- Before, there was a separate AST traversal that collected `next-dynamic` imports
- Now, we use transitions (which are inserted by the existing next-custom-transform) to tell Turbopack about the next-dynamic imports
- Unlike Webpack (where the react-loadable identifiers are `path/to/parent-module -> ./imported-client`), Turbopack just uses the module id: `[project]/path/to/imported-client.js [app-client] (ecmascript, next/dynamic entry)`
- the `next-dynamic` transition only inserts the marker module (i.e. for the SSR `import()`)
- the `next-dynamic-client` transition inserts the marker module and changes the module context (used to get the client module name)
- Now, we use the correct `AvailabilityInfo` for the next/dynamic chunks to preload, and they are not explicitly emitted. If they are for some reason not the same chunks that are actually emitted for the dynamic import, you get a hard error. (Previously, they didn't use the same `AvailabilityInfo` as the real async chunks, so they usually bundled another copy of React, and also had a different chunk name hash)
- For Turbopack, react-loadable manifests are now scoped per page instead of globally
since we no longer have a single compiler for browser & edge, we can enable
source map generation by default for all edge functions. This would make it
much easier to debug and understand what's happening when deploying to prod
as the log statements will show us the actual code location instead of post
bundling and minified location.
This also removes the experimental flag as it's not needed anymore.
_Hello Next.js team! First PR here, I hope I've followed the right practices._
### What's in there?
It has been decided to only support the following uses cases in Next.js' middleware:
- rewrite the URL (`x-middleware-rewrite` response header)
- redirect to another URL (`Location` response header)
- pass on to the next piece in the request pipeline (`x-middleware-next` response header)
1. during development, a warning on console tells developers when they are returning a response (either with `Response` or `NextResponse`).
2. at build time, this warning becomes an error.
3. at run time, returning a response body will trigger a 500 HTTP error with a JSON payload containing the detailed error.
All returned/thrown errors contain a link to the documentation.
This is a breaking feature compared to the _beta_ middleware implementation, and also removes `NextResponse.json()` which makes no sense any more.
### How to try it?
- runtime behavior: `HEADLESS=true yarn jest test/integration/middleware/core`
- build behavior : `yarn jest test/integration/middleware/build-errors`
- development behavior: `HEADLESS=true yarn jest test/development/middleware-warnings`
### Notes to reviewers
The limitation happens in next's web adapter. ~The initial implementation was to check `response.body` existence, but it turns out [`Response.redirect()`](https://github.com/vercel/next.js/blob/canary/packages/next/server/web/spec-compliant/response.ts#L42-L53) may set the response body (https://github.com/vercel/next.js/pull/31886). Hence why the proposed implementation specifically looks at response headers.~
`Response.redirect()` and `NextResponse.redirect()` do not need to include the final location in their body: it is handled by next server https://github.com/vercel/next.js/blob/canary/packages/next/server/next-server.ts#L1142
Because this is a breaking change, I had to adjust several tests cases, previously returning JSON/stream/text bodies. When relevant, these middlewares are returning data using response headers.
About DevEx: relying on AST analysis to detect forbidden use cases is not as good as running the code.
Such cases are easy to detect:
```js
new Response('a text value')
new Response(JSON.stringify({ /* whatever */ })
```
But these are false-positive cases:
```js
function returnNull() { return null }
new Response(returnNull())
function doesNothing() {}
new Response(doesNothing())
```
However, I see no good reasons to let users ship middleware such as the one above, hence why the build will fail, even if _technically speaking_, they are not setting the response body.
## Feature
- [x] 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`
- [x] Integration tests added
- [x] Documentation added
- [ ] Telemetry added. In case of a feature if it's used or not.
- [x] Errors have helpful link attached, see `contributing.md`
## Documentation / Examples
- [x] Make sure the linting passes by running `yarn lint`
This PR deprecates declaring a middleware under `pages` in favour of the project root naming it after `middleware` instead of `_middleware`. This is in the context of having a simpler execution model for middleware and also ships some refactor work. There is a ton of a code to be simplified after this deprecation but I think it is best to do it progressively.
With this PR, when in development, we will **fail** whenever we find a nested middleware but we do **not** include it in the compiler so if the project is using it, it will no longer work. For production we will **fail** too so it will not be possible to build and deploy a deprecated middleware. The error points to a page that should also be reviewed as part of **documentation**.
Aside from the deprecation, this migrates all middleware tests to work with a single middleware. It also splits tests into multiple folders to make them easier to isolate and work with. Finally it ships some small code refactor and simplifications.
Niklas Mischkulnig