On Mon, Jul 25, 2022 at 9:44 AM Jason A. Donenfeld via Libc-alpha <libc-alpha@xxxxxxxxxxxxxx> wrote: > Or, if you insist on providing these functions t o d a y, and won't heed > my warnings about designing the APIs alongside the implementations, then > just make them thin wrappers over getrandom(0) *without* doing fancy > buffering, and then optimizations later can improve it. That would be > the incremental approach, which wouldn't harm potential users. It also > wouldn't shut the door on doing the buffering: if the kernel > optimization improvements go nowhere, and you decide it's a lost cause, > you can always change the way it works later, and make that decision > then. My 2CLP here if that matters..I agree with this sentiment/approach. provide this functions for source compat which all juist call getrandom and abort on failure *for now* and then a future iteration can have something done about the syscall overhead with kernel help.
