[Flight] Assume __turbopack_load_by_url__ returns a cached Promise

[eps1lon]

React doesn't have a signal it can hook into to dispose of chunk cache. Bundlers used to re-evaluate the React Client module to get a fresh cache. However, this is problematic if React Client is in the same bundle (but different module layer) as React Server.

Re-evaluating React Server is not safe since it contains side-effects (console patching and async_hooks).

We can drop the chunk cache in react-server-dom-turbopack by assuming Turbopack returns a cached Promise from __turbopack_load_by_url__ which is implemented in vercel/next.js#81663.

As a follow-up, Turbopack will return already instrumented Promises (setting status).

[eps1lon]

Question is we still need this or if we're ok with Promise.all([Promise.resolve(undefined)]) instead of Promise.all([null]). Or we change Turbopack to return null if the chunk is already loaded.

[sebmarkbage]

It avoids potentially a lot of unnecessary overhead when we end up blocking a large number of chunks downstream for just a microtask. Especially on the server where all of these are either fully sync for basically every request or every request will have to go through some of this.

However, if we trust that Turbopack returns cached promises, then we can also instrument those promises with our .status/.value convention, or even better, have Turbopack yield already instrumented Promises.

[eps1lon]

Yeah, I misread the code originally. It's not Promise.all([null]) if the chunk is already loaded. For sync chunks, preloadModule becomes fully sync. I only accounted for that later in https://github.com/facebook/react/compare/e6ed827a0a52671417a7e34a1707bfda1a209c44..6a4296437e7eb12b18e7b5ee10d213a03cc1b6a1

[react-sizebot]

The size diff is too large to display in a single comment. The GitHub action for this pull request contains an artifact called 'sizebot-message.md' with the full message.

Generated by 🚫 dangerJS against 690f10d