This updates to handle detecting `getStaticProps`/`getServerSideProps` correctly during build when `experimental-edge` is being used. This also fixes not parsing dynamic route params correctly with the edge runtime and sets up the handling needed for the static generation for app opened in the below mentioned PR.
## Bug
- [x] Related issues linked using `fixes #number`
- [x] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
Fixes: [slack thread](https://vercel.slack.com/archives/C0289CGVAR2/p1661554455121189)
x-ref: https://github.com/vercel/next.js/pull/39884
<!--
Thanks for opening a PR! Your contribution is much appreciated.
In order 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 that you're making:
-->
This is intended to refactor the router code to reduce the overhead of executing routes. This is related to #32314 that may help reduce the memory overhead as this also replaced the `Set` with a `WeakMap`.
Co-authored-by: JJ Kasper <jj@jjsweb.site>
This PR improves the Testing documentation in [Setting up Jest (with the Rust Compiler)](https://nextjs.org/docs/testing#setting-up-jest-with-the-rust-compiler). It adds JSDoc typing in `jest.config.js`.
## Documentation / Examples
- [x] Make sure the linting passes by running `pnpm lint`
Currently when the resolved path is `undefined` it generates `filePath: 'undefined'` in the tree object, but should be `filePath: undefined` instead.
## Bug
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
## Feature
- [ ] Implements an existing feature request or RFC. Make sure the feature request has been accepted for implementation before opening a PR.
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Documentation added
- [ ] Telemetry added. In case of a feature if it's used or not.
- [ ] Errors have helpful link attached, see `contributing.md`
## Documentation / Examples
- [ ] Make sure the linting passes by running `pnpm lint`
- [ ] The examples guidelines are followed from [our contributing doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
This PR removes `<noscript>` from `next/future/image` since it is no longer needed now that we use a svg blur placeholder (as opposed to css filter).
- Fixes #38621
- Closes#39736
This PR updates the `next/future/image` component so that CSS `filter` is never needed. Previously, we used SVG to blur for prod and CSS to blur for dev, but now we use SVG for both.
This required a shared function `getImageBlurSvg()` used between both client and server because `next dev` doesn't create Data URIs and instead defers blur generation until request time. So we also need to defer svg generation to request time (on the server) during next dev.
This is the first step to removing `<noscript>` completely (see #39736).
<!--
Thanks for opening a PR! Your contribution is much appreciated.
In order 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 that you're making:
-->
## Documentation / Examples
- [x] Make sure the linting passes by running `pnpm lint`
- [x] The examples guidelines are followed from [our contributing
doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
Convert example to use typescript.
Co-authored-by: Titus Moore <tmoore@edgewebware.com>
Co-authored-by: JJ Kasper <jj@jjsweb.site>
Typescript published 4.8.2 today and it fails CI, bump our typescript version to 4.8.2 and tweak some typings to make existing e2e typescript tests work properly
* Bump web-vitals from 3.0.0-beta to 3.0.0 stable for typing fix (there's an undefined type but it wasn't caught by ts 4.7), also force compiled it as CJS for pre-compiled
* Bump ncc to 3.34.0 for ts-loader compatibility for new typescript version, ncc 3.33.x cannot work with ts 4.8
* Update pre-compiled
### Problem
esm modules imports from client components will be compiled to `m = import('module-name')` when webpack bundles them for server components flight rendering. In this case, they will all become async modules since dyanmic imports will return a promise which react flight cannot handle it then results into module resolving error on server flight rendering.
### Solution
* React flight renderer supports handling async modules in https://github.com/facebook/react/pull/25138
* On next.js side leverage the module proxy change for each client reference, to make sure it always resolve the correct client module
The idea is wrapping each module with a module proxy, and if the module is async and accessed as thenable, it will return a new module reference with `async` label to tell react to handle it as async modules:
exported client reference `*` --> not async module (non thenable) --> original module reference `''`
exported client reference `*` --> it's async module (thenable) --> wrapped module reference `'*'` with `async` label
### Note
Since we need to check if user having incorrect gSSP/gSP specifying in layout client componet, so we still need to parse it and assign those info to the proxy (Does client module containing `ssr`, `ssg` exports). Otherwise the proxy will return the cached module reference
Adding a precision because when you change the `pageExtensions` configuration in `next.config.js`, `middleware.ts` isn't recognized anymore.
Took me some time to debug this one, not sure if it should be considered a bug or just lack of documentation.
I'll let you be the judge of that.
## Documentation / Examples
- [x] Make sure the linting passes by running `pnpm lint`
- [x] The examples guidelines are followed from [our contributing doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
Co-authored-by: JJ Kasper <22380829+ijjk@users.noreply.github.com>
We only use `if (child.type === Script)` on the server side to check the
element type, that's unnecessary because we can add a special flag for
that (`__nextScript` in this PR).
This reduces the server bundle by ~13kb if `next/script` is not imported
by the user.
## Bug
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
## Feature
- [ ] Implements an existing feature request or RFC. Make sure the
feature request has been accepted for implementation before opening a
PR.
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Documentation added
- [ ] Telemetry added. In case of a feature if it's used or not.
- [ ] Errors have helpful link attached, see `contributing.md`
## Documentation / Examples
- [ ] Make sure the linting passes by running `pnpm lint`
- [ ] The examples guidelines are followed from [our contributing
doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
Fixes#39993.
Before the PR:
- `next/script` component mount, `useEffect` for `onReady` executes
- The script's cacheKey is not added to `LoadCache`, skip `onReady`
- The second `useEffect` for `loadScript` executes
- The script's cacheKey is added to `LoadCache` even if it might not fully load yet
- Because of React's strict mode, `useEffect` for `onReady` executes again
- Since the script's cacheKey is in `LoadCache`, `onReady` is called (even when the script is not loaded yet)
- After the script is actually loaded, inside the `script.onload` event handler the `onReady` is called again
After the PR:
- `next/script` component mount, `useEffect` for `onReady` executes
- The script's cacheKey is not added to `LoadCache`, `useEffect` skips `onReady`
- The second `useEffect` for `loadScript` executes
- The script's cacheKey is added to `LoadCache` only if it is an inline script
- Because of React's strict mode, `useEffect` for `onReady` executes again
- The script is not yet loaded, its cacheKey is not in `LoadCache`, `useEffect` skips `onReady` again
- After the script is actually loaded, inside the `script.onload` event handler the `onReady` is finally called
In short, the PR resolves a race condition that only occurs under React strict mode (and makes the `next/script` component more concurrent rendering resilient).
## Bug
- [x] Related issues linked using `fixes #number`
- [x] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
Now we are covering all 8 scenarios in our tests: server-client × global-CSS Modules × page-layout. Next step is to cover HMR and page navigation cases.
## Bug
- [ ] Related issues linked using `fixes #number`
- [x] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
## Feature
- [ ] Implements an existing feature request or RFC. Make sure the feature request has been accepted for implementation before opening a PR.
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Documentation added
- [ ] Telemetry added. In case of a feature if it's used or not.
- [ ] Errors have helpful link attached, see `contributing.md`
## Documentation / Examples
- [ ] Make sure the linting passes by running `pnpm lint`
- [ ] The examples guidelines are followed from [our contributing doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
Removes extra log shown when checking `moduleResolution` in
`tsconfig.json` and adds regression test.
## Bug
- [ ] Related issues linked using `fixes #number`
- [x] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
For historical reasons, Next.js has been falling back to polyfill certain Node.js APIs in the browser. `webpack` itself stopped doing it, and so should Next.js. This might unexpectedly break some packages in the ecosystem though, so it is being introduced as an experimental flag. These imports will now throw a `Module not found` error and the package maintainer should make sure that the library isn't relying on these Node.js APIs when the package is meant for browser usage.
Let's take a look at a common example, the `crypto` API, which can be imported as `import crypto from "crypto"` but [should already be available in browsers](https://caniuse.com/cryptography). Until now, Next.js has fallen back to use a polyfilled version for the import, which resulted in a bundle-size increase.
```js
import crypto from 'crypto'
import { useEffect } from 'react'
export default function Page() {
useEffect(() => {
console.log(crypto)
}, [])
}
```
it imports `crypto`, which currently resolves to [`crypto-browserify`](https://www.npmjs.com/package/crypto-browserify).
So the bundle will include `crypto-browserify` as well:
```sh
Page Size First Load JS
┌ ○ / 131 kB 213 kB # <--
└ ○ /404 194 B 82.2 kB
+ First Load JS shared by all 82 kB
├ chunks/framework-bcc2dc0ea27ab0c6.js 45.1 kB
├ chunks/main-dc2421aef72299b4.js 35.4 kB
├ chunks/pages/_app-a85935458980c5c2.js 708 B
└ chunks/webpack-9b312e20a4e32339.js 836 B
```
Here, we can just remove the import, as we are [safely accessing](https://nextjs.org/docs/migrating/from-create-react-app#safely-accessing-web-apis) the [Crypto Web API](https://caniuse.com/cryptography):
```diff
- import crypto from 'crypto'
import { useEffect } from 'react'
export default function Page() {
useEffect(() => {
console.log(crypto)
}, [])
}
```
Which will reduce the bundle size:
```sh
Page Size First Load JS
┌ ○ / 269 B 82.2 kB # <--
└ ○ /404 194 B 82.1 kB
+ First Load JS shared by all 81.9 kB
├ chunks/framework-bcc2dc0ea27ab0c6.js 45.1 kB
├ chunks/main-dc2421aef72299b4.js 35.4 kB
├ chunks/pages/_app-a85935458980c5c2.js 708 B
└ chunks/webpack-fd82975a6094609f.js 727 B
```
This is harder to detect if the `crypto` import is in a third-party package though. By setting `experimental: { fallbackNodePolyfills: false }`, Next.js will now fail at build-time and should show where the unnecessary import comes from, so the developer can reach out to the package maintainer to fix this issue.
Note: There might be differences between the living standard and some of these older polyfills, so you have to make sure your code works well without the polyfilled version.
Related feedback: https://twitter.com/lfredolo/status/1539608666026000384
This fixes the prefetching handling when middleware is present as currently we are incorrectly triggering data requests for non-SSG pages to handle the case where a middleware rewrite is present and pointing to an SSG path. Since the majority use case won't be rewriting in middleware and this incurs a lot of potentially heavy requests to `getServerSideProps` paths this matches the prefetch handling when middleware isn't present and only prefetches the data route when we match an SSG path.
Fixes: [slack thread](https://vercel.slack.com/archives/C0289CGVAR2/p1661364670589519?thread_ts=1661199566.675759&cid=C0289CGVAR2)
Fixes: https://github.com/vercel/next.js/issues/38918
## Bug
- [x] Related issues linked using `fixes #number`
- [x] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
This warning was incorrectly printed when `fill` and `placeholder="blur"` were used together:
> Image with src "/test.jpg" is smaller than 40x40. Consider removing the "placeholder='blur'" property to improve performance.
Hello, this PR simply configures the `next/mdx` package to use `{ providerImportSource: '@mdx-js/react' }` by default. This is now [required in version 2](https://mdxjs.com/packages/mdx/#optionsproviderimportsource) and causes a lot of confusion personally because `MDXProvider` seems to not work when all is needed is to set this option.
Co-authored-by: JJ Kasper <jj@jjsweb.site>
Update the cache behavior with the correct Header name. Updated header name from `x-nextjs-cache` to `x-vercel-cache`
Co-authored-by: JJ Kasper <jj@jjsweb.site>
## Bug
Adding a components folder inside a catch-all page triggers the error: "Catch-all must be the last part of the URL", having the component outside works, for example `app/[...slug]/components/*.tsx` fails but `app/components/*.tsx` works as expected.
### Fix
The reason it errors because they're treated as route and inserted into the url node tree where they shouldn't be treated in this way. Adding a helper to skip collecting and normailizing every file as route for app dir, but only do it for page files
Co-authored-by: JJ Kasper <jj@jjsweb.site>
Reverts vercel/next.js#39931
Let's instead add this to the custom server documentation, in a section on "Using Middleware". Since such a low percentage of folks eject out of the default server, this shouldn't be so high up on the Middleware docs page.
This makes it more clear that using Document is recommended for the best font experience, versus manually adding it every page. More updates coming to this page in the future as we add support for more features here.
I did some experimenting to see if we can grab the latest version from npm instead of GitHub releases since that's the version the user can actually use.
The npm registry sends the package metadata in a gigantic (~8mb for `next`) JSON response and parsing it with `res.json()` takes several seconds, but using `ReadableStream`, we only care about the first chunk. This way we can grab the relevant version information in a few ms. ~I added `undici` which implements `res.body.getReader()`.~ Using Node 18/native `fetch`.
Follow-up to #39758, this PR makes sure that CSS imports (both global and CSS modules) from client components are not handled by mini-css-extract's HMR logic. Instead, we trigger a server component update and let the client to refetch the RSC payload.
However, we are still leveraging the mini-css-extract plugin to emit CSS assets. So in this PR we add a new pitch loader to calculate the original file hash, but replace the final content to eliminate HMR logic but only keep the hash (so hot reloader can keep tracking that).
## Bug
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Errors have helpful link attached, see `contributing.md`
## Feature
- [ ] Implements an existing feature request or RFC. Make sure the feature request has been accepted for implementation before opening a PR.
- [ ] Related issues linked using `fixes #number`
- [ ] Integration tests added
- [ ] Documentation added
- [ ] Telemetry added. In case of a feature if it's used or not.
- [ ] Errors have helpful link attached, see `contributing.md`
## Documentation / Examples
- [ ] Make sure the linting passes by running `pnpm lint`
- [ ] The examples guidelines are followed from [our contributing doc](https://github.com/vercel/next.js/blob/canary/contributing.md#adding-examples)
This wasn't obvious to me when trying out middleware in our project, and the error message was `Invalid URL` - took me embarrassingly long to remember that we run a custom server. Would it be possibly to detect that middleware is used and then throw a descriptive error message when trying to call `next` without `hostname` and `port` in a custom server? If someone can point me to where that could be done, I would be happy to make a pull request.
## Documentation / Examples
- [ ] Make sure the linting passes by running `pnpm lint`
I made the change in Github ui, so couldn't run `pnpm lint`. Is this automated or should I clone and run it locally?
- Group edge ssr tests, merge `edge-vs.-non-edge-api-route-priority` and `react-18-streaming-ssr` into 1 e2e test suite
- Add rewrite case to edge ssr in switchable runtime