This uses an IPC server (if available) for incremental cache methods to help prevent race conditions when reading/writing from cache and also to dedupe requests in cases where multiple cache reads are in flight. This cuts down on data fetching across the different build-time workers.
Co-authored-by: JJ Kasper <22380829+ijjk@users.noreply.github.com>
### Why
In #52985 the not found solution introduces `NODE_ENV` to determine if it his the not found boundary and should render the not found, as in the next build mode, we have `/_not-found` as a special route which has a empty parallel route, but in next dev mode so far it his the `parallel-default-route`. This could dependend on the `NODE_ENV` passing to next server but not necessarily.
### What
Fixes#53082Fixes#53083
### How
When server actions `not-found` hits, now we create a new loader tree based on the previous one, including `layout` and other components but not the children parallel routes
For production case, to make the rendering independent from the `NODE_ENV`, we're using original pathname to check if it's `/_not-found` to determine if it's production build 404 page
To support replace the loader tree of action, did a little refactor that passing down the loader tree from top level to `bodyResult`. Then we can change the loader tree itself before rendering, in short, we tweak it from original tree to one for not-found case, so server actions could render it properly
### Why
We calculate the “next url” depending on the router state and the previous router state so that when you navigate to a route, the proxy matches with that header and returns you the intercepted route if matching
### What
- Fixes#52745
Co-authored-by: Jiachi Liu <4800338+huozhi@users.noreply.github.com>
This breaks out routing handling from `next-server`, `next-dev-server`,
and `base-server` so that these are only handling the "render" work and
eventually these will be deleted completely in favor of the bundling
work being done.
The `router` process and separate `render` processes are still
maintained here although will be investigated further in follow-up to
see if we can reduce the need for these.
We are also changing the `require-cache` IPC to a single call instead of
call per entry to reduce overhead and also de-dupes handling for
starting the server between the standalone server and normal server.
To maintain support for existing turbopack route resolving this
implements the new route resolving in place of the existing
`route-resolver` until the new nextturbo API is fully landed.
After these initial changes we should continue to eliminate non-render
related code from `next-server`, `base-server`, and `next-dev-server`.
This PR tries to address some feedback around prefetching, like in #49607, where they were some warnings because we were over prefetching.
The tweaks in this PR:
- if there are no loading boundary below, we don't prefetch the full page anymore. I made that change a while ago but I think it wasn't the original intent from @sebmarkbage. Fixing that now. So now, we will prefetch the page content until the nearest loading boundary, only if there is any.
- there's now a queue for limiting the number of concurrent prefetches. This is to not ruin the bandwidth for complex pages. Thanks @alvarlagerlof for that one.
- also, if the prefetch is in the queue when navigating, it will get bumped.
- bonus: fixes a bug where we were wrongly stripping headers in dev for static pages
Test plan:
<img width="976" alt="CleanShot 2023-07-20 at 17 53 43@2x" src="https://github.com/vercel/next.js/assets/11064311/2ea34419-c002-4aea-94df-57576e3ecb2e">
In the screenshot you can see that:
- only 5 requests at a time are authorised
- when I clicked on 15, it got prioritised in the queue (did not cancel the rest)
- the prefetch only included the content until the loading boundary
Co-authored-by: JJ Kasper <22380829+ijjk@users.noreply.github.com>
### What
This PR changes the flow of not-found rendering process.
### Why
`not-found.js` was rendered in two ways before:
* 1 is SSR rendering the not-found as 404
* 2 is triggering the error on RSC rendering then the error will be
preserved in inline flight data, on the client it will recover the error
and trigger the proper error boundary.
The solution has been through a jounery:
No top-level not found boundary -> introduce metadata API -> then we
create a top level root not found boundary -> then we delete it due to
duplicated rendering of root layout -> now this
So the solution before this PR is still having a root not found boundary
wrapped in the `AppRouter`, it's being used in a lot of places including
HMR. As we discovered it's doing duplicated rendering of root layout,
then we removed it and it started failing with rendering `not-found` but
missing root layout. In this PR we redesign the process.
### How
Now the rendering architecture looks like:
* For normal root not-found and certain level of not-found boundary
they're still covered by `LayoutRouter`
* For other error renderings including not-found
* Fully remove the top level not-found boundary, when it renders with
404 error it goes to render the fallback page
* During rendering the fallback page it will check if it should just
renders a 404 error page or render nothing and let the error from inline
flight data to trigger the error boundary
pseudo code
```
try {
render AppRouter > PageComponent
} catch (err) {
create ErrorComponent by determine err
render AppRouter > ErrorComponent
}
```
In this way if the error is thrown from top-level like the page itself
or even from metadata, we can still catch them and render the proper
error page based on the error type.
The problematic is the HMR: introduces a new development mode meta tag
`<meta name="next-error">` to indicate it's 404 so that we don't do
refresh. This reverts the change brougt in #51637 as it will also has
the duplicated rendering problem for root layout if it's included in the
top level not found boundary.
Also fixes the root layout missing issue:
Fixes#52718Fixes#52739
---------
Co-authored-by: Shu Ding <g@shud.in>
We have the logic to group the client compiler's entry names to make sure we generate one single manifest file for the page. This is complicated and requires a special step to "group" the entry names because a page can depend on a bunch of files from everywhere.
And currently, the normalization of "entryName → groupName" doesn't cover interception routes' conventions (`(.)`, `(..)` and `(...)`). This PR fixes that.
Closes#52862, closes#52681, closes#52958.
### What?
- Visiting a page in the app router without a proper component export doesn't show the dev overlay, but logs errors to the console
- When it does show the error overlay (e.g. during an HMR event), the error message was sharing the module code itself rather than the component path, making it hard to debug
### Why?
`createComponentTree` can throw these errors before the AppRouter tree is mounted, leaving the errors uncaught by the dev overlay.
### How?
This wraps the server root in the `ReactDevOverlay` when in dev mode with a minimal "HMR" for when the server component is edited (to reload the page).
Closes NEXT-308
This updates our `moduleResolution` to `bundler` as this matches our heuristics much more closely so is more accurate. This shouldn't be a breaking change is it should be compatible with our previous default.
Co-authored-by: JJ Kasper <22380829+ijjk@users.noreply.github.com>
When doing dynamic imports like `import(variable)`, Webpack _tries_ to statically analyze it and creates a regex like context module for it (which includes all possible modules). This `ContextModule` doesn't have a resource path so we need to use the identifier to track it.
Tested with @alexkirsz's repro here https://github.com/vercel/next.js/issues/50243#issuecomment-1628675346 and confirmed that it fixes the problem.
Closes#50243.
This ensures we properly detect `not-found` as `runtime = 'edge'` and treat it as such instead of attempting to render it like normal.
x-ref: https://github.com/vercel/next.js/issues/52648
### What?
Support metadata exports for `not-found.js` conventions
### Why?
We want to define metadata such as title or description basic properties for error pages, including 404 and 500 which referrs to `error.js` and `not-found.js` convention. See more requests in #45620
Did some research around metadata support for not-found and error convention. It's possible to support in `not-found.js` when it's server components as currently metadata is only available but for `error.js` it has to be client components for now so it's hard to support it for now as it's a boundary.
### How?
We determine the convention if we're going to render is page or `not-found` boundary then we traverse the loader tree based on the convention type. One special case is for redirection the temporary metadata is not generated yet, we leave it as default now.
Fixes#52636
### What
This was an issue introduced in #52416 when we check the images and icons property while merging with static file conventions. Where we should check if the properties are present but we only checked if they value is falsy. So when you're using `icons` with array value it breaks.
Now we properly check all the possible case of `metadata.icons` value, so then decide if we are using the static file convention or the icons property of defined metadata.
Also add few more tests cases to cover icons resolving.
References for Client Components need to be aggregated to the page entry level, and emitted as files in the correct directory for the SSR server to read from. For normal routes (e.g. `app/foo/bar/page`), we can go through and collect all entries (layout, loading, error, ...) from the current and parent segments, to aggregate all necessary client references.
However, for routes with special conventions like `app/(group)/@named/foo/bar/page`, it needs to be normalized (remove the named slot and group segments) so it can be grouped together with `app/(group)/@named2/foo/bar/loading`.
When there's a runtime error showing in root layout (server components), it should be able to catch by `global-error`.
For server components, we caught it and gonna render the error fallback components (either not-found or error page), and the response status is `200`, and since we'll display error dev overlay in developmenet mode so we only render `global-error` for production.
So that you can catch more errors with `global-error` and maybe do potential error tracking on client side.
Follow up of #52573
Closes NEXT-1442
minor refactor: move `appUsingSizeAdjust` into `Metadata` component so that we can just tune the flag as prop
We have some special bundle path handling logic for `/index` routes in
`normalizePagePath`, which is missing in the new
`AppBundlePathNormalizer`. This already broke `/index/page.js` in dev in
the past, and now become noticeable in prod as well because of the
manifest change.
b98469c86b/packages/next/src/shared/lib/page-path/normalize-page-path.ts (L5-L14)
Co-authored-by: kodiakhq[bot] <49736102+kodiakhq[bot]@users.noreply.github.com>
Introduce a new way to search for `not-found` component that based on
the request pathname and current loader tree of that route. And we
search the proper not-found in the finall catch closure of app
rendering, so that we don't have to pass down the root layout to
app-router to create the extra error boundary.
This ensures the root layout doesn't have duplicated rendering for
normal requests
Fixes NEXT-1220
Fixes#49115
Transitions the App Pages renderer into the entrypoint bundle.
- Adjusts the static path detection to handle route module's with App Pages
- Fixes bug in font manifest loading on Edge
This PR ensures that both Webpack and the config won't be initiated in render workers. This is great for performance but also avoids potential issues (e.g. Next.js plugin with side effects). Instead, we pass a serialized config from the router worker to the render workers.
Closes#52366.
Previously `global-error` only caught the error on client side, this PR adds the support for catching the errors thrown during client components SSR or server components RSC rendering.
Closes#46572Closes#50119Closes#50723
The problem was introduced in #52450, that the client reference manifest isn't being tracked and included in the function.
Verified that this fixes the issue.
This PR changes client manifest generation process. Instead of one big
manifest file that contains client references for the entire app, we're
now generating one manifest file per entry which only covers client
components that can be reached in the module graph.
Metadata API should provide a way to override the filebased metadata
images. As usually for child routes, if there's new social images or
icons are provided, the ones from parent routes should be overridden /
skipped.
The `metadata` object export or `generateMetadata` should be able to do
that. Sometimes users still add other og info (besides images) to
metadata export (both object and `generateMetadata`).
I think we should check if they really have returned images property,
then decide to override.
- For the same level of routes:
- If there's no `openGraph.images` in the returned value, merge with
file based images
- If there's any `openGraph.images` in the returned value, ignore file
based ones
- For child level of routes:
Always override the parent level, ignoring parent level file based
images unless they use `generateMetadata` to merge from
`resolvingParentMetadata` value, then the parent level's file based ones
will present there
Closes NEXT-1418
After we separating the metadata routes to a separate layer, we didn't apply the webpack alias rules properly to it as it's should still be treated as pure "server" side
This PR fixes the aliasing for that new metadata layer and make it working properly with "server-only"
Fixes#52390
### What?
HMR causes infinite reloads for parallel routes when the corresponding page component is nested
### Why?
In 4900fa21b078fd1ec1adc5d570fcfb560be8aeb6, code was added to remove `/@children` from the page path (if present) but in 59b36349eb86427ac7b679ac62fa6628c9fc4886, `normalizeParallelKey` removes the @ prefix from children, so this doesn't seem to be catching the scenario it was intended to prevent
### How?
This updates the existing replace logic to consider `/children/page` rather than `/@children/page` -- it doesn't seem like `/@children` is a valid scenario given the `normalizeParallelKey` behavior
Fixes#52342 and addresses the concerns in https://github.com/vercel/next.js/pull/52061#issuecomment-1619145129
Adds a regression test for testing progressive enhancement of Server
Actions. Both when passed from a Server Component and when imported into
a Client Component.
#51723 landed a bit too early which broke this but it'll be fixed again
once React is upgraded.
Co-authored-by: Shu Ding <g@shud.in>
### Issue
x-ref:
https://github.com/vercel/next.js/actions/runs/5469070005/jobs/9957658991?pr=52282#step:27:526
metadata tests are failing as flaky recently, then after digging into
it, noticed it as a hmr issue.
**Steps to repro with metadata test app**
The later (after the 1st one) visited page with client components
imports sometimes throw an full reload refresh warning
> Fast Refresh will perform a full reload when you edit a file that's
imported by modules
outside of the React rendering tree.
Turns out there's some unexpected hmr events when `"sync"` event is
triggered, and client tries to apply the updates but then failed. This
PR excludes the triggering by `"sync"` for RSC pages updates. `'sync'`
event with errors/warnings will still be handled and then `'built'`
event will be handled with hot reload updates
Possibly related to #40184
### Issue
When the og module is a shared module being imported in both page and metadata image routes, it will be shared in the module graph. Especially in the edge runtime, since the `default` export is being used in the metadata image routes, then it can't be easily tree-shaked out.
### Solution
Separate the image route to a separate layer which won't share modules with the page, so that image route is always bundling separately and the `@vercel/og` module only stays inside in that layer, when import image metadata named exports (size / alt / etc..) it can be still tree shaked.
Co-authored-by: Jiachi Liu <4800338+huozhi@users.noreply.github.com>
### What?
When there's a parallel route adjacent to a tree that has no page
component, it's treated as an invalid entry in `handleAppPing` during
dev HMR, which causes an infinite refresh cycle
### Why?
In #51413, an update was made to `next-app-loader` to support layout
files in parallel routes. Part of this change updated the parallel
segment matching logic to mark the parallel page entry as `[
'@parallel', [ 'page$' ] ]` rather than `[ '@parallel', 'page$' ]`.
This resulted in `handleAppPing` looking for the corresponding page
entry at `client@app@/@parallel/page$/page` (note the `PAGE_SEGMENT`
marker) rather than `client@app@/@parallel/page`, causing the path to be
marked invalid on HMR pings, and triggering an endless fastRefresh.
### How?
A simple patch to fix this would fix this is to update `getEntryKey` to
replace any `PAGE_SEGMENT`'s that leak into the entry which I did in
59a972f53339cf6e444e3bf5be45bf115a24c31a.
The other option that's currently implemented here is to only insert
PAGE_SEGMENT as an array in the scenario where there isn't a page
adjacent to the parallel segment. This is to ensure that the
`parallelKey` is `children` rather than the `@parallel` slot when in
[`createSubtreePropsFromSegmentPath`](59a972f533/packages/next/src/build/webpack/loaders/next-app-loader.ts (L298)).
This seems to not cause any regressions with the issue being fixed in
51413, and also solves this case, but I'm just not 100% sure if this
might break another scenario that I'm not thinking of.
Closes NEXT-1337
Fixes#51951
### What
Support `scroll={false}` for Link component in app router. This can be
used when you don't need to scroll back to top again when route url
changes. For instance hash query changes, if you want to keep the
scrolling as it is, you can use this option.
### How
Handling the `scroll` option in navigation reducer on client side.
Fixes#50105
Fixes NEXT-1377
---------
Co-authored-by: Jiachi Liu <inbox@huozhi.im>
Currently we are validating the `experimental.serverActions` flag when creating the actual entries for Server Actions, this causes two problems. One is that syntax errors caught at compilation time are still shown, even if you don't have this flag enabled. Another problem is we still traverse the client graph to collect these action modules even if the flag isn't enabled.
This PR moves that check to be happening at compilation time, which addresses the two above but also brings the extra benefit of showing the exact span and module trace that errors:
<img width="974" alt="CleanShot 2023-07-03 at 20 26 34@2x" src="https://github.com/vercel/next.js/assets/3676859/1676b1f6-e205-4963-bce4-5b515a698e9c">
When looking at [some sites](https://rsc-llm-on-the-edge.vercel.app/) with a large amount of chunks streamed, I noticed that the inlined Flight data array can be optimized quite a lot. Currently we do:
```js
self.__next_f.push([1,"d5:[\"4\",[\"$\",\"$a\",null,..."])
```
1. The `self.` isn't needed (except for the initial bootstrap tag) as React itself has `<script>$RC("B:f","S:f")</script>` too.
2. After the bootstrap script tag, all items are an array with `[1, flight_data]` and `flight_data` is always a string. We can just push only these strings.
3. We use `JSON.stringify(flight_payload)` to inline the payload where the payload itself is a string with a lot of double quotes (`"`), this results in a huge amount of backslashes (`\`). Here we can instead replace it to use a pair of single quotes on the outside and un-escape the double quotes inside.
Here's a side-by-side comparison of a small page:
<img width="1710" alt="CleanShot 2023-06-30 at 11 41 02@2x" src="https://github.com/vercel/next.js/assets/3676859/398356ec-91d5-435c-892d-16fb996029e8">
For a real production page I saw the HTML payload reduced by 11,031 bytes, a 3% improvement.
Note that all the tests are not considering gzip here, so the actual traffic impact will be smaller.
We show the "Application error: a client-side exception has occurred (see the browser console for more information)" error incorrectly when a server-side error occurs (a digest is present) when we should be showing an error saying it is in fact a server side error and should check the logs there. This change will make the error message more accurate for users to look up
Fixes NEXT-1263