Post

This is great. I'm looking at it as "React as a templating language" - this API makes it trivial to use static (eg. no client component) server components - without even needing special bundler integration. That's awesome.

I mean yeah but they’re not server components they just dehydrated client components same as before with renderToString except you can suspend. If you try to hydrate that you’ll hit issues with the async/await and hooks

I know you know that, just sharing so no one gets confused

Given that this will wait until the whole thing is loaded before starting to emit, why not make it a Promise instead of a Stream? It seems it’d fit the semantics more.

prerender is a promise, the response is a stream because it supports more modern use cases bsky.app/profile/rick...

Possibly silly question: what's the difference between this vs onAllReady via the standard react streaming APIs? I'm assuming the latter is still the recommended way to fully buffer pages for bots, for example? Curious if there are limitations in one or the other I'm overlooking

The former with onAllReady will emit all the intermediate suspense fallbacks plus the content, this will just contain the html with content, so it’s more optimized for this use case.

Ah, thanks - that's interesting since the bot usecase in my mind seems like it's actually closer to wanting just the html (since presumably intermediate fallbacks require js to patch up?), which would seem to favour prerender over onAllReady This is assuming we want zero-js-required first paint

that's the most RTFM "troubleshooting" question i've written

Definitively noticed that 😀 BTW wonder if this is exactly the same as this, or has subtle differences? github.com/facebook/doc...

the difference is that it only outputs the content instead of emitting a stream with the intermediate fallback states and then the content

Ah great I don't use Suspense in Docusaurus 3 yet so that's probably why I didn't really see those intermediate states. Would it emit the JS to replace Suspense placeholders to the stream too?

Also renderToString returns a string, not a web stream, so it's not directly equivalent although you can still convert the web stream to string that could be nice to document 😄

That's the first example in the usage section isn't it? There's an example for using the stream and for rendering the stream to a string?

link react.dev/reference/re...

If this is for SSG and waits for all suspense boundaries, why does it return a stream and not a string?

Even for SSG, if you’re uploading to a CDN it’s more efficient to use their streaming APIs and skip decoding and encoding again, but in the future prerender could be used to stream the shell in a response and resume the shell for dynamic content.