Re: [PATCH 08/12] rust: pin-init: rewrite the initializer macros using `syn`

["Benno Lossin" <lossin@kernel.org>]

On Fri Jan 9, 2026 at 2:45 PM CET, Gary Guo wrote:
> On Thu Jan 8, 2026 at 1:50 PM GMT, Benno Lossin wrote:
>> +    let init_kind = get_init_kind(rest, &mut errors);
>> +    let zeroable_check = match init_kind {
>> +        InitKind::Normal => quote!(),
>> +        InitKind::Zeroing => quote! {
>> +            // The user specified `..Zeroable::zeroed()` at the end of the list of fields.
>> +            // Therefore we check if the struct implements `Zeroable` and then zero the memory.
>> +            // This allows us to also remove the check that all fields are present (since we
>> +            // already set the memory to zero and that is a valid bit pattern).
>> +            fn assert_zeroable<T: ?::core::marker::Sized>(_: *mut T)
>> +            where T: ::pin_init::Zeroable
>> +            {}
>> +            // Ensure that the struct is indeed `Zeroable`.
>> +            assert_zeroable(#slot);
>> +            // SAFETY: The type implements `Zeroable` by the check above.
>> +            unsafe { ::core::ptr::write_bytes(#slot, 0, 1) };
>
> Can this be `#slot.write(::pin_init::zeroed())`?

That could overflow the stack?

>> +        },
>> +    };

>> +        InitKind::Zeroing => quote! {
>> +            // We use unreachable code to ensure that all fields have been mentioned at most once.
>> +            // Since the user specified `..Zeroable::zeroed()` at the end, all missing fields will
>> +            // be zeroed. This struct initializer will still be type-checked and complain with a
>> +            // very natural error message if a field is mentioned more than once, or doesn't exist.
>> +            #[allow(unreachable_code, clippy::diverging_sub_expression, unused_assignments)]
>> +            // SAFETY: this code is never executed.
>> +            let _ = || unsafe {
>> +                let mut zeroed = ::core::mem::zeroed();
>> +                ::core::ptr::write(slot, zeroed);
>
> Looks like the comment explaining why this is done gets missed.

Good catch!

>> +                zeroed = ::core::mem::zeroed();
>> +                ::core::ptr::write(slot, #path {
>> +                    #(
>> +                        #fields: ::core::panic!(),
>> +                    )*
>> +                    ..zeroed
>
> Would just ::core::mem::zeroed() here work or does it have same inference issue?
> IIUC the type inference should work here as ..Default::default() works.

I haven't checked this, will do so.

Cheers,
Benno