Somewhat of a tangent - did you ever feed "forced" into io_uring because zig lacked native async io? I know you already pre-allocate, and I guess io_uring is effectively pre-allocating a core to doing io work since you have to spin it
Regardless of language, there was never any question that TB would be using io_uring—the interface is that good! If anything, Zig made it easier, bringing all the primitives/atomics I needed to merge io_uring directly into the std lib.
Note that io_uring doesn’t require a pre-allocated core for spinning. There are multiple modes. We just use the kernel thread pool, which we prefer to a user space thread pool!
