feat: add experimental option to skip SSR transform

[cyco130]

Description

This PR adds an experimental option to skip SSR transform so that we can use Vite directly in Node's (currently experimental) ESM loaders. You can think of it as a minimal alternative to #9740. This tiny change enables the ecosystem to experiment with this approach without modifying Vite's core.

Building a Node ESM loader on top of Vite using this approach would render #3288 and #3928 less urgent.

The PR also exposes shouldExternalizeForSSR which does most of the work in deciding whether a dep should be externalized. Exposing it makes it possible to use the same logic in loaders. (not necessary, import analysis plugin takes care of it, a simple "is first char dot or slash" check is enough)

Additional context

In SSR dev mode, Vite transforms ESM imports and exports into special forms like __vite_ssr_import__ and __vite_ssr_exports__. This PR adds an option to skip that transform so that transform output is valid native ESM. Of course it breaks normal operation but it allows using Vite in Node's ESM loaders.

---

What is the purpose of this pull request?

• Bug fix
• New Feature
• Documentation update
• Other

Before submitting the PR, please make sure you do the following

• Read the Contributing Guidelines.
• Read the Pull Request Guidelines and follow the PR Title Convention.
• Check that there isn't already a PR that solves the problem the same way to avoid creating a duplicate.
• Provide a description in this PR that addresses what the PR is solving, or reference the issue that it solves (e.g. fixes #123).
• Ideally, include relevant tests that fail without this PR but pass with it.

[cyco130]

A good loader implementation also requires access to the shouldExternalizeForSSR function so that it can let Node handle the loading when appropriate. The alternative would be to duplicate the exact logic. Ideas on how to expose it are welcome. (On my local branch, I just exported it with an @experimental annotation).

[bluwy]

Looks like a nice middle ground for now

[bluwy]

A good loader implementation also requires access to the shouldExternalizeForSSR function so that it can let Node handle the loading when appropriate. The alternative would be to duplicate the exact logic. Ideas on how to expose it are welcome. (On my local branch, I just exported it with an @experimental annotation).

When does something need to be externalized or not? IIUC this feature requires externalizing everything by default, which today works partially for non-linked packages.

[cyco130]

When does something need to be externalized or not?

I think most things are externalized for SSR by default. But you can opt out of externalization with ssr.noExternal, ssr.optimizeDeps or you can force externalization with ssr.external. Rakkas, for example, uses ssr.noExternal to force Vite to process Rakkas's own library so that it can access virtual files etc.

shouldExternalizeForSSR does most of the work in deciding whether a dep should be externalized. Exposing it would make it possible to use the same logic in loaders.

(I just pushed a version that exposes it, updating the OP).

[cyco130]

I published this under @cyco130/vite and created a proof-of-concept repo here: https://github.com/cyco130/vite-loader-poc

It doesn't cover many edge cases but it works. Breakpoints and sourcemaps work too.

(Doesn't work on StackBlitz or CodeSandbox, not sure why yet)

Update: A more complete example with the upcoming vavite loader (cyco130/vavite#17), vite-plugin-ssr, and React: https://github.com/cyco130/vavite-vps-loader

Update 2: Another one, this time with SvelteKit.

[bluwy]

shouldExternalizeForSSR does most of the work in deciding whether a dep should be externalized. Exposing it would make it possible to use the same logic in loaders.

Ah ok I misunderstood your point initially. I think it make sense to expose it perhaps as an experimental flag too. (EDIT: looks like you already did it 😄)

[bluwy]

I quite like the idea of using loaders so putting my approval for now. But we may need a quick discussion before proceeding with this.

[cyco130]

I retracted the bit that exposes shouldExternalizeForSSR. Not necessary, see the updated original post.

[patak-dev]

We discussed this in the last meeting, and we decided to move forward with this as an experimental option to let the ecosystem play with this new SSR mode. Let's review later if we should move it out of experimental once we have enough use cases in the wild.

[patak-dev]

We are gathering feedback about this feature for future stabilization here:

• Stabilizing `experimental.skipSsrTransform` #13812