kevy-sys

The network-boundary layer for kevy — a tiny, zero-dependency, pure-Rust wrapper over the sockets and readiness poller the kevy server needs.

This is one of kevy's three OS-boundary crates. The other two are publishable, general-purpose crates:

• kevy-uring — pure-Rust io_uring (Linux completion engine), carved out of this crate.
• kevy-madvise — pure-Rust madvise(MADV_HUGEPAGE) hint, also carved out.

kevy-sys is server-internal — its API surface is hand-curated to the exact subset of sockets / pollers kevy's reactor uses. A third party would compare against libc / nix / rustix / mio and find it missing too much; it's not a generic foundation, it's the OS-boundary piece of the kevy server.

• Sockets — tcp_listen / tcp_listen_reuseport, non-blocking I/O, TCP_NODELAY, owned fds that close on drop.
• Readiness poller — one API over kqueue (macOS) and epoll (Linux).
• Cross-thread Waker — a self-pipe to wake a blocked poller.
• Cross-platform sockaddr_in / kevent / epoll_event layouts (incl. the x86_64 packed epoll_event).

use kevy_sys::{Poller, tcp_listen};

let listener = tcp_listen([127, 0, 0, 1], 6379, 1024)?;
listener.set_nonblocking()?;
let poller = Poller::new()?;
poller.add(listener.raw(), true, false)?;
# Ok::<(), std::io::Error>(())

Safety

unsafe is confined to the ffi module and its wrappers; the public API is safe. See the crate docs' Safety section for the ABI invariants.

License

Licensed under either of MIT or Apache-2.0 at your option.