Re: [PATCH 06/12] rust: pin-init: rewrite `#[pin_data]` using `syn`

[Gary Guo]

On Fri Jan 9, 2026 at 4:39 PM GMT, Benno Lossin wrote:
> On Fri Jan 9, 2026 at 1:47 PM CET, Gary Guo wrote:
>> On Thu Jan 8, 2026 at 1:50 PM GMT, Benno Lossin wrote:
>>> +impl Parse for Args {
>>> +    fn parse(input: syn::parse::ParseStream) -> Result<Self> {
>>> +        let lh = input.lookahead1();
>>> +        if lh.peek(End) {
>>> +            input.parse().map(Self::Nothing)
>>
>> How about make this `impl Parse for Option<Args>` and remove the nothing
>> variant? It looks a bit weird.
>
> `Option` is not fundamental, so I can't change the impl. Parse doesn't
> have a blanket impl on `Option`. Maybe I'm not fully understanding the
> change you're proposing.

nvm. I forgot about orphan rules from time to time.

>
>>> +fn strip_pin_annotations(struct_: &mut syn::ItemStruct) {
>>> +    for field in &mut struct_.fields {
>>> +        field.attrs.retain(|a| !a.path().is_ident("pin"));
>>> +    }
>>> +}
>>
>> Multiple places have similar things for stripping annotations and checking if
>> structurally pinned. Would it make sense to do this at the very beginning, and
>> build a `HashSet` of structurally pinned fields, and use that as canonical
>> source for all generate_ functions?
>
> There is https://github.com/Rust-for-Linux/pin-init/pull/94, and I don't
> want to take away their work. I'll contact them and see if we can fold
> it into this commit/series.

I don't think that's the same? I am proposing remove `#[pin]` and build a set at
the very beginning of `fn pin_data` and just use that set. Which is not what the
reference PR does.

Best,
Gary