can be very helpful when the hosted app has many layers of reverse proxies ahead.
Closes#57397.
Scope all `serverActions` config in one group "serverActions" to make it
more semantics
---------
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
### What?
Adds support for Server Actions imported by both server and client.
### Why?
If an Action is imported by both the Client and RSC layers, we need to
support them as if they're the same action.
### How?
First, we need to ensure both layers create the same action hash ids,
which we can then use to deduplicate actions imported by both layers. If
a conflict is found, we prefer the RSC layer's action.
Closes WEB-1879
---------
Co-authored-by: Tobias Koppers <tobias.koppers@googlemail.com>
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
Instead of `Readable.toWeb` we're gonna manually convert the Node.js stream to a Web stream. `toWeb` is either having a bug, or not compatible with middleware-cloned `PassThrough` streams.
Closes#56286. The case should be already covered with existing tests.
We introduced a data fetching logging before, and the control option was under experimental. After a bit experiments turns out users really loves it. We decide to move it to a stable option.
### Changes
We're going to move the `logging` option outside of `experimental`, and scope the `fetches` related config under `logging.fetches`.
```js
// next.config.js
logging?: {
fetches?: {
fullUrl?: boolean
}
}
```
This PR implements encryption and decryption for Server Action bound values that are from the closure level. Explicit `.bind` values, function arguments and module-level values are NOT handled.
### Compiler
The compiler now groups all closure bound values to an array which gets wrapped with `encrypt`. And then inside the action body, it prepends an expression to recreate the values via `await decrypt`.
Since closure-closed variables will only exist on the server layer, the encryption utility has `"server-only"` annotated.
### Encryption
During build time, a private AES-GCM encryption key is randomly generated and stored in the built server manifest. Before encrypting/decrypting, an extra round of Flight server and client will be used to serialize/deserialize the value.
When encrypting, a salt that contains the action ID is provided to prevent replay attack towards different API endpoints. The encryption key can be overridden via the `NEXT_SERVER_ACTIONS_ENCRYPTION_KEY` env variable so it can be built on multiple machines on scale.
A global singleton for storing the client reference manifest was made for Flight's serialization/deserialization as that might happen outside of rendering.
After encryption, we then serialize the ArrayBuffer as Base64 to send it to the client.
Remove the experimental `serverActions` flag
Co-authored-by: Shu Ding <3676859+shuding@users.noreply.github.com>
Co-authored-by: Jiachi Liu <4800338+huozhi@users.noreply.github.com>
This PR adds the optional `limit` parameter on String.prototype.split uses.
> If provided, splits the string at each occurrence of the specified separator, but stops when limit entries have been placed in the array. Any leftover text is not included in the array at all.
[MDN](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/String/split#syntax)
While the performance gain may not be significant for small texts, it can be huge for large ones.
I made a benchmark on the following repository : https://github.com/Yovach/benchmark-nodejs
On my machine, I get the following results:
`node index.js`
> normal 1: 570.092ms
> normal 50: 2.284s
> normal 100: 3.543s
`node index-optimized.js`
> optmized 1: 644.301ms
> optmized 50: 929.39ms
> optmized 100: 1.020s
The "benchmarks" numbers are :
- "lorem-1" file contains 1 paragraph of "lorem ipsum"
- "lorem-50" file contains 50 paragraphes of "lorem ipsum"
- "lorem-100" file contains 100 paragraphes of "lorem ipsum"
### What?
BREAKING CHANGE: Bump the minimum required Node.js version.
### Why?
Node.js 16 has reached end-of-life in September.
Bumped to `18.18.2` since it contained some security-related patches: https://nodejs.org/en/blog/vulnerability/october-2023-security-releases
### How?
Bumped `engines` where needed, upgraded the workflows.
This will allow us to remove quite a few polyfills, I'll open separate PRs for those.
This PR implements progressive enhancement for `useFormState`:
1. Inline the form state in `__next_f` and use it for hydration.
2. Encode the new form state based on the action return value, and pass
it to Flight/Fizz.
Also fixed a problem caused by inconsistent React tree structure between
static-generation and non-static-generation.
This fixes some scenarios where executing a server action after navigation can cause the action to behave incorrectly (double submitting, not resolving). There are two separate issues:
- `canonicalUrl` and `pendingNavigatePath` were not constructed using the same function (`createHrefFromUrl`) so in certain situations they'd be comparing different values
- a fulfilled inFlightServerAction should not be invoked again
Closes NEXT-1655
Closes NEXT-1654
Fixes#55845Fixes#55814Fixes#55805
### What?
Pages marked with `generateStaticParams` don't currently support server actions, and instead return a 405 Method Not Allowed, with no action being taken on the client. Additionally, pages that are marked static & use server actions are opted into dynamic rendering.
### Why?
The page that has `generateStaticParams` is marked as `isSSG` [here](ee2ec3dd1d/packages/next/src/server/base-server.ts (L1337)).
As a result, the request is short-circuited because a POST request isn't supported on static pages. Upon detecting a server action on a page marked SSG, we bypass the static cache and go straight to the lambda.
This PR introduces an experimental option to the prerender manifest that will allow for selectively bypassing the static cache
This also removes the need to bail out of static generation
Closes NEXT-1167
Closes NEXT-1453
Fixes#49408Fixes#52840Fixes#50932
## What?
In Next, rendering a route involves 3 layers:
- the routing layer, which will direct the request to the correct route to render
- the rendering layer, which will take a route and render it appropriately
- the user layer, which contains the user code
In #51831, in order to optimise the boot time of Next.js, I introduced a change that allowed the routing layer to be bundled. In this PR, I'm doing the same for the rendering layer. This is building up on @wyattjoh's work that initially split the routing and the rendering layer into separate entry-points.
The benefits of having this approach is that this allows us to compartmentalise the different part of Next, optimise them individually and making sure that serving a request is as efficient as possible, e.g. rendering a `pages` route should not need code from the `app router` to be used.
There are now 4 different rendering runtimes, depending on the route type:
- app pages: for App Router pages
- app routes: for App Router route handlers
- pages: for legacy pages
- pages api: for legacy API routes
This change should be transparent to the end user, beside faster cold boots.
## Notable changes
Doing this change required a lot of changes for Next.js under the hood in order to make the different layers play well together.
### New conventions for externals/shared modules
The big issue of bundling the rendering runtimes is that the user code needs to be able to reference an instance of a module/value created in Next during the render. This is the case when the user wants to access the router context during SSR via `next/link` for example; when you call `useContext(value)` the value needs to be the exact same reference to one as the one created by `createContext` earlier.
Previously, we were handling this case by making all files from Next that were affected by this `externals`, meaning that we were marking them not to be bundled.
**Why not keep it this way?**
The goal of this PR as stated previously was to make the rendering process as efficient as possible, so I really wanted to avoid extraneous fs reads to unoptimised code.
In order to "fix" it, I introduced two new conventions to the codebase:
- all files that explicitly need to be shared between a rendering runtime and the user code must be suffixed by `.shared-runtime` and exposed via adding a reference in the relevant `externals` file. At compilation time, a reference to a file ending with this will get re-written to the appropriate runtime.
- all files that need to be truly externals need to be suffixed by `.external`. At compilation time, a reference to it will stay as-is. This special case is needed mostly only for the async local storages that need to be shared with all three layers of Next.
As a side effect, we should be bundling more of the Next code in the user bundles, so it should be slightly more efficient.
### App route handlers are compiled on their own layer
App route handlers should be compiled in their own layer, this allows us to separate more cleanly the compilation logic here (we don't need to run the RSC logic for example).
### New rendering bundles
We now generate a prod and a dev bundle for:
- the routing server
- the app/pages SSR rendering process
- the API routes process
The development bundle is needed because:
- there is code in Next that relies on NODE_ENV
- because we opt out of the logic referencing the correct rendering runtime in dev for a `shared-runtime` file. This is because we don't need to and that Turbopack does not support rewriting an external to something that looks like this `require('foo').bar.baz` yet. We will need to fix that when Turbopack build ships.
### New development pipeline
Bundling Next is now required when developing on the repo so I extended the taskfile setup to account for that. The webpack config for Next itself lives in `webpack.config.js` and contains the logic for all the new bundles generated.
### Misc changes
There are some misc reshuffling in the code to better use the tree shaking abilities that we can now use.
fixes NEXT-1573
Co-authored-by: Alex Kirszenberg <1621758+alexkirsz@users.noreply.github.com>
The original logic here was introduced to unblock client side navigations if a server action was in flight, however this introduced a bug where subsequent actions would fetch twice after navigation.
This was happening because the promise handling was in the wrong spot: previously this would potentially cause both the `then` callback to fire while simultaneously the action reducer would handle the result. Moving it to where we're first checking if there's a pending navigation will more reliably indicate if the action was resolved after we discarded it in the reducer.
Closes NEXT-1589
Fixes#54746
Optimizes how we handle cache tags for soft tags (auto-added by Next.js)
and normal tags (added manually) and adds differentiating between
`revalidatePath('/blog/first')` and page/layout.
Soft tags are not stored across cache entry and instead auto sent along
when checking cache entries. This allows us to prevent storing
exponential amounts of tags across cache entries while still having the
relationship between them so that single path revalidation can work
properly.
x-ref: [slack
thread](https://vercel.slack.com/archives/C042LHPJ1NX/p1690586837903309)
