Rewrite node:http to not use Request & Response

[Jarred-Sumner]

What does this PR do?

Extremely WIP.

Rewrite node:http to not use Request & Response for Bun.serve

Introduce a new internal API that currently looks like this and may change at any point:

import { serve } from "bun";

serve({
  port: 3002,

  onNodeHTTPRequest(server, url: string, method: string, headers: Record<string, string>, rawHeaders: string[], handle, hasBody: boolean) {
    handle.writeHead(200, { "Content-Type": "text/plain" });
    handle.end("Hello World");
  },
});

This does not replace fetch in Bun.serve()

How did you verify your code works?

it doesn't really work yet. We aren't emitting socket.

[robobun]

❌ @Jarred-Sumner, your commit 1a59948 has some failures in #7923

[kravetsone]

will it be available for the first time through the flag? it sounds dangerous if suddenly the insides change so much

[Jarred-Sumner]

@kravetsone i agree dangerous, but we have far better test coverage than we used to.

[kravetsone]

@kravetsone i agree dangerous, but we have far better test coverage than we used to.

It's great
But it maybe more painful for developers which don't pin bun docker image version

Test coverage can't find all bugs

[huseeiin]

super cool

[huseeiin]

there was a trick in bun to get the Request object (from a symbol) in node:http:

import { createServer } from "http";

createServer(async (req, res) => {
    const request: Request = req[Object.getOwnPropertySymbols(req)[3]];

    const json = await request.json().catch(() => null);

    res.end(json);
}).listen(3000);

does this PR mean this is no longer gonna be possible?

[huseeiin]

so this idea is completely dismissed or just the pr?

[mtsewrs]

As a library author this functional approach would be great to have imo since it has less overhead and also more compatible with node and i guess uws. Hope you reconsider ❤️

[Jarred-Sumner]

so this idea is completely dismissed or just the pr?

this PR was very stale. we shipped this in Bun v.1.2.6 where we made Express 9% faster, with the actual PR landing the changes in #17093