Custom derives for buffer traits

[thomaseizinger]

It is sometimes useful to define your own buffers as wrappers around Vec<u8> or other types that have a valid IoBufMut implementation. In my case, I am getting buffers from a buffer pool (https://docs.rs/lockfree-object-pool/latest/lockfree_object_pool) and therefore I am not holding a Vec<u8> directly but rather something that DerefMuts to Vec<u8>.

It would be nice to have derive macros that allow on to derive IoBufMut for a custom type that just delegates to the inner one.

[Berrysoft]

Well, there's a PR about buffer pool: #358 , but it's unsafe to implement IoBuf or IoBufMut for a reference to buffer from an extern buffer pool. I'm not sure how do you keep the lifetime of the buffers, but it's better to make the unsafe explicit.

I have noticed that edition 2024 added "unsafe" attributes. I'm not sure if it is possible to implement a derive macro that could be only used with #[unsafe(derive(IoBufMut))]. I would like to accept a PR for the derive macro if the keyword unsafe is still explicitly required.

[thomaseizinger]

but it's unsafe to implement IoBuf or IoBufMut for a reference to buffer from an extern buffer pool. I'm not sure how do you keep the lifetime of the buffers, but it's better to make the unsafe explicit.

My wrapper owns the buffer and the buffer internally has an Arc of the pool and returns it to the pool on Drop. Therefore, it is like wrapping a Vec<u8> in terms of lifetimes.

[Berrysoft]

I think it should be safe then. PRs are welcome.

[thomaseizinger]

IoBufMut is only implemented for owned buffers. Unless I am missing something, that enforces already that any wrapper struct also needs to have ownership of the buffer. IoBuf also requires Self: 'static.

[Berrysoft]

Well, yes. But your wrapper already owns the buffer, isn't it?

[thomaseizinger]

Yes. What I meant was: The current trait implementations are only for types that own the buffer and thus, implementing it via deriving should always be safe?

Here is my code: https://github.com/firezone/firezone/pull/8178/files#diff-c3200f3bb4414a260c930e656ddb673a1ad2a8ee292e508bfebf5dbf670c2b8dR320

[Berrysoft]

Seems that your code is OK. The implementation satisfies the safety constraints of IoBuf.