Commit graph

2 commits

Author SHA1 Message Date
JJ Kasper
8bf6a87303
Rename allowDynamic to unstable_allowDynamic (#40496)
Follow-up to https://github.com/vercel/next.js/pull/39539 as discussed
this renames to `unstable_` prefix initially while we test this out
further.

## 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)
2022-09-12 15:32:18 -07:00
Damien Simonin Feugas
97ac344468
feat(edge): allows configuring Dynamic code execution guard (#39539)
### 📖  What's in there?

Dynamic code evaluation (`eval()`, `new Function()`, ...) is not
supported on the edge runtime, hence why we fail the build when
detecting such statement in the middleware or `experimental-edge` routes
at build time.

However, there could be false positives, which static analysis and
tree-shaking can not exclude:
- `qs` through these dependencies (get-intrinsic:
[source](https://github.com/ljharb/get-intrinsic/blob/main/index.js#L12))
- `function-bind`
([source](https://github.com/Raynos/function-bind/blob/master/implementation.js#L42))
- `has`
([source](https://github.com/tarruda/has/blob/master/src/index.js#L5))

This PR leverages the existing `config` export to let user allow some of
their files.
it’s meant for allowing users to import 3rd party modules who embed
dynamic code evaluation, but do not use it (because or code paths), and
can't be tree-shaked.

By default, it’s keeping the existing behavior: warn in dev, fails to
build.
If users allow dynamic code, and that code is reached at runtime, their
app stills breaks.

### 🧪 How to test?

- (existing) integration tests for disallowing dynamic code evaluation:
`pnpm testheadless --testPathPattern=runtime-dynamic`
- (new) integration tests for allowing dynamic code evaluation: `pnpm
testheadless --testPathPattern=runtime-configurable`
- (amended) production tests for validating the new configuration keys:
`pnpm testheadless --testPathPattern=config-validations`

To try it live, you could have an application such as:
```js
// lib/index.js
/* eslint-disable no-eval */
export function hasUnusedDynamic() {
  if ((() => false)()) {
    eval('100')
  }
}

export function hasDynamic() {
  eval('100')
}

// pages/index.jsx
export default function Page({ edgeRoute }) {
  return <p>{edgeRoute}</p>
}

export const getServerSideProps = async (req) => {
  const res = await fetch(`http://localhost:3000/api/route`)
  const data = await res.json()
  return { props: { edgeRoute: data.ok ? `Hi from the edge route` : '' } }
}

// pages/api/route.js
import { hasDynamic } from '../../lib'

export default async function handle() {
  hasDynamic()
  return Response.json({ ok: true })
}

export const config = { 
  runtime: 'experimental-edge' ,
  allowDynamic: '/lib/**'
}
```

Playing with `config.allowDynamic`, you should be able to:
- build the app even if it uses `eval()` (it will obviously fail at
runtime)
- build the app that _imports but does not use_ `eval()`
- run the app in dev, even if it uses `eval()` with no warning

### 🆙 Notes to reviewers

Before adding documentation and telemetry, I'd like to collect comments
on a couple of points:
- the overall design for this feature: is a list of globs useful and
easy enough?
- should the globs be relative to the application root (current
implementation) to to the edge route/middleware file?
- (especially to @sokra) is the implementation idiomatic enough? I've
leverage loaders to read the _entry point_ configuration once, then the
ModuleGraph to get it back during the parsing phase. I couldn't re-use
the existing `getExtractMetadata()` facility since it's happening late
after the parsing.
- there's a glitch with `import { ServerRuntime } from '../../types'` in
`get-page-static-info.ts`
([here](https://github.com/vercel/next.js/pull/39539/files#diff-cb7ac6392c3dd707c5edab159c3144ec114eafea92dad5d98f4eedfc612174d2L12)).
I had to use `next/types` because it was failing during lint. Any clue
why?

### ☑️ Checklist

- [ ] 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
- [x] Telemetry added. In case of a feature if it's used or not.
- [x] Errors have helpful link attached, see `contributing.md`
2022-09-12 15:01:00 -07:00