Re: [PATCH v1 6/7] rust: workqueue: add safe API to workqueue

[Alice Ryhl]

Benno Lossin <benno.lossin@xxxxxxxxx> writes:
> On Wednesday, May 17th, 2023 at 22:31, Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:
>> +macro_rules! define_work_adapter_newtype {
>> +    (
>> +        $(#[$outer:meta])*
>> +        $pub:vis struct $name:ident(
>> +            $(#[$innermeta:meta])*
>> +            $fpub:vis Arc<$inner:ty> $(,)?
>> +        );
>> +        $($rest:tt)*
>> +    ) => {
>> +        $(#[$outer])*
>> +        $pub struct $name($(#[$innermeta])* $fpub $crate::sync::Arc<$inner>);
> 
> I am a bit confused as to why these types *contain* a pointer. Shouldn't
> these be exactly the same `Work<$inner>`, except they allow multiple `run`
> functions? So IMO they should embed a `Work<$inner>` and the
> manually defined `run` function would take a `$inner`.

No, that's not how this particular patch was designed. With the design I
used, the way you tell `enqueue` which `work_struct` field you want to
use is by using a different pointer type for each `work_struct` field.
This macro defines those pointer types.

So, for example, if you have only one `work_struct` field, then you just
use `Arc<MyStruct>` as your pointer type, and the field has type
`Work<Arc<MyStruct>>`.

On the other hand, if you have two `work_struct` fields, then you
instead use the macro to define `MyPointerType1` and `MyPoinerType2`
that both wrap an `Arc<MyStruct>`, and the fields then have types
`Work<MyPointerType1>` and `Work<MyPointerType2>`.

Alice