Is porting usockets to Rust a good idea? Any plan for it?

[acoBOYZ]

Hi Bun team,

I saw that packages/bun-usockets and packages/bun-uws are Bun's own fork and not synced from upstream uNetworking anymore. They also changed a lot (QUIC, fault inject, only epoll/kqueue and libuv left, many Bun__* hooks).

Since the Rust rewrite is going on, I am wondering is it a good idea to port this layer to pure Rust too? Now src/uws_sys is mostly FFI bindings with a lot of unsafe blocks, so a native port could remove most of them and also the C/Rust boundary.

Also it can be nice for Rust users, maybe one day it can be a standalone crate for people who want to use native usockets in Rust instead of tokio.

So my questions:

• Is it a good idea or there is a reason to keep it in C/C++?
• Is there any short or long term plan to port it?

Thanks for the great work!

[shamil-tp]

Hey bro, porting that layer to pure Rust is definitely a great idea. The main reason it is still in C and C++ right now is just to save time and rely on code that is already battle-tested. uSockets is insanely fast, and rewriting the entire low-level networking stack from scratch is a massive job. During the recent move to Rust, keeping the C library and using FFI bindings was simply the fastest way to get the IO layer working without breaking anything.

For the short term, it looks like the plan is to keep the C version around to maintain stability while the massive new Rust codebase settles down. But in the long term, a pure Rust port makes total sense because Rust natively supports the exact single-thread reactor model that Bun relies on. Doing a full native port would wipe out the C boundary, get rid of all those unsafe FFI blocks, and make the whole core much safer to maintain. Plus, like you mentioned, dropping a native Rust crate based on this event loop model would be a huge win for developers who want a fast alternative to Tokio.

[acoBOYZ]

Thanks bro, glad you agree! Yes, it makes sense to keep the C version for now until the Rust codebase settles down. I guess we should just wait, since the Bun team likes to move things internally until finalize.