Re: [PATCH v1 0/2] Add support for print exactly once

[Boqun Feng <boqun.feng@gmail.com>]

On Fri, Nov 14, 2025 at 10:12:08AM +0900, FUJITA Tomonori wrote:
> On Thu, 13 Nov 2025 16:57:18 -0800
> Boqun Feng <boqun.feng@gmail.com> wrote:
> 
> > On Fri, Nov 14, 2025 at 09:47:30AM +0900, FUJITA Tomonori wrote:
> > [...]
> >> >> Using u8 as atomic type instead of i32 would require a fair amount of
> >> >> work, since at the moment only i32 and i64 are supported as atomic
> >> >> types, right?
> >> > 
> >> > Supporting u8 atomics in general is tricky, but I think *specifically*
> >> > load/store should not be a problem.
> >> 
> >> You meant using read_volatile/write_volatile? Since we only need
> >> "Relaxed" order so something like the following?
> >> 
> >> #[repr(transparent)]
> >> struct AtomicU8(UnsafeCell<u8>);
> >> 
> > 
> > Please don't. Please make it work with Atomic<T>, also store release and
> > load acquire should be introduced as well.
> 
> I wanted to confirm what kind of u8 load/store implementation Alice
> has in mind, and whether it would be used only for OnceLite.
> 
> Of course. If we add something like that to kernel/sync, I agree that
> it should be implemented in that way. If not, I'm not sure.
> 

Yeah, I'm afraid that you want to implement something only for OneLite,
and that'll introduce fragments for maintaining.

For Atomic<T>, supporting load/store/xchg/cmpxchg is definitely the
plan, so appreciate if you could help while working on this pr_*_once().

Alternatively, we can start with i32, and improve that later.

Thanks!

Regards,
Boqun